upload config files

LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2025 MiniMax
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

config.json

@@ -1,6 +1,6 @@
 {
   "architectures": [
-    "AbabForCausalLM"
+    "MiniMaxM1ForCausalLM"
   ],
   "attention_dropout": 0.0,
   "attn_type_list": [
@@ -86,21 +86,24 @@
     1
   ],
   "auto_map": {
-    "AutoConfig": "configuration_abab.AbabConfig",
-    "AutoModelForCausalLM": "modeling_abab.AbabForCausalLM"
+    "AutoConfig": "configuration_minimax_m1.MiniMaxM1Config",
+    "AutoModelForCausalLM": "modeling_minimax_m1.MiniMaxM1ForCausalLM"
   },
-  "bos_token_id": 1,
-  "eos_token_id": 2,
+  "bos_token_id": null,
+  "eos_token_id": null,
   "head_dim": 128,
   "hidden_act": "silu",
   "hidden_size": 6144,
   "initializer_range": 0.02,
   "intermediate_size": 9216,
   "layernorm_full_attention_alpha": 3.5565588200778455,
+  "layernorm_full_attention_beta": 1.0,
   "layernorm_linear_attention_alpha": 3.5565588200778455,
+  "layernorm_linear_attention_beta": 1.0,
   "layernorm_mlp_alpha": 3.5565588200778455,
-  "max_position_embeddings": 131072,
-  "model_type": "mixtral",
+  "layernorm_mlp_beta": 1.0,
+  "max_position_embeddings": 10240000,
+  "model_type": "minimax_m1",
   "num_attention_heads": 64,
   "num_experts_per_tok": 2,
   "num_hidden_layers": 80,
@@ -117,7 +120,8 @@
   "shared_moe_mode": "sigmoid",
   "sliding_window": null,
   "tie_word_embeddings": false,
-  "transformers_version": "4.49.0",
+  "transformers_version": "4.45.2",
   "use_cache": true,
   "vocab_size": 200064
 }
+

configuration_minimax_m1.py

@@ -0,0 +1,152 @@
+""" MiniMaxM1 model configuration"""
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+
+class MiniMaxM1Config(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`MiniMaxM1Model`]. It is used to instantiate an
+    MiniMaxM1 model according to the specified arguments, defining the model architecture. Instantiating a configuration
+    with the defaults will yield a similar configuration to that of the MiniMaxM1.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 32000):
+            Vocabulary size of the MiniMaxM1 model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`MiniMaxM1Model`]
+        hidden_size (`int`, *optional*, defaults to 4096):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 14336):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        num_key_value_heads (`int`, *optional*, defaults to 8):
+            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
+            `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
+            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
+            by meanpooling all the original heads within that group. For more details checkout [this
+            paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to `8`.
+        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (function or string) in the decoder.
+        max_position_embeddings (`int`, *optional*, defaults to `4096*32`):
+            The maximum sequence length that this model might ever be used with. MiniMaxM1's sliding window attention
+            allows sequence of up to 4096*32 tokens.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-05):
+            The epsilon used by the rms normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        pad_token_id (`int`, *optional*):
+            The id of the padding token.
+        bos_token_id (`int`, *optional*, defaults to 1):
+            The id of the "beginning-of-sequence" token.
+        eos_token_id (`int`, *optional*, defaults to 2):
+            The id of the "end-of-sequence" token.
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether the model's input and output word embeddings should be tied.
+        rope_theta (`float`, *optional*, defaults to 1000000.0):
+            The base period of the RoPE embeddings.
+        sliding_window (`int`, *optional*):
+            Sliding window attention window size. If not specified, will default to `4096`.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        num_experts_per_tok (`int`, *optional*, defaults to 2):
+            The number of experts to route per-token, can be also interpreted as the `top-k` routing
+            parameter
+        num_local_experts (`int`, *optional*, defaults to 8):
+            Number of experts per Sparse MLP layer.
+        output_router_logits (`bool`, *optional*, defaults to `False`):
+            Whether or not the router logits should be returned by the model. Enabeling this will also
+            allow the model to output the auxiliary loss. See [here]() for more details
+        router_aux_loss_coef (`float`, *optional*, defaults to 0.001):
+            The aux loss factor for the total loss.
+        router_jitter_noise (`float`, *optional*, defaults to 0.0):
+            Amount of noise to add to the router.
+
+    ```python
+    >>> from transformers import MiniMaxM1Model, MiniMaxM1Config
+
+    >>> # Initializing a MiniMaxM1 style configuration
+    >>> configuration = MiniMaxM1Config()
+
+    >>> # Initializing a model from the MiniMaxM1 style configuration
+    >>> model = MiniMaxM1Model(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "MiniMaxM1"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=32000,
+        hidden_size=4096,
+        intermediate_size=14336,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=8,
+        hidden_act="silu",
+        max_position_embeddings=4096 * 32,
+        initializer_range=0.02,
+        rms_norm_eps=1e-5,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=None,
+        eos_token_id=None,
+        tie_word_embeddings=False,
+        rope_theta=1e6,
+        sliding_window=None,
+        attention_dropout=0.0,
+        num_experts_per_tok=2,
+        num_local_experts=8,
+        output_router_logits=False,
+        router_aux_loss_coef=0.001,
+        router_jitter_noise=0.0,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.sliding_window = sliding_window
+
+        # for backward compatibility
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.attention_dropout = attention_dropout
+
+        self.num_experts_per_tok = num_experts_per_tok
+        self.num_local_experts = num_local_experts
+        self.output_router_logits = output_router_logits
+        self.router_aux_loss_coef = router_aux_loss_coef
+        self.router_jitter_noise = router_jitter_noise
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )

function_call_guide.md

@@ -0,0 +1,270 @@
+# MiniMax-M1 Function Call Guide
+
+[FunctionCall中文使用指南](./function_call_guide_cn.md)
+
+## 📖 Introduction
+
+The MiniMax-M1 model supports function calling capabilities, enabling the model to identify when external functions need to be called and output function call parameters in a structured format. This document provides detailed instructions on how to use the function calling feature of MiniMax-M1.
+
+## 🚀 Quick Start
+
+### Using Chat Template
+
+MiniMax-M1 uses a specific chat template format to handle function calls. The chat template is defined in `tokenizer_config.json`, and you can use it in your code through the template.
+
+```python
+from transformers import AutoTokenizer
+
+def get_default_tools():
+    return [
+        {
+          {
+            "name": "get_current_weather",
+            "description": "Get the latest weather for a location",
+            "parameters": {
+                "type": "object", 
+                "properties": {
+                    "location": {
+                        "type": "string", 
+                        "description": "A certain city, such as Beijing, Shanghai"
+                    }
+                }, 
+            }
+            "required": ["location"],
+            "type": "object"
+          }
+        }
+    ]
+
+# Load model and tokenizer
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+prompt = "What's the weather like in Shanghai today?"
+messages = [
+    {"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant created by Minimax based on MiniMax-M1 model."}]},
+    {"role": "user", "content": [{"type": "text", "text": prompt}]},
+]
+
+# Enable function call tools
+tools = get_default_tools()
+
+# Apply chat template and add tool definitions
+text = tokenizer.apply_chat_template(
+    messages,
+    tokenize=False,
+    add_generation_prompt=True,
+    tools=tools
+)
+```
+
+## 🛠️ Function Call Definition
+
+### Function Structure
+
+Function calls need to be defined in the `tools` field of the request body. Each function consists of the following components:
+
+```json
+{
+  "tools": [
+    {
+      "name": "search_web",
+      "description": "Search function.",
+      "parameters": {
+        "properties": {
+          "query_list": {
+            "description": "Keywords for search, with list element count of 1.",
+            "items": { "type": "string" },
+            "type": "array"
+          },
+          "query_tag": {
+            "description": "Classification of the query",
+            "items": { "type": "string" },
+            "type": "array"
+          }
+        },
+        "required": [ "query_list", "query_tag" ],
+        "type": "object"
+      }
+    }
+  ]
+}
+```
+
+**Field Descriptions:**
+- `name`: Function name
+- `description`: Function description
+- `parameters`: Function parameter definition
+  - `properties`: Parameter property definitions, where key is the parameter name and value contains detailed parameter description
+  - `required`: List of required parameters
+  - `type`: Parameter type (usually "object")
+
+### Internal Model Processing Format
+
+When processed internally by the model, function definitions are converted to a special format and concatenated to the input text:
+
+```
+]~!b[]~b]system ai_setting=MiniMax AI
+MiniMax AI is an AI assistant independently developed by MiniMax. [e~[
+]~b]system tool_setting=tools
+You are provided with these tools:
+<tools>
+{"name": "search_web", "description": "Search function.", "parameters": {"properties": {"query_list": {"description": "Keywords for search, with list element count of 1.", "items": {"type": "string"}, "type": "array"}, "query_tag": {"description": "Classification of the query", "items": {"type": "string"}, "type": "array"}}, "required": ["query_list", "query_tag"], "type": "object"}}
+</tools>
+
+If you need to call tools, please respond with <tool_calls></tool_calls> XML tags, and provide tool-name and json-object of arguments, following the format below:
+<tool_calls>
+{"name": <tool-name>, "arguments": <args-json-object>}
+...
+</tool_calls>[e~[
+]~b]user name=User
+When were the most recent launch events for OpenAI and Gemini?[e~[
+]~b]ai name=MiniMax AI
+```
+
+### Model Output Format
+
+The model outputs function calls in the following format:
+
+```xml
+<think>
+Okay, I will search for the OpenAI and Gemini latest release.
+</think>
+<tool_calls>
+{"name": "search_web", "arguments": {"query_tag": ["technology", "events"], "query_list": ["\"OpenAI\" \"latest\" \"release\""]}}
+{"name": "search_web", "arguments": {"query_tag": ["technology", "events"], "query_list": ["\"Gemini\" \"latest\" \"release\""]}}
+</tool_calls>
+```
+
+## 📥 Function Call Result Processing
+
+### Parsing Function Calls
+
+You can use the following code to parse function calls from the model output:
+
+```python
+import re
+import json
+
+def parse_function_calls(content: str):
+    """
+    Parse function calls from model output
+    """
+    function_calls = []
+    
+    # Match content within <tool_calls> tags
+    tool_calls_pattern = r"<tool_calls>(.*?)</tool_calls>"
+    tool_calls_match = re.search(tool_calls_pattern, content, re.DOTALL)
+    
+    if not tool_calls_match:
+        return function_calls
+    
+    tool_calls_content = tool_calls_match.group(1).strip()
+    
+    # Parse each function call (one JSON object per line)
+    for line in tool_calls_content.split('\n'):
+        line = line.strip()
+        if not line:
+            continue
+            
+        try:
+            # Parse JSON format function call
+            call_data = json.loads(line)
+            function_name = call_data.get("name")
+            arguments = call_data.get("arguments", {})
+            
+            function_calls.append({
+                "name": function_name,
+                "arguments": arguments
+            })
+            
+            print(f"Function call: {function_name}, Arguments: {arguments}")
+            
+        except json.JSONDecodeError as e:
+            print(f"Parameter parsing failed: {line}, Error: {e}")
+    
+    return function_calls
+
+# Example: Handle weather query function
+def execute_function_call(function_name: str, arguments: dict):
+    """
+    Execute function call and return result
+    """
+    if function_name == "get_current_weather":
+        location = arguments.get("location", "Unknown location")
+        # Build function execution result
+        return {
+            "role": "tool", 
+            "name": function_name, 
+            "content": json.dumps({
+                "location": location, 
+                "temperature": "25", 
+                "unit": "celsius", 
+                "weather": "Sunny"
+            }, ensure_ascii=False)
+        }
+    elif function_name == "search_web":
+        query_list = arguments.get("query_list", [])
+        query_tag = arguments.get("query_tag", [])
+        # Simulate search results
+        return {
+            "role": "tool",
+            "name": function_name,
+            "content": f"Search keywords: {query_list}, Categories: {query_tag}\nSearch results: Relevant information found"
+        }
+    
+    return None
+```
+
+### Returning Function Execution Results to the Model
+
+After successfully parsing function calls, you should add the function execution results to the conversation history so that the model can access and utilize this information in subsequent interactions.
+
+#### Single Result
+
+If the model decides to call `search_web`, we suggest you to return the function result in the following format, with the `name` field set to the specific tool name.
+
+```json
+{
+  "data": [
+     {
+       "role": "tool", 
+       "name": "search_web", 
+       "content": "search_result"
+     }
+  ]
+}
+```
+
+Corresponding model input format:
+```
+]~b]tool name=search_web
+search_result[e~[
+```
+
+
+#### Multiple Result
+If the model decides to call `search_web` and `get_current_weather` at the same time, we suggest you to return the multiple function results in the following format, with the `name` field set to "tools", and use the `content` field to contain multiple results.
+
+
+```json
+{
+  "data": [
+     {
+       "role": "tool", 
+       "name": "tools", 
+       "content": "Tool name: search_web\nTool result: test_result1\n\nTool name: get_current_weather\nTool result: test_result2"
+     }
+  ]
+}
+```
+
+Corresponding model input format:
+```
+]~b]tool name=tools
+Tool name: search_web
+Tool result: test_result1
+
+Tool name: get_current_weather
+Tool result: test_result2[e~[
+```
+
+While we suggest following the above formats, as long as the model input is easy to understand, the specific values of `name` and `content` is entirely up to the caller.

function_call_guide_cn.md

@@ -0,0 +1,267 @@
+# MiniMax-M1 函数调用（Function Call）功能指南
+
+## 📖 简介
+
+MiniMax-M1 模型支持函数调用功能，使模型能够识别何时需要调用外部函数，并以结构化格式输出函数调用参数。本文档详细介绍了如何使用 MiniMax-M1 的函数调用功能。
+
+## 🚀 快速开始
+
+### 聊天模板使用
+
+MiniMax-M1 使用特定的聊天模板格式处理函数调用。聊天模板定义在 `tokenizer_config.json` 中，你可以在代码中通过 template 来进行使用。
+
+```python
+from transformers import AutoTokenizer
+
+def get_default_tools():
+    return [
+        {
+          {
+            "name": "get_current_weather",
+            "description": "Get the latest weather for a location",
+            "parameters": {
+                "type": "object", 
+                "properties": {
+                    "location": {
+                        "type": "string", 
+                        "description": "A certain city, such as Beijing, Shanghai"
+                    }
+                }, 
+            }
+            "required": ["location"],
+            "type": "object"
+          }
+        }
+    ]
+
+# 加载模型和分词器
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+prompt = "What's the weather like in Shanghai today?"
+messages = [
+    {"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant created by Minimax based on MiniMax-M1 model."}]},
+    {"role": "user", "content": [{"type": "text", "text": prompt}]},
+]
+
+# 启用函数调用工具
+tools = get_default_tools()
+
+# 应用聊天模板，并加入工具定义
+text = tokenizer.apply_chat_template(
+    messages,
+    tokenize=False,
+    add_generation_prompt=True,
+    tools=tools
+)
+```
+
+## 🛠️ 函数调用的定义
+
+### 函数结构体
+
+函数调用需要在请求体中定义 `tools` 字段，每个函数由以下部分组成：
+
+```json
+{
+  "tools": [
+    {
+      "name": "search_web",
+      "description": "搜索函数。",
+      "parameters": {
+        "properties": {
+          "query_list": {
+            "description": "进行搜索的关键词，列表元素个数为1。",
+            "items": { "type": "string" },
+            "type": "array"
+          },
+          "query_tag": {
+            "description": "query的分类",
+            "items": { "type": "string" },
+            "type": "array"
+          }
+        },
+        "required": [ "query_list", "query_tag" ],
+        "type": "object"
+      }
+    }
+  ]
+}
+```
+
+**字段说明：**
+- `name`: 函数名称
+- `description`: 函数功能描述
+- `parameters`: 函数参数定义
+  - `properties`: 参数属性定义，key 是参数名，value 包含参数的详细描述
+  - `required`: 必填参数列表
+  - `type`: 参数类型（通常为 "object"）
+
+### 模型内部处理格式
+
+在模型内部处理时，函数定义会被转换为特殊格式并拼接到输入文本中：
+
+```
+]~!b[]~b]system ai_setting=MiniMax AI
+MiniMax AI是由上海稀宇科技有限公司（MiniMax）自主研发的AI助理。[e~[
+]~b]system tool_setting=tools
+You are provided with these tools:
+<tools>
+{"name": "search_web", "description": "搜索函数。", "parameters": {"properties": {"query_list": {"description": "进行搜索的关键词，列表元素个数为1。", "items": {"type": "string"}, "type": "array"}, "query_tag": {"description": "query的分类", "items": {"type": "string"}, "type": "array"}}, "required": ["query_list", "query_tag"], "type": "object"}}
+</tools>
+
+If you need to call tools, please respond with <tool_calls></tool_calls> XML tags, and provide tool-name and json-object of arguments, following the format below:
+<tool_calls>
+{"name": <tool-name>, "arguments": <args-json-object>}
+...
+</tool_calls>[e~[
+]~b]user name=用户
+OpenAI 和 Gemini 的最近一次发布会都是什么时候?[e~[
+]~b]ai name=MiniMax AI
+```
+
+### 模型输出格式
+
+模型会以以下格式输出函数调用：
+
+```xml
+<think>
+Okay, I will search for the OpenAI and Gemini latest release.
+</think>
+<tool_calls>
+{"name": "search_web", "arguments": {"query_tag": ["technology", "events"], "query_list": ["\"OpenAI\" \"latest\" \"release\""]}}
+{"name": "search_web", "arguments": {"query_tag": ["technology", "events"], "query_list": ["\"Gemini\" \"latest\" \"release\""]}}
+</tool_calls>
+```
+
+## 📥 函数调用结果处理
+
+### 解析函数调用
+
+您可以使用以下代码解析模型输出的函数调用：
+
+```python
+import re
+import json
+
+def parse_function_calls(content: str):
+    """
+    解析模型输出中的函数调用
+    """
+    function_calls = []
+    
+    # 匹配 <tool_calls> 标签内的内容
+    tool_calls_pattern = r"<tool_calls>(.*?)</tool_calls>"
+    tool_calls_match = re.search(tool_calls_pattern, content, re.DOTALL)
+    
+    if not tool_calls_match:
+        return function_calls
+    
+    tool_calls_content = tool_calls_match.group(1).strip()
+    
+    # 解析每个函数调用（每行一个JSON对象）
+    for line in tool_calls_content.split('\n'):
+        line = line.strip()
+        if not line:
+            continue
+            
+        try:
+            # 解析JSON格式的函数调用
+            call_data = json.loads(line)
+            function_name = call_data.get("name")
+            arguments = call_data.get("arguments", {})
+            
+            function_calls.append({
+                "name": function_name,
+                "arguments": arguments
+            })
+            
+            print(f"调用函数: {function_name}, 参数: {arguments}")
+            
+        except json.JSONDecodeError as e:
+            print(f"参数解析失败: {line}, 错误: {e}")
+    
+    return function_calls
+
+# 示例：处理天气查询函数
+def execute_function_call(function_name: str, arguments: dict):
+    """
+    执行函数调用并返回结果
+    """
+    if function_name == "get_current_weather":
+        location = arguments.get("location", "未知位置")
+        # 构建函数执行结果
+        return {
+            "role": "tool", 
+            "name": function_name, 
+            "content": json.dumps({
+                "location": location, 
+                "temperature": "25", 
+                "unit": "celsius", 
+                "weather": "晴朗"
+            }, ensure_ascii=False)
+        }
+    elif function_name == "search_web":
+        query_list = arguments.get("query_list", [])
+        query_tag = arguments.get("query_tag", [])
+        # 模拟搜索结果
+        return {
+            "role": "tool",
+            "name": function_name,
+            "content": f"搜索关键词: {query_list}, 分类: {query_tag}\n搜索结果: 相关信息已找到"
+        }
+    
+    return None
+```
+
+### 将函数执行结果返回给模型
+
+成功解析函数调用后，您应将函数执行结果添加到对话历史中，以便模型在后续交互中能够访问和利用这些信息。
+
+#### 单个结果
+
+假如模型调用了 `search_web` 函数，您可以参考如下格式添加执行结果，`name` 字段为具体的函数名称。
+
+```json
+{
+  "data": [
+     {
+       "role": "tool", 
+       "name": "search_web", 
+       "content": "search_result"
+     }
+  ]
+}
+```
+
+对应如下的模型输入格式：
+```
+]~b]tool name=search_web
+search_result[e~[
+```
+
+
+#### 多个结果
+假如模型同时调用了 `search_web` 和 `get_current_weather` 函数，您可以参考如下格式添加执行结果，`name` 字段为"tools"，`content`包含多个结果。
+
+```json
+{
+  "data": [
+     {
+       "role": "tool", 
+       "name": "tools", 
+       "content": "Tool name: search_web\nTool result: test_result1\n\nTool name: get_current_weather\nTool result: test_result2"
+     }
+  ]
+}
+```
+
+对应如下的模型输入格式：
+```
+]~b]tool name=tools
+Tool name: search_web
+Tool result: test_result1
+
+Tool name: get_current_weather
+Tool result: test_result2[e~[
+```
+
+虽然我们建议您参考以上格式，但只要返回给模型的输入易于理解，`name` 和 `content` 的具体内容完全由您自主决定。

main.py

@@ -0,0 +1,106 @@
+from transformers import AutoModelForCausalLM, AutoTokenizer, AutoConfig, QuantoConfig, GenerationConfig
+import torch
+import argparse
+
+"""
+ usage:
+    export SAFETENSORS_FAST_GPU=1
+    python main.py --quant_type int8 --world_size 8 --model_id <model_path>
+"""
+
+def generate_quanto_config(hf_config: AutoConfig, quant_type: str):
+    QUANT_TYPE_MAP = {
+        "default": None,
+        "int8": QuantoConfig(
+            weights="int8",
+            modules_to_not_convert=[
+                "lm_head",
+                "embed_tokens",
+            ] + [f"model.layers.{i}.coefficient" for i in range(hf_config.num_hidden_layers)]
+            + [f"model.layers.{i}.block_sparse_moe.gate" for i in range(hf_config.num_hidden_layers)]
+        ),
+    }
+    return QUANT_TYPE_MAP[quant_type]
+
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--quant_type", type=str, default="default", choices=["default", "int8"])
+    parser.add_argument("--model_id", type=str, required=True)
+    parser.add_argument("--world_size", type=int, required=True)
+    return parser.parse_args()
+
+
+def check_params(args, hf_config: AutoConfig):
+    if args.quant_type == "int8":
+        assert args.world_size >= 8, "int8 weight-only quantization requires at least 8 GPUs"
+
+    assert hf_config.num_hidden_layers % args.world_size == 0, f"num_hidden_layers({hf_config.num_hidden_layers}) must be divisible by world_size({args.world_size})"
+
+
+@torch.no_grad()
+def main():
+    args = parse_args()
+    print("\n=============== Argument ===============")
+    for key in vars(args):
+        print(f"{key}: {vars(args)[key]}")
+    print("========================================")
+
+    model_id = args.model_id
+
+    hf_config = AutoConfig.from_pretrained(model_id, trust_remote_code=True)
+    check_params(args, hf_config)
+    quantization_config = generate_quanto_config(hf_config, args.quant_type)
+ 
+    device_map = {
+        'model.embed_tokens': 'cuda:0',
+        'model.norm': f'cuda:{args.world_size - 1}',
+        'lm_head': f'cuda:{args.world_size - 1}'
+    }
+    layers_per_device = hf_config.num_hidden_layers // args.world_size
+    for i in range(args.world_size):
+        for j in range(layers_per_device):
+            device_map[f'model.layers.{i * layers_per_device + j}'] = f'cuda:{i}'
+
+    tokenizer = AutoTokenizer.from_pretrained(model_id)
+    message = [
+        {"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant."}]},
+        {"role": "user", "content": [{"type": "text", "text": "Hello, what is the weather today?"}]}
+    ]
+    tools = [
+        {"name": "get_location", "description": "Get the location of the user.", "parameters": {"type": "object", "properties": {}}},
+        {"name": "get_weather", "description": "Get the weather of a city.", "parameters": {"type": "object", "properties": {"city": {"type": "string", "description": "The name of the city"}}}},
+        {"name": "get_news", "description": "Get the news.", "parameters": {"type": "object", "properties": {"domain": {"type": "string", "description": "The domain of the news"}}}}
+    ]
+    text = tokenizer.apply_chat_template(
+        message,
+        tools,
+        tokenize=False,
+        add_generation_prompt=True
+    )
+    model_inputs = tokenizer(text, return_tensors="pt").to("cuda")
+    quantized_model = AutoModelForCausalLM.from_pretrained(
+        model_id,
+        torch_dtype="bfloat16",
+        device_map=device_map,
+        quantization_config=quantization_config,
+        trust_remote_code=True,
+        offload_buffers=True,
+    )
+    generation_config = GenerationConfig(
+        max_new_tokens=20,
+        eos_token_id=200020,
+        use_cache=True,
+    )
+    generated_ids = quantized_model.generate(**model_inputs, generation_config=generation_config)
+    print(f"generated_ids: {generated_ids}")
+    generated_ids = [
+        output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
+    ]
+    response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+    print(response)
+
+if __name__ == "__main__":
+    main()
+
+

modeling_minimax_m1.py

@@ -0,0 +1,1701 @@
+""" PyTorch MiniMaxM1 model."""
+import inspect
+import math
+import warnings
+from typing import List, Optional, Tuple, Union
+import os
+import copy
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
+from einops import rearrange, repeat
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache
+from transformers.modeling_attn_mask_utils import (
+    _prepare_4d_causal_attention_mask,
+)
+from transformers.modeling_outputs import (
+    MoeCausalLMOutputWithPast,
+    MoeModelOutputWithPast,
+    SequenceClassifierOutputWithPast,
+)
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import (
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    is_flash_attn_2_available,
+    is_flash_attn_greater_or_equal_2_10,
+    logging,
+    replace_return_docstrings,
+)
+from transformers.utils.import_utils import is_torch_fx_available
+from .configuration_minimax_m1 import MiniMaxM1Config
+
+if is_flash_attn_2_available():
+    from flash_attn import flash_attn_func, flash_attn_varlen_func
+    from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input  # noqa
+
+    _flash_supports_window_size = "window_size" in list(inspect.signature(flash_attn_func).parameters)
+
+# This makes `_prepare_4d_causal_attention_mask` a leaf function in the FX graph.
+# It means that the function will not be traced through and simply appear as a node in the graph.
+if is_torch_fx_available():
+    _prepare_4d_causal_attention_mask = torch.fx.wrap(_prepare_4d_causal_attention_mask)
+    
+use_triton = eval(os.environ.get("use_triton", default="False"))
+debug = eval(os.environ.get("debug", default="False"))
+do_eval = eval(os.environ.get("do_eval", default="False"))
+eval_and_not_generate = eval(os.environ.get("eval_and_not_generate", default="False"))
+BLOCK = 256
+
+logger = logging.get_logger(__name__)
+
+_CONFIG_FOR_DOC = "MiniMaxM1Config"
+
+
+def get_activation_fn(activation):
+    if debug:
+        logger.info(f"activation: {activation}")
+    if activation == "gelu":
+        return F.gelu
+    elif activation == "relu":
+        return F.relu
+    elif activation == "elu":
+        return F.elu
+    elif activation == "sigmoid":
+        return F.sigmoid
+    elif activation == "exp":
+
+        def f(x):
+            with torch.no_grad():
+                x_max = torch.max(x, dim=-1, keepdims=True).values
+            y = torch.exp(x - x_max)
+
+            return y
+
+        return f
+    elif activation == "leak":
+        return F.leaky_relu
+    elif activation == "1+elu":
+
+        def f(x):
+            return 1 + F.elu(x)
+
+        return f
+    elif activation == "2+elu":
+
+        def f(x):
+            return 2 + F.elu(x)
+
+        return f
+    elif activation == "silu" or activation == "swish":
+        return F.silu
+    elif activation == "sine":
+        return torch.sin
+    else:
+        logger.info(
+            f"activation: does not support {activation}, use Identity!!!")
+        return lambda x: x
+
+
+def load_balancing_loss_func(
+        gate_logits: torch.Tensor, num_experts: torch.Tensor = None, top_k=2,
+        attention_mask: Optional[torch.Tensor] = None
+) -> float:
+    r"""
+    Computes auxiliary load balancing loss as in Switch Transformer - implemented in Pytorch.
+
+    See Switch Transformer (https://arxiv.org/abs/2101.03961) for more details. This function implements the loss
+    function presented in equations (4) - (6) of the paper. It aims at penalizing cases where the routing between
+    experts is too unbalanced.
+
+    Args:
+        gate_logits (Union[`torch.Tensor`, Tuple[torch.Tensor]):
+            Logits from the `gate`, should be a tuple of model.config.num_hidden_layers tensors of
+            shape [batch_size X sequence_length, num_experts].
+        attention_mask (`torch.Tensor`, None):
+            The attention_mask used in forward function
+            shape [batch_size X sequence_length] if not None.
+        num_experts (`int`, *optional*):
+            Number of experts
+
+    Returns:
+        The auxiliary loss.
+    """
+    if gate_logits is None or not isinstance(gate_logits, tuple):
+        return 0
+
+    if isinstance(gate_logits, tuple):
+        compute_device = gate_logits[0].device
+        concatenated_gate_logits = torch.cat([layer_gate.to(compute_device) for layer_gate in gate_logits], dim=0)
+
+    routing_weights = torch.nn.functional.softmax(concatenated_gate_logits, dim=-1)
+
+    _, selected_experts = torch.topk(routing_weights, top_k, dim=-1)
+
+    expert_mask = torch.nn.functional.one_hot(selected_experts, num_experts)
+
+    if attention_mask is None:
+        # Compute the percentage of tokens routed to each experts
+        tokens_per_expert = torch.mean(expert_mask.float(), dim=0)
+
+        # Compute the average probability of routing to these experts
+        router_prob_per_expert = torch.mean(routing_weights, dim=0)
+    else:
+        batch_size, sequence_length = attention_mask.shape
+        num_hidden_layers = concatenated_gate_logits.shape[0] // (batch_size * sequence_length)
+
+        # Compute the mask that masks all padding tokens as 0 with the same shape of expert_mask
+        expert_attention_mask = (
+            attention_mask[None, :, :, None, None]
+            .expand((num_hidden_layers, batch_size, sequence_length, top_k, num_experts))
+            .reshape(-1, top_k, num_experts)
+            .to(compute_device)
+        )
+
+        # Compute the percentage of tokens routed to each experts
+        tokens_per_expert = torch.sum(expert_mask.float() * expert_attention_mask, dim=0) / torch.sum(
+            expert_attention_mask, dim=0
+        )
+
+        # Compute the mask that masks all padding tokens as 0 with the same shape of tokens_per_expert
+        router_per_expert_attention_mask = (
+            attention_mask[None, :, :, None]
+            .expand((num_hidden_layers, batch_size, sequence_length, num_experts))
+            .reshape(-1, num_experts)
+            .to(compute_device)
+        )
+
+        # Compute the average probability of routing to these experts
+        router_prob_per_expert = torch.sum(routing_weights * router_per_expert_attention_mask, dim=0) / torch.sum(
+            router_per_expert_attention_mask, dim=0
+        )
+
+    overall_loss = torch.sum(tokens_per_expert * router_prob_per_expert.unsqueeze(0))
+    return overall_loss * num_experts
+
+
+# Copied from transformers.models.llama.modeling_llama._get_unpad_data
+def _get_unpad_data(attention_mask):
+    seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
+    indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
+    max_seqlen_in_batch = seqlens_in_batch.max().item()
+    cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
+    return (
+        indices,
+        cu_seqlens,
+        max_seqlen_in_batch,
+    )
+
+
+class GLU(nn.Module):
+
+    def __init__(self, d1, d2, bias=False):
+        super().__init__()
+
+        self.l1 = nn.Linear(d1, d2, bias=bias)
+        self.l2 = nn.Linear(d1, d2, bias=bias)
+        self.l3 = nn.Linear(d2, d1, bias=bias)
+
+    def forward(self, x):
+        o1 = self.l1(x)
+        o2 = self.l2(x)
+        output = o1 * o2
+        output = self.l3(output)
+        return output
+
+
+class MiniMaxM1LightningAttention(nn.Module):
+    def __init__(self, config: MiniMaxM1Config, layer_idx: Optional[int] = None):
+        super().__init__()
+        bias = False
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = getattr(config, 'head_dim', self.hidden_size // self.num_heads)
+
+        self.out_proj = nn.Linear(self.head_dim * self.num_heads, self.hidden_size, bias=bias)
+        self.act = get_activation_fn(config.hidden_act)
+        self.norm = MiniMaxM1RMSNorm(self.head_dim * self.num_heads)
+
+        self.qkv_proj = nn.Linear(self.hidden_size, 3 * self.head_dim * self.num_heads, bias=bias)
+        self.output_gate = nn.Linear(self.hidden_size, self.head_dim * self.num_heads, bias=bias)
+
+        # for inference only
+        self.offset = 0
+        self.layer_idx = layer_idx
+
+    def forward(
+            self,
+            hidden_states,
+            attn_mask: Optional[torch.Tensor] = None,  # (b, h, n, m)
+            output_attentions: bool = False,
+            past_key_value: Optional[Tuple[torch.Tensor]] = None,
+            use_cache: bool = False,
+            slope_rate: Optional[torch.Tensor] = None,
+            **kwargs
+    ):
+        if (not self.training) and (not do_eval):
+            return self.inference(
+                hidden_states,
+                attn_mask,
+                output_attentions,
+                past_key_value,
+                use_cache,
+                slope_rate,
+            )
+
+    def inference(
+            self,
+            x,
+            attn_mask: Optional[torch.Tensor] = None,  # (b, n)
+            output_attentions: bool = False,
+            past_key_value: Optional[Tuple[torch.Tensor]] = None,
+            use_cache: bool = False,
+            slope_rate: Optional[torch.Tensor] = None,  # (h, 1, 1)
+    ):
+        # x: b n d
+        b, n, d = x.shape
+        # linear map
+        qkv = self.act(self.qkv_proj(x))
+        new_shape = qkv.size()[:-1] + (self.num_heads, -1)
+        qkv = qkv.view(*new_shape)
+        q, k, v = torch.split(qkv, [self.head_dim] * 3, dim=3)
+        q = q.transpose(1, 2)
+        k = k.transpose(1, 2)
+        v = v.transpose(1, 2)
+
+        if past_key_value is None:
+            self.offset = q.shape[-2]
+        else:
+            self.offset += 1
+
+        # for align with metaseq
+        ratio = torch.exp(-slope_rate)
+
+        # only use for the first time
+        if past_key_value is None:
+            slope_rate = slope_rate.to(torch.float32)
+            if attn_mask is not None:
+                v = v.masked_fill((1 - attn_mask).unsqueeze(1).unsqueeze(-1).to(torch.bool), 0)
+            NUM_BLOCK = (n + BLOCK - 1) // BLOCK
+            b, h, n, d = q.shape
+            e = v.shape[-1]
+            # other
+            array = torch.arange(BLOCK).to(q) + 1
+            q_decay = torch.exp(-slope_rate * array.reshape(-1, 1))
+            k_decay = torch.exp(-slope_rate * (BLOCK - array.reshape(-1, 1)))
+            index = array[:, None] - array[None, :]
+            s_index = slope_rate * index[
+                None,
+                None,
+            ]
+            s_index = torch.where(index >= 0, -s_index, float("-inf"))
+            diag_decay = torch.exp(s_index)
+
+            kv = torch.zeros(b, h, d, e).to(torch.float32).to(q.device)
+            output = torch.empty((b, h, n, e), dtype=q.dtype, device=q.device)
+            for i in range(NUM_BLOCK):
+                si = i * BLOCK
+                ei = min(si + BLOCK, n)
+                m = ei - si
+                qi = q[:, :, si:ei].contiguous()
+                ki = k[:, :, si:ei].contiguous()
+                vi = v[:, :, si:ei].contiguous()
+                qkv_none_diag = torch.matmul(qi * q_decay[:, :m], kv).to(torch.float32)
+
+                # diag
+                qk = torch.matmul(qi, ki.transpose(-1, -2)).to(torch.float32) * diag_decay[:, :, :m, :m]
+                qkv_diag = torch.matmul(qk, vi.to(torch.float32))
+                block_decay = torch.exp(-slope_rate * m)
+                output[:, :, si:ei] = qkv_none_diag + qkv_diag
+                kv = block_decay * kv + torch.matmul((ki * k_decay[:, -m:]).transpose(-1, -2).to(vi.dtype), vi)
+
+        else:
+            kv = past_key_value
+            output = []
+            for i in range(n):
+                kv = ratio * kv + torch.einsum(
+                    "... n d, ... n e -> ... d e",
+                    k[:, :, i:i + 1],
+                    v[:, :, i:i + 1],
+                )
+                qkv = torch.einsum("... n e, ... e d -> ... n d", q[:, :, i:i + 1], kv.to(q.dtype))
+                output.append(qkv)
+            output = torch.concat(output, dim=-2)
+        # reshape
+        output = rearrange(output, "b h n d -> b n (h d)")
+        # normalize
+        output = self.norm(output)
+        # gate
+        output = F.sigmoid(self.output_gate(x)) * output
+        # outproj
+        output = self.out_proj(output)
+
+        attn_weights = None
+
+        return output, attn_weights, kv
+
+
+# Copied from transformers.models.llama.modeling_llama.LlamaRMSNorm with Llama->MiniMaxM1
+class MiniMaxM1RMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        MiniMaxM1RMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(torch.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+        return self.weight * hidden_states.to(input_dtype)
+
+
+# Copied from transformers.models.mistral.modeling_mistral.MistralRotaryEmbedding with Mistral->MiniMaxM1
+class MiniMaxM1RotaryEmbedding(nn.Module):
+    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None):
+        super().__init__()
+
+        self.dim = dim
+        self.max_position_embeddings = max_position_embeddings
+        self.base = base
+        inv_freq = 1.0 / (self.base ** (torch.arange(0, self.dim, 2, dtype=torch.int64).float().to(device) / self.dim))
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+
+        # Build here to make `torch.jit.trace` work.
+        self._set_cos_sin_cache(
+            seq_len=max_position_embeddings, device=self.inv_freq.device, dtype=torch.float32
+        )
+
+    def _set_cos_sin_cache(self, seq_len, device, dtype):
+        self.max_seq_len_cached = seq_len
+        t = torch.arange(self.max_seq_len_cached, device=device, dtype=torch.int64).type_as(self.inv_freq)
+
+        freqs = torch.outer(t, self.inv_freq)
+        # Different from paper, but it uses a different permutation in order to obtain the same calculation
+        emb = torch.cat((freqs, freqs), dim=-1)
+        self.register_buffer("cos_cached", emb.cos().to(dtype), persistent=False)
+        self.register_buffer("sin_cached", emb.sin().to(dtype), persistent=False)
+
+    def forward(self, x, seq_len=None):
+        # x: [bs, num_attention_heads, seq_len, head_size]
+        if seq_len > self.max_seq_len_cached:
+            self._set_cos_sin_cache(seq_len=seq_len, device=x.device, dtype=torch.float32)
+
+        return (
+            self.cos_cached[:seq_len].to(dtype=torch.float32),
+            self.sin_cached[:seq_len].to(dtype=torch.float32),
+        )
+
+
+# Copied from transformers.models.llama.modeling_llama.rotate_half
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2:]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+# Copied from transformers.models.mistral.modeling_mistral.apply_rotary_pos_emb
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids, unsqueeze_dim=1):
+    """Applies Rotary Position Embedding to the query and key tensors.
+
+    Args:
+        q (`torch.Tensor`): The query tensor.
+        k (`torch.Tensor`): The key tensor.
+        cos (`torch.Tensor`): The cosine part of the rotary embedding.
+        sin (`torch.Tensor`): The sine part of the rotary embedding.
+        position_ids (`torch.Tensor`):
+            The position indices of the tokens corresponding to the query and key tensors. For example, this can be
+            used to pass offsetted position ids when working with a KV-cache.
+        unsqueeze_dim (`int`, *optional*, defaults to 1):
+            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
+            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
+            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
+            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
+            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
+            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
+    Returns:
+        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
+    """
+    dtype = q.dtype
+    rot_dim = cos.shape[-1]
+    q_, q_pass = q[..., :rot_dim], q[..., rot_dim:]
+    k_, k_pass = k[..., :rot_dim], k[..., rot_dim:]
+    cos = cos[position_ids].unsqueeze(unsqueeze_dim)
+    sin = sin[position_ids].unsqueeze(unsqueeze_dim)
+    q_embed = (q_ * cos) + (rotate_half(q_) * sin)
+    k_embed = (k_ * cos) + (rotate_half(k_) * sin)
+    return torch.cat((q_embed, q_pass), dim=-1).to(dtype), torch.cat((k_embed, k_pass), dim=-1).to(dtype)
+
+
+# Copied from transformers.models.llama.modeling_llama.repeat_kv
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """
+    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+# Copied from transformers.models.mistral.modeling_mistral.MistralAttention with Mistral->MiniMaxM1
+class MiniMaxM1Attention(nn.Module):
+    """
+    Multi-headed attention from 'Attention Is All You Need' paper. Modified to use sliding window attention: Longformer
+    and "Generating Long Sequences with Sparse Transformers".
+    """
+
+    def __init__(self, config: MiniMaxM1Config, layer_idx: Optional[int] = None):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        if layer_idx is None:
+            logger.warning_once(
+                f"Instantiating {self.__class__.__name__} without passing a `layer_idx` is not recommended and will "
+                "lead to errors during the forward call if caching is used. Please make sure to provide a `layer_idx` "
+                "when creating this class."
+            )
+
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = getattr(config, 'head_dim', self.hidden_size // self.num_heads)
+        self.num_key_value_heads = config.num_key_value_heads
+        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
+        self.max_position_embeddings = config.max_position_embeddings
+        self.rope_theta = config.rope_theta
+        self.is_causal = True
+        self.attention_dropout = config.attention_dropout
+
+        self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
+        self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
+        self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
+        self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
+        self.rotary_dim = getattr(config, 'rotary_dim', self.head_dim)
+
+        self.rotary_emb = MiniMaxM1RotaryEmbedding(
+            self.rotary_dim,
+            max_position_embeddings=self.max_position_embeddings,
+            base=self.rope_theta,
+        )
+
+    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
+        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_value: Optional[Cache] = None,
+            output_attentions: bool = False,
+            use_cache: bool = False,
+            **kwargs,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        if "padding_mask" in kwargs:
+            warnings.warn(
+                "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
+            )
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            if self.layer_idx is None:
+                raise ValueError(
+                    f"The cache structure has changed since version v4.36. If you are using {self.__class__.__name__} "
+                    "for auto-regressive decoding with k/v caching, please make sure to initialize the attention class "
+                    "with a layer index."
+                )
+            kv_seq_len += past_key_value.get_usable_length(kv_seq_len, self.layer_idx)
+        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+
+        if past_key_value is not None:
+            cache_kwargs = {"sin": sin, "cos": cos}  # Specific to RoPE models
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        # repeat k/v heads if n_kv_heads < n_heads
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz, self.num_heads, q_len, kv_seq_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+
+            attn_weights = attn_weights + attention_mask
+
+        # upcast attention to fp32
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
+        attn_weights = nn.functional.dropout(attn_weights, p=self.attention_dropout, training=self.training)
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        attn_output = attn_output.transpose(1, 2).contiguous()
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+
+# Copied from transformers.models.mistral.modeling_mistral.MistralFlashAttention2 with Mistral->MiniMaxM1
+class MiniMaxM1FlashAttention2(MiniMaxM1Attention):
+    """
+    MiniMaxM1 flash attention module. This module inherits from `MiniMaxM1Attention` as the weights of the module stays
+    untouched. The only required change would be on the forward pass where it needs to correctly call the public API of
+    flash attention and deal with padding tokens in case the input contains any of them.
+    """
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2.__init__
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
+        # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
+        # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
+        self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_value: Optional[Union[Cache, Tuple[torch.Tensor]]] = None,
+            output_attentions: bool = False,
+            use_cache: bool = False,
+            **kwargs,
+    ):
+        if "padding_mask" in kwargs:
+            warnings.warn(
+                "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
+            )
+
+            # overwrite attention_mask with padding_mask
+            attention_mask = kwargs.pop("padding_mask")
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value[0].shape[-3]
+
+        # Because the input can be padded, the absolute sequence length depends on the max position id.
+        rotary_seq_len = max(kv_seq_len, position_ids[:, -1].max().item()) + 1
+        cos, sin = self.rotary_emb(value_states, seq_len=rotary_seq_len)
+
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+
+        use_sliding_windows = (
+                _flash_supports_window_size
+                and getattr(self.config, "sliding_window", None) is not None
+                and kv_seq_len > self.config.sliding_window
+        )
+
+        if not _flash_supports_window_size:
+            logger.warning_once(
+                "The current flash attention version does not support sliding window attention, for a more memory efficient implementation"
+                " make sure to upgrade flash-attn library."
+            )
+
+        dropout_rate = 0.0 if not self.training else self.attention_dropout
+
+        # In PEFT, usually we cast the layer norms in float32 for training stability reasons
+        # therefore the input hidden states gets silently casted in float32. Hence, we need
+        # cast them back in float16 just to be sure everything works as expected.
+        input_dtype = query_states.dtype
+        if input_dtype == torch.float32:
+            if torch.is_autocast_enabled():
+                target_dtype = torch.get_autocast_gpu_dtype()
+            # Handle the case where the model is quantized
+            elif hasattr(self.config, "_pre_quantization_dtype"):
+                target_dtype = self.config._pre_quantization_dtype
+            else:
+                target_dtype = self.q_proj.weight.dtype
+
+            logger.warning_once(
+                f"The input hidden states seems to be silently casted in float32, this might be related to"
+                f" the fact you have upcasted embedding or layer norm layers in float32. We will cast back the input in"
+                f" {target_dtype}."
+            )
+
+            query_states = query_states.to(target_dtype)
+            key_states = key_states.to(target_dtype)
+            value_states = value_states.to(target_dtype)
+
+        # Reshape to the expected shape for Flash Attention
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+
+        if past_key_value is not None:
+            # reuse k, v, for evaluation only
+            key_states = torch.cat([past_key_value[0], key_states], dim=-3)
+            value_states = torch.cat([past_key_value[1], value_states], dim=-3)
+        
+        past_key_value = (key_states, value_states) if use_cache else None
+
+        attn_output = self._flash_attention_forward(
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            q_len,
+            dropout=dropout_rate,
+            use_sliding_windows=use_sliding_windows,
+        )
+
+        attn_output = attn_output.reshape(bsz, q_len, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+    def _flash_attention_forward(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            query_length,
+            dropout=0.0,
+            softmax_scale=None,
+            use_sliding_windows=False,
+    ):
+        """
+        Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
+        first unpad the input, then computes the attention scores and pad the final attention scores.
+
+        Args:
+            query_states (`torch.Tensor`):
+                Input query states to be passed to Flash Attention API
+            key_states (`torch.Tensor`):
+                Input key states to be passed to Flash Attention API
+            value_states (`torch.Tensor`):
+                Input value states to be passed to Flash Attention API
+            attention_mask (`torch.Tensor`):
+                The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
+                position of padding tokens and 1 for the position of non-padding tokens.
+            dropout (`float`):
+                Attention dropout
+            softmax_scale (`float`, *optional*):
+                The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
+            use_sliding_windows (`bool`, *optional*):
+                Whether to activate sliding window attention.
+        """
+        if not self._flash_attn_uses_top_left_mask:
+            causal = self.is_causal
+        else:
+            # TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
+            causal = self.is_causal and query_length != 1
+
+        # Contains at least one padding token in the sequence
+        if attention_mask is not None:
+            batch_size = query_states.shape[0]
+            query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
+                query_states, key_states, value_states, attention_mask, query_length
+            )
+
+            cu_seqlens_q, cu_seqlens_k = cu_seq_lens
+            max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
+
+            if not use_sliding_windows:
+                attn_output_unpad = flash_attn_varlen_func(
+                    query_states,
+                    key_states,
+                    value_states,
+                    cu_seqlens_q=cu_seqlens_q,
+                    cu_seqlens_k=cu_seqlens_k,
+                    max_seqlen_q=max_seqlen_in_batch_q,
+                    max_seqlen_k=max_seqlen_in_batch_k,
+                    dropout_p=dropout,
+                    softmax_scale=softmax_scale,
+                    causal=causal,
+                )
+            else:
+                attn_output_unpad = flash_attn_varlen_func(
+                    query_states,
+                    key_states,
+                    value_states,
+                    cu_seqlens_q=cu_seqlens_q,
+                    cu_seqlens_k=cu_seqlens_k,
+                    max_seqlen_q=max_seqlen_in_batch_q,
+                    max_seqlen_k=max_seqlen_in_batch_k,
+                    dropout_p=dropout,
+                    softmax_scale=softmax_scale,
+                    causal=causal,
+                    window_size=(self.config.sliding_window, self.config.sliding_window),
+                )
+
+            attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
+        else:
+            if not use_sliding_windows:
+                attn_output = flash_attn_func(
+                    query_states,
+                    key_states,
+                    value_states,
+                    dropout,
+                    softmax_scale=softmax_scale,
+                    causal=causal,
+                )
+            else:
+                attn_output = flash_attn_func(
+                    query_states,
+                    key_states,
+                    value_states,
+                    dropout,
+                    softmax_scale=softmax_scale,
+                    causal=causal,
+                    window_size=(self.config.sliding_window, self.config.sliding_window),
+                )
+
+        return attn_output
+
+    def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
+        batch_size, kv_seq_len, num_heads, head_dim = key_layer.shape
+
+        # On the first iteration we need to properly re-create the padding mask
+        # by slicing it on the proper place
+        if kv_seq_len != attention_mask.shape[-1]:
+            attention_mask_num_tokens = attention_mask.shape[-1]
+            attention_mask = attention_mask[:, attention_mask_num_tokens - kv_seq_len:]
+
+        indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
+
+        key_layer = index_first_axis(key_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k)
+        value_layer = index_first_axis(value_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k)
+
+        if query_length == kv_seq_len:
+            query_layer = index_first_axis(
+                query_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k
+            )
+            cu_seqlens_q = cu_seqlens_k
+            max_seqlen_in_batch_q = max_seqlen_in_batch_k
+            indices_q = indices_k
+        elif query_length == 1:
+            max_seqlen_in_batch_q = 1
+            cu_seqlens_q = torch.arange(
+                batch_size + 1, dtype=torch.int32, device=query_layer.device
+            )  # There is a memcpy here, that is very bad.
+            indices_q = cu_seqlens_q[:-1]
+            query_layer = query_layer.squeeze(1)
+        else:
+            # The -q_len: slice assumes left padding.
+            attention_mask = attention_mask[:, -query_length:]
+            query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
+
+        return (
+            query_layer,
+            key_layer,
+            value_layer,
+            indices_q,
+            (cu_seqlens_q, cu_seqlens_k),
+            (max_seqlen_in_batch_q, max_seqlen_in_batch_k),
+        )
+
+
+class MiniMaxM1MLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = config.intermediate_size
+        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, x):
+        down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+        return down_proj
+
+
+class MiniMaxM1BlockSparseTop2MLP(nn.Module):
+    def __init__(self, config: MiniMaxM1Config):
+        super().__init__()
+        self.ffn_dim = config.intermediate_size
+        self.hidden_dim = config.hidden_size
+
+        self.w1 = nn.Linear(self.hidden_dim, self.ffn_dim, bias=False)
+        self.w2 = nn.Linear(self.ffn_dim, self.hidden_dim, bias=False)
+        self.w3 = nn.Linear(self.hidden_dim, self.ffn_dim, bias=False)
+
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, hidden_states):
+        current_hidden_states = self.act_fn(self.w1(hidden_states)) * self.w3(hidden_states)
+        current_hidden_states = self.w2(current_hidden_states)
+        return current_hidden_states
+
+
+class MiniMaxM1BLockSparseTop2MLP(MiniMaxM1BlockSparseTop2MLP):
+    def __init__(self, *args, **kwargs):
+        logger.warning_once(
+            "MiniMaxM1BLockSparseTop2MLP is deprecated by MiniMaxM1BlockSparseTop2MLP and will be removed in v4.40."
+        )
+        super().__init__(*args, **kwargs)
+
+
+class MiniMaxM1SparseMoeBlock(nn.Module):
+    """
+    This implementation is
+    strictly equivalent to standard MoE with full capacity (no
+    dropped tokens). It's faster since it formulates MoE operations
+    in terms of block-sparse operations to accomodate imbalanced
+    assignments of tokens to experts, whereas standard MoE either
+    (1) drop tokens at the cost of reduced performance or (2) set
+    capacity factor to number of experts and thus waste computation
+    and memory on padding.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        self.hidden_dim = config.hidden_size
+        self.ffn_dim = config.intermediate_size
+        self.num_experts = config.num_local_experts
+        self.top_k = config.num_experts_per_tok
+
+        # gating
+        self.gate = nn.Linear(self.hidden_dim, self.num_experts, bias=False)
+
+        self.experts = nn.ModuleList([MiniMaxM1BlockSparseTop2MLP(config) for _ in range(self.num_experts)])
+
+        # Jitter parameters
+        self.jitter_noise = config.router_jitter_noise
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        """ """
+        batch_size, sequence_length, hidden_dim = hidden_states.shape
+        if self.training and self.jitter_noise > 0:
+            hidden_states *= torch.empty_like(hidden_states).uniform_(1.0 - self.jitter_noise, 1.0 + self.jitter_noise)
+        hidden_states = hidden_states.view(-1, hidden_dim)
+        # router_logits: (batch * sequence_length, n_experts)
+        router_logits = self.gate(hidden_states)
+
+        routing_weights = F.softmax(router_logits, dim=1, dtype=torch.float)
+        routing_weights, selected_experts = torch.topk(routing_weights, self.top_k, dim=-1)
+        routing_weights /= routing_weights.sum(dim=-1, keepdim=True)
+        # we cast back to the input dtype
+        routing_weights = routing_weights.to(hidden_states.dtype)
+
+        final_hidden_states = torch.zeros(
+            (batch_size * sequence_length, hidden_dim), dtype=hidden_states.dtype, device=hidden_states.device
+        )
+
+        # One hot encode the selected experts to create an expert mask
+        # this will be used to easily index which expert is going to be sollicitated
+        expert_mask = torch.nn.functional.one_hot(selected_experts, num_classes=self.num_experts).permute(2, 1, 0)
+
+        # Loop over all available experts in the model and perform the computation on each expert
+        for expert_idx in range(self.num_experts):
+            expert_layer = self.experts[expert_idx]
+            idx, top_x = torch.where(expert_mask[expert_idx])
+
+            # Index the correct hidden states and compute the expert hidden state for
+            # the current expert. We need to make sure to multiply the output hidden
+            # states by `routing_weights` on the corresponding tokens (top-1 and top-2)
+            current_state = hidden_states[None, top_x].reshape(-1, hidden_dim)
+            current_hidden_states = expert_layer(current_state) * routing_weights[top_x, idx, None]
+
+            # However `index_add_` only support torch tensors for indexing so we'll use
+            # the `top_x` tensor here.
+            final_hidden_states.index_add_(0, top_x, current_hidden_states.to(hidden_states.dtype))
+        final_hidden_states = final_hidden_states.reshape(batch_size, sequence_length, hidden_dim)
+        return final_hidden_states, router_logits
+
+
+class MiniMaxM1DecoderLayer(nn.Module):
+    def __init__(self, config: MiniMaxM1Config, layer_idx: int):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+
+        self.self_attn = self.build_attn(config, layer_idx)
+
+        self.layer_idx = layer_idx
+
+        self.block_sparse_moe = MiniMaxM1SparseMoeBlock(config)
+        self.input_layernorm = MiniMaxM1RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = MiniMaxM1RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+        self.postnorm = getattr(config, 'postnorm', False)
+        self.layernorm_attention_alpha = getattr(config, 'layernorm_linear_attention_alpha', 1) \
+            if config.attention_type == 0 else getattr(config, 'layernorm_full_attention_alpha', 1)
+        self.layernorm_attention_beta = getattr(config, 'layernorm_linear_attention_beta', 1) \
+            if config.attention_type == 0 else getattr(config, 'layernorm_full_attention_beta', 1)
+        self.layernorm_mlp_alpha = getattr(config, 'layernorm_mlp_alpha', 1)
+        self.layernorm_mlp_beta = getattr(config, 'layernorm_mlp_beta', 1)
+
+        shared_intermediate = getattr(config, 'shared_intermediate_size', 0)
+        self.shared_moe = False
+        if shared_intermediate > 0:
+            self.shared_moe = True
+            self.shared_mlp = MiniMaxM1MLP(config)
+            self.coefficient = torch.nn.Linear(self.hidden_size, 1, bias=False)
+
+    def build_attn(self, config, layer_idx):
+        if config.attention_type == 0:
+            Attention_module = MiniMaxM1LightningAttention
+        else:
+            Attention_module = MiniMaxM1FlashAttention2
+
+        return Attention_module(
+            config,
+            layer_idx
+        )
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_value: Optional[Tuple[torch.Tensor]] = None,
+            output_attentions: Optional[bool] = False,
+            output_router_logits: Optional[bool] = False,
+            use_cache: Optional[bool] = False,
+            slope_rate: Optional[float] = None,
+            **kwargs,
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        if "padding_mask" in kwargs:
+            warnings.warn(
+                "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
+            )
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
+                `(batch, sequence_length)` where padding elements are indicated by 0.
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            output_router_logits (`bool`, *optional*):
+                Whether or not to return the logits of all the routers. They are useful for computing the router loss, and
+                should not be returned during inference.
+            use_cache (`bool`, *optional*):
+                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
+                (see `past_key_values`).
+        """
+
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+        if self.postnorm:
+            residual = hidden_states
+
+        hidden_states, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            position_ids=position_ids,
+            attn_mask=attention_mask,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+            slope_rate=slope_rate,
+        )
+
+        hidden_states = residual * self.layernorm_attention_alpha \
+                        + hidden_states * self.layernorm_attention_beta
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        if self.postnorm:
+            residual = hidden_states
+
+        moe_hidden_states, router_logits = self.block_sparse_moe(hidden_states)
+        if self.shared_moe:
+            output_mlp = self.shared_mlp(hidden_states)
+            weight_fp32 = self.coefficient.weight.float()
+            coef = hidden_states.to(torch.float32) @ weight_fp32.T
+            coef = torch.nn.functional.sigmoid(coef).to(hidden_states.dtype)
+            hidden_states = moe_hidden_states * (1 - coef) + output_mlp * coef
+        else:
+            hidden_states = moe_hidden_states
+
+        hidden_states = residual * self.layernorm_mlp_alpha \
+                        + hidden_states * self.layernorm_mlp_beta
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+
+        if output_router_logits:
+            outputs += (router_logits,)
+
+        return outputs
+
+
+MIXTRAL_START_DOCSTRING = r"""
+    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
+    etc.)
+
+    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`MiniMaxM1Config`]):
+            Model configuration class with all the parameters of the model. Initializing with a config file does not
+            load the weights associated with the model, only the configuration. Check out the
+            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+
+@add_start_docstrings(
+    "The bare MiniMaxM1 Model outputting raw hidden-states without any specific head on top.",
+    MIXTRAL_START_DOCSTRING,
+)
+# Copied from transformers.models.mistral.modeling_mistral.MistralPreTrainedModel with Mistral->MiniMaxM1
+class MiniMaxM1PreTrainedModel(PreTrainedModel):
+    config_class = MiniMaxM1Config
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["MiniMaxM1DecoderLayer"]
+    _skip_keys_device_placement = "past_key_values"
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
+
+    def _init_weights(self, module):
+        std = self.config.initializer_range
+        if isinstance(module, nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+
+MIXTRAL_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+            Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
+            it.
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            If `past_key_values` is used, optionally only the last `decoder_input_ids` have to be input (see
+            `past_key_values`).
+
+            If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`]
+            and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more
+            information on the default strategy.
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+        position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+            config.n_positions - 1]`.
+
+            [What are position IDs?](../glossary#position-ids)
+        past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of shape
+            `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`.
+
+            Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
+            blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
+
+            If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that
+            don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all
+            `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
+            model's internal embedding lookup matrix.
+        use_cache (`bool`, *optional*):
+            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
+            `past_key_values`).
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        output_router_logits (`bool`, *optional*):
+            Whether or not to return the logits of all the routers. They are useful for computing the router loss, and
+            should not be returned during inference.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare MiniMaxM1 Model outputting raw hidden-states without any specific head on top.",
+    MIXTRAL_START_DOCSTRING,
+)
+# Copied from transformers.models.mistral.modeling_mistral.MistralModel with MISTRAL->MIXTRAL,Mistral->MiniMaxM1
+class MiniMaxM1Model(MiniMaxM1PreTrainedModel):
+    """
+    Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`MiniMaxM1DecoderLayer`]
+
+    Args:
+        config: MiniMaxM1Config
+    """
+
+    def __init__(self, config: MiniMaxM1Config):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+        self.attn_type_list = config.attn_type_list
+        config_copy = copy.deepcopy(config)
+
+        self.layers = nn.ModuleList([])
+        for i in range(config.num_hidden_layers):
+            _config = copy.deepcopy(config)
+            if self.attn_type_list[i] == 0:
+                _config._attn_implementation = 'linear_attention'
+                _config.attention_type = 0
+            else:
+                _config._attn_implementation = config_copy._attn_implementation
+                _config.attention_type = 1
+            self.layers.append(MiniMaxM1DecoderLayer(_config, i))
+
+        self._attn_implementation = config_copy._attn_implementation
+        self.norm = MiniMaxM1RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+        self.gradient_checkpointing = False
+        self.slopes = self._build_slope_tensor(config.num_attention_heads)
+        # mask
+        self._linear_attn_mask = torch.empty(0)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+    @staticmethod
+    def _build_slope_tensor(n_attention_heads: int):
+
+        def get_slopes(n):
+
+            def get_slopes_power_of_2(n):
+                start = 2 ** (-(2 ** -(math.log2(n) - 3)))
+                ratio = start
+                return [start * ratio ** i for i in range(n)]
+
+            if math.log2(n).is_integer():
+                return get_slopes_power_of_2(
+                    n)  # In the paper, we only train models that have 2^a heads for some a. This function has
+            else:  # some good properties that only occur when the input is a power of 2. To maintain that even
+                closest_power_of_2 = 2 ** math.floor(
+                    math.log2(n))  # when the number of heads is not a power of 2, we use this workaround.
+                return (get_slopes_power_of_2(closest_power_of_2)
+                        + get_slopes(2 * closest_power_of_2)[0::2][:n - closest_power_of_2])
+
+        # h, 1, 1
+        slopes = torch.tensor(get_slopes(n_attention_heads), dtype=torch.float32).reshape(n_attention_heads, 1, 1)
+
+        return slopes
+
+    # Ignore copy
+    @add_start_docstrings_to_model_forward(MIXTRAL_INPUTS_DOCSTRING)
+    def forward(
+            self,
+            input_ids: torch.LongTensor = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[List[torch.FloatTensor]] = None,
+            inputs_embeds: Optional[torch.FloatTensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            output_router_logits: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, MoeModelOutputWithPast]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_router_logits = (
+            output_router_logits if output_router_logits is not None else self.config.output_router_logits
+        )
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # retrieve input_ids and inputs_embeds
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
+        elif input_ids is not None:
+            batch_size, seq_length = input_ids.shape
+            default_device = input_ids.device
+        elif inputs_embeds is not None:
+            batch_size, seq_length, _ = inputs_embeds.shape
+            default_device = inputs_embeds.device
+        else:
+            raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")
+
+        past_key_values_length = 0
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
+        seq_length_with_past = seq_length
+        if past_key_values is not None:
+            for idx in range(len(past_key_values)):
+                if self.attn_type_list[idx] == 1:
+                    past_key_values_length = past_key_values[idx][0].shape[-3]
+                    seq_length_with_past = seq_length_with_past + past_key_values_length
+                    break
+
+        if position_ids is None:
+            device = input_ids.device if input_ids is not None else inputs_embeds.device
+            position_ids = torch.arange(
+                past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
+            )
+            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+        else:
+            position_ids = position_ids.view(-1, seq_length).long()
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        if attention_mask is not None and self._attn_implementation == "flash_attention_2" and use_cache:
+            is_padding_right = attention_mask[:, -1].sum().item() != batch_size
+            if is_padding_right:
+                raise ValueError(
+                    "You are attempting to perform batched generation with padding_side='right'"
+                    " this may lead to unexpected behaviour for Flash Attention version of MiniMaxM1. Make sure to "
+                    " call `tokenizer.padding_side  = 'left'` before tokenizing the input. "
+                )
+        slope_rates = [self.slopes.to(default_device) for _ in range(len(self.layers))]
+        hidden_states = inputs_embeds
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        all_router_logits = () if output_router_logits else None
+        next_decoder_cache = () if use_cache else None
+
+        for idx, decoder_layer in enumerate(self.layers):
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            past_key_value = (past_key_values[idx] if past_key_values is not None else None)
+            attn_mask = attention_mask
+            slope_rate = slope_rates[idx]
+            slope_rate = slope_rate * (1 - idx / (len(self.layers) - 1) + 1e-5)
+            if self.gradient_checkpointing and self.training:
+                layer_outputs = self._gradient_checkpointing_func(
+                    decoder_layer.__call__,
+                    hidden_states,
+                    attention_mask,
+                    position_ids,
+                    past_key_values,
+                    output_attentions,
+                    output_router_logits,
+                    use_cache,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=attn_mask,
+                    position_ids=position_ids,
+                    past_key_value=past_key_value,
+                    output_attentions=output_attentions,
+                    output_router_logits=output_router_logits,
+                    use_cache=use_cache,
+                    slope_rate=slope_rate
+                )
+
+            hidden_states = layer_outputs[0]
+
+            if use_cache:
+                next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+            if output_router_logits:
+                all_router_logits += (layer_outputs[-1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+        next_cache = next_decoder_cache if use_cache else None
+        if not return_dict:
+            return tuple(
+                v
+                for v in [hidden_states, next_cache, all_hidden_states, all_self_attns, all_router_logits]
+                if v is not None
+            )
+        return MoeModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+            router_logits=all_router_logits,
+        )
+
+
+class MiniMaxM1ForCausalLM(MiniMaxM1PreTrainedModel):
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = MiniMaxM1Model(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+        self.router_aux_loss_coef = config.router_aux_loss_coef
+        self.num_experts = config.num_local_experts
+        self.num_experts_per_tok = config.num_experts_per_tok
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    @add_start_docstrings_to_model_forward(MIXTRAL_INPUTS_DOCSTRING)
+    @replace_return_docstrings(output_type=MoeCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
+    # Ignore copy
+    def forward(
+            self,
+            input_ids: torch.LongTensor = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[List[torch.FloatTensor]] = None,
+            inputs_embeds: Optional[torch.FloatTensor] = None,
+            labels: Optional[torch.LongTensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            output_router_logits: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, MoeCausalLMOutputWithPast]:
+        r"""
+        Args:
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, MiniMaxM1ForCausalLM
+
+        >>> model = MiniMaxM1ForCausalLM.from_pretrained(PATH_TO_WEIGHTS)
+        >>> tokenizer = AutoTokenizer.from_pretrained(PATH_TO_WEIGHTS)
+
+        >>> prompt = "Hey, are you conscious? Can you talk to me?"
+        >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+        ```"""
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_router_logits = (
+            output_router_logits if output_router_logits is not None else self.config.output_router_logits
+        )
+
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            output_router_logits=output_router_logits,
+            return_dict=return_dict,
+        )
+
+        hidden_states = outputs[0]
+        logits = self.lm_head(hidden_states)
+        logits = logits.float()
+
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            # Enable model parallelism
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+
+        aux_loss = None
+        if output_router_logits:
+            aux_loss = load_balancing_loss_func(
+                outputs.router_logits if return_dict else outputs[-1],
+                self.num_experts,
+                self.num_experts_per_tok,
+                attention_mask,
+            )
+            if labels is not None:
+                loss += self.router_aux_loss_coef * aux_loss.to(loss.device)  # make sure to reside in the same device
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            if output_router_logits:
+                output = (aux_loss,) + output
+            return (loss,) + output if loss is not None else output
+
+        torch.cuda.empty_cache()
+        return MoeCausalLMOutputWithPast(
+            loss=loss,
+            aux_loss=aux_loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            router_logits=outputs.router_logits,
+        )
+
+    def prepare_inputs_for_generation(
+            self,
+            input_ids,
+            past_key_values=None,
+            attention_mask=None,
+            inputs_embeds=None,
+            **kwargs,
+    ):
+        if past_key_values:
+            input_ids = input_ids[:, -1:]
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids}
+
+        model_inputs.update({
+            "past_key_values": past_key_values,
+            "use_cache": kwargs.get("use_cache"),
+            "attention_mask": attention_mask,
+        })
+        return model_inputs
+
+    @staticmethod
+    def _reorder_cache(past_key_values, beam_idx):
+        reordered_past = ()
+        for layer_past in past_key_values:
+            reordered_past += (
+                tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past),
+            )
+        return reordered_past
+
+
+@add_start_docstrings(
+    """
+    The MiniMaxM1 Model transformer with a sequence classification head on top (linear layer).
+
+    [`MiniMaxM1ForSequenceClassification`] uses the last token in order to do the classification, as other causal models
+    (e.g. GPT-2) do.
+
+    Since it does classification on the last token, it requires to know the position of the last token. If a
+    `pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If
+    no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the
+    padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in
+    each row of the batch).
+    """,
+    MIXTRAL_START_DOCSTRING,
+)
+# Copied from transformers.models.llama.modeling_llama.LlamaForSequenceClassification with Llama->MiniMaxM1, LLAMA->MIXTRAL
+class MiniMaxM1ForSequenceClassification(MiniMaxM1PreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+        self.model = MiniMaxM1Model(config)
+        self.score = nn.Linear(config.hidden_size, self.num_labels, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    @add_start_docstrings_to_model_forward(MIXTRAL_INPUTS_DOCSTRING)
+    def forward(
+            self,
+            input_ids: torch.LongTensor = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+            inputs_embeds: Optional[torch.FloatTensor] = None,
+            labels: Optional[torch.LongTensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
+            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
+            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        transformer_outputs = self.model(
+            input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        hidden_states = transformer_outputs[0]
+        logits = self.score(hidden_states)
+
+        if input_ids is not None:
+            batch_size = input_ids.shape[0]
+        else:
+            batch_size = inputs_embeds.shape[0]
+
+        if self.config.pad_token_id is None and batch_size != 1:
+            raise ValueError("Cannot handle batch sizes > 1 if no padding token is defined.")
+        if self.config.pad_token_id is None:
+            sequence_lengths = -1
+        else:
+            if input_ids is not None:
+                # if no pad token found, use modulo instead of reverse indexing for ONNX compatibility
+                sequence_lengths = torch.eq(input_ids, self.config.pad_token_id).int().argmax(-1) - 1
+                sequence_lengths = sequence_lengths % input_ids.shape[-1]
+                sequence_lengths = sequence_lengths.to(logits.device)
+            else:
+                sequence_lengths = -1
+
+        pooled_logits = logits[torch.arange(batch_size, device=logits.device), sequence_lengths]
+
+        loss = None
+        if labels is not None:
+            labels = labels.to(logits.device)
+            if self.config.problem_type is None:
+                if self.num_labels == 1:
+                    self.config.problem_type = "regression"
+                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
+                    self.config.problem_type = "single_label_classification"
+                else:
+                    self.config.problem_type = "multi_label_classification"
+
+            if self.config.problem_type == "regression":
+                loss_fct = MSELoss()
+                if self.num_labels == 1:
+                    loss = loss_fct(pooled_logits.squeeze(), labels.squeeze())
+                else:
+                    loss = loss_fct(pooled_logits, labels)
+            elif self.config.problem_type == "single_label_classification":
+                loss_fct = CrossEntropyLoss()
+                loss = loss_fct(pooled_logits.view(-1, self.num_labels), labels.view(-1))
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = BCEWithLogitsLoss()
+                loss = loss_fct(pooled_logits, labels)
+        if not return_dict:
+            output = (pooled_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return SequenceClassifierOutputWithPast(
+            loss=loss,
+            logits=pooled_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )

tokenizer_config.json

@@ -0,0 +1,10 @@
+{
+  "add_prefix_space": false,
+  "bos_token": "<beginning_of_sentence>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<end_of_sentence>",
+  "model_max_length": 40960000,
+  "tokenizer_class": "GPT2Tokenizer",
+  "unk_token": "<end_of_document>",
+  "chat_template": "{{ '<begin_of_document>' -}}{% set ns = namespace(system_prompt='') -%}{% for message in messages -%}{% if message['role'] == 'system' -%}{% set ns.system_prompt = ns.system_prompt + message['content'][0]['text'] -%}{% endif -%}{%- endfor -%}{% if ns.system_prompt != '' -%}{{ '<beginning_of_sentence>system ai_setting=assistant\n' + ns.system_prompt + '<end_of_sentence>\n' -}}{%- endif -%}{% if tools -%}{{ '<beginning_of_sentence>system tool_setting=tools\nYou are provided with these tools:\n<tools>\n' -}}{% for tool in tools -%}{{ tool | tojson ~ '\n' -}}{%- endfor -%}{{ '</tools>\n\nIf you need to call tools, please respond with <tool_calls></tool_calls> XML tags, and provide tool-name and json-object of arguments, following the format below:\n<tool_calls>\n{''name'': <tool-name-1>, ''arguments'': <args-json-object-1>}\n...\n</tool_calls><end_of_sentence>\n' -}}{%- endif -%}{% for message in messages -%}{% if message['role'] == 'user' -%}{{ '<beginning_of_sentence>user name=user\n' + message['content'][0]['text'] + '<end_of_sentence>\n' -}}{% elif message['role'] == 'assistant' -%}{{ '<beginning_of_sentence>ai name=assistant\n' -}}{% for content in message['content'] | selectattr('type', 'equalto', 'text') -%}{{ content['text'] -}}{%- endfor -%}{{ '<end_of_sentence>\n' -}}{% elif message['role'] == 'tool' -%}{{ '<beginning_of_sentence>tool name=tools\n' }} {%- for content in message['content'] -%}{{- 'tool name: ' + content['name'] + '\n' + 'tool result: ' + content['text'] + '\n\n' -}} {%- endfor -%}{{- '<end_of_sentence>\n' -}}{% endif -%}{%- endfor -%}{% if add_generation_prompt -%}{{ '<beginning_of_sentence>ai name=assistant\n' -}}{%- endif -%}"
+}

transformers_deployment_guide.md

@@ -0,0 +1,97 @@
+# 🚀 MiniMax Model Transformers Deployment Guide
+
+[Transformers中文版部署指南](./transformers_deployment_guide_cn.md)
+
+## 📖 Introduction
+
+This guide will help you deploy the MiniMax-M1 model using the [Transformers](https://huggingface.co/docs/transformers/index) library. Transformers is a widely used deep learning library that provides a rich collection of pre-trained models and flexible model operation interfaces.
+
+## 🛠️ Environment Setup
+
+### Installing Transformers
+
+```bash
+pip install transformers torch accelerate
+```
+
+## 📋 Basic Usage Example
+
+The pre-trained model can be used as follows:
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
+
+MODEL_PATH = "{MODEL_PATH}"
+model = AutoModelForCausalLM.from_pretrained(MODEL_PATH, device_map="auto", trust_remote_code=True)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, trust_remote_code=True)
+
+messages = [
+    {"role": "user", "content": "What is your favourite condiment?"},
+    {"role": "assistant", "content": "Well, I'm quite partial to a good squeeze of fresh lemon juice. It adds just the right amount of zesty flavour to whatever I'm cooking up in the kitchen!"},
+    {"role": "user", "content": "Do you have mayonnaise recipes?"}
+]
+
+text = tokenizer.apply_chat_template(
+    messages,
+    tokenize=False,
+    add_generation_prompt=True
+)
+
+model_inputs = tokenizer(text, return_tensors="pt").to(model.device)
+
+generation_config = GenerationConfig(
+    max_new_tokens=20,
+    eos_token_id=tokenizer.eos_token_id,
+    use_cache=True,
+)
+
+generated_ids = model.generate(**model_inputs, generation_config=generation_config)
+
+generated_ids = [
+    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
+]
+
+response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+print(response)
+```
+
+## ⚡ Performance Optimization
+
+### Speeding up with Flash Attention
+
+The code snippet above showcases inference without any optimization tricks. However, one can drastically speed up the model by leveraging [Flash Attention](../perf_train_gpu_one#flash-attention-2), which is a faster implementation of the attention mechanism used inside the model.
+
+First, make sure to install the latest version of Flash Attention 2 to include the sliding window attention feature:
+
+```bash
+pip install -U flash-attn --no-build-isolation
+```
+
+Also make sure that you have hardware that is compatible with Flash-Attention 2. Read more about it in the official documentation of the [Flash Attention repository](https://github.com/Dao-AILab/flash-attention). Additionally, ensure you load your model in half-precision (e.g. `torch.float16`).
+
+To load and run a model using Flash Attention-2, refer to the snippet below:
+
+```python
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+MODEL_PATH = "{MODEL_PATH}"
+model = AutoModelForCausalLM.from_pretrained(MODEL_PATH, trust_remote_code=True, torch_dtype=torch.float16, attn_implementation="flash_attention_2", device_map="auto")
+tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, trust_remote_code=True)
+
+prompt = "My favourite condiment is"
+
+model_inputs = tokenizer([prompt], return_tensors="pt").to("cuda")
+generated_ids = model.generate(**model_inputs, max_new_tokens=100, do_sample=True)
+response = tokenizer.batch_decode(generated_ids)[0]
+print(response)
+```
+
+## 📮 Getting Support
+
+If you encounter any issues while deploying the MiniMax-M1 model:
+- Please check our official documentation
+- Contact our technical support team through official channels
+- Submit an Issue on our GitHub repository
+
+We continuously optimize the deployment experience on Transformers and welcome your feedback!

transformers_deployment_guide_cn.md

@@ -0,0 +1,95 @@
+# 🚀 MiniMax 模型 Transformers 部署指南
+
+## 📖 简介
+
+本指南将帮助您使用 [Transformers](https://huggingface.co/docs/transformers/index) 库部署 MiniMax-M1 模型。Transformers 是一个广泛使用的深度学习库，提供了丰富的预训练模型和灵活的模型操作接口。
+
+## 🛠️ 环境准备
+
+### 安装 Transformers
+
+```bash
+pip install transformers torch accelerate
+```
+
+## 📋 基本使用示例
+
+预训练模型可以按照以下方式使用：
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
+
+MODEL_PATH = "{MODEL_PATH}"
+model = AutoModelForCausalLM.from_pretrained(MODEL_PATH, device_map="auto", trust_remote_code=True)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, trust_remote_code=True)
+
+messages = [
+    {"role": "user", "content": "What is your favourite condiment?"},
+    {"role": "assistant", "content": "Well, I'm quite partial to a good squeeze of fresh lemon juice. It adds just the right amount of zesty flavour to whatever I'm cooking up in the kitchen!"},
+    {"role": "user", "content": "Do you have mayonnaise recipes?"}
+]
+
+text = tokenizer.apply_chat_template(
+    messages,
+    tokenize=False,
+    add_generation_prompt=True
+)
+
+model_inputs = tokenizer(text, return_tensors="pt").to(model.device)
+
+generation_config = GenerationConfig(
+    max_new_tokens=20,
+    eos_token_id=tokenizer.eos_token_id,
+    use_cache=True,
+)
+
+generated_ids = model.generate(**model_inputs, generation_config=generation_config)
+
+generated_ids = [
+    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
+]
+
+response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+print(response)
+```
+
+## ⚡ 性能优化
+
+### 使用 Flash Attention 加速
+
+上面的代码片段展示了不使用任何优化技巧的推理过程。但通过利用 [Flash Attention](../perf_train_gpu_one#flash-attention-2)，可以大幅加速模型，因为它提供了模型内部使用的注意力机制的更快实现。
+
+首先，确保安装最新版本的 Flash Attention 2 以包含滑动窗口注意力功能：
+
+```bash
+pip install -U flash-attn --no-build-isolation
+```
+
+还要确保您拥有与 Flash-Attention 2 兼容的硬件。在[Flash Attention 官方仓库](https://github.com/Dao-AILab/flash-attention)的官方文档中了解更多信息。此外，请确保以半精度（例如 `torch.float16`）加载模型。
+
+要使用 Flash Attention-2 加载和运行模型，请参考以下代码片段：
+
+```python
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+MODEL_PATH = "{MODEL_PATH}"
+model = AutoModelForCausalLM.from_pretrained(MODEL_PATH, trust_remote_code=True, torch_dtype=torch.float16, attn_implementation="flash_attention_2", device_map="auto")
+tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, trust_remote_code=True)
+
+prompt = "My favourite condiment is"
+
+model_inputs = tokenizer([prompt], return_tensors="pt").to("cuda")
+generated_ids = model.generate(**model_inputs, max_new_tokens=100, do_sample=True)
+response = tokenizer.batch_decode(generated_ids)[0]
+print(response)
+```
+
+## 📮 获取支持
+
+如果您在部署 MiniMax-M1 模型过程中遇到任何问题：
+- 请查看我们的官方文档
+- 通过官方渠道联系我们的技术支持团队
+- 在我们的 GitHub 仓库提交 Issue
+
+我们会持续优化 Transformers 上的部署体验，欢迎您的反馈！

vllm_deployment_guide.md

@@ -0,0 +1,166 @@
+# 🚀 MiniMax Models vLLM Deployment Guide
+
+[vLLM中文版部署指南](./vllm_deployment_guide_cn.md)
+
+## 📖 Introduction
+
+We recommend using [vLLM](https://docs.vllm.ai/en/latest/) to deploy [MiniMax-M1](https://huggingface.co/MiniMaxAI/MiniMax-M1-40k) model. Based on our testing, vLLM performs excellently when deploying this model, with the following features:
+
+- 🔥 Outstanding service throughput performance
+- ⚡ Efficient and intelligent memory management
+- 📦 Powerful batch request processing capability
+- ⚙️ Deeply optimized underlying performance
+
+The MiniMax-M1 model can run efficiently on a single server equipped with 8 H800 or 8 H20 GPUs. In terms of hardware configuration, a server with 8 H800 GPUs can process context inputs up to 2 million tokens, while a server equipped with 8 H20 GPUs can support ultra-long context processing capabilities of up to 5 million tokens.
+
+## 💾 Obtaining MiniMax Models
+
+### MiniMax-M1 Model Obtaining
+
+You can download the model from our official HuggingFace repository: [MiniMax-M1](https://huggingface.co/MiniMaxAI/MiniMax-M1-40k)
+
+Download command:
+```
+pip install -U huggingface-hub
+huggingface-cli download MiniMaxAI/MiniMax-M1
+
+# If you encounter network issues, you can set a proxy
+export HF_ENDPOINT=https://hf-mirror.com
+```
+
+Or download using git:
+
+```bash
+git lfs install
+git clone https://huggingface.co/MiniMaxAI/MiniMax-M1-40k
+```
+
+⚠️ **Important Note**: Please ensure that [Git LFS](https://git-lfs.github.com/) is installed on your system, which is necessary for completely downloading the model weight files.
+
+## 🛠️ Deployment Options
+
+### Option 1: Deploy Using Docker (Recommended)
+
+To ensure consistency and stability of the deployment environment, we recommend using Docker for deployment.
+
+⚠️ **Version Requirements**: 
+- MiniMax-M1 model requires vLLM version 0.8.3 or later for full support
+- If you are using a Docker image with vLLM version lower than the required version, you will need to:
+  1. Update to the latest vLLM code
+  2. Recompile vLLM from source. Follow the compilation instructions in Solution 2 of the Common Issues section
+
+1. Get the container image:
+```bash
+docker pull vllm/vllm-openai:v0.8.3
+```
+
+2. Run the container:
+```bash
+# Set environment variables
+IMAGE=vllm/vllm-openai:v0.8.3
+MODEL_DIR=<model storage path>
+CODE_DIR=<code path>
+NAME=MiniMaxImage
+
+# Docker run configuration
+DOCKER_RUN_CMD="--network=host --privileged --ipc=host --ulimit memlock=-1 --shm-size=2gb --rm --gpus all --ulimit stack=67108864"
+
+# Start the container
+sudo docker run -it \
+    -v $MODEL_DIR:$MODEL_DIR \
+    -v $CODE_DIR:$CODE_DIR \
+    --name $NAME \
+    $DOCKER_RUN_CMD \
+    $IMAGE /bin/bash
+```
+
+
+### Option 2: Direct Installation of vLLM
+
+If your environment meets the following requirements:
+
+- CUDA 12.1
+- PyTorch 2.1
+
+You can directly install vLLM
+
+Installation command:
+```bash
+pip install vllm
+```
+
+💡 If you are using other environment configurations, please refer to the [vLLM Installation Guide](https://docs.vllm.ai/en/latest/getting_started/installation.html)
+
+## 🚀 Starting the Service
+
+### Launch MiniMax-M1 Service
+
+```bash
+export SAFETENSORS_FAST_GPU=1
+export VLLM_USE_V1=0
+python3 -m vllm.entrypoints.openai.api_server \
+--model <model storage path> \
+--tensor-parallel-size 8 \
+--trust-remote-code \
+--quantization experts_int8  \
+--max_model_len 4096 \
+--dtype bfloat16
+```
+
+### API Call Example
+
+```bash
+curl http://localhost:8000/v1/chat/completions \
+    -H "Content-Type: application/json" \
+    -d '{
+        "model": "MiniMaxAI/MiniMax-M1",
+        "messages": [
+            {"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant."}]},
+            {"role": "user", "content": [{"type": "text", "text": "Who won the world series in 2020?"}]}
+        ]
+    }'
+```
+
+## ❗ Common Issues
+
+### Module Loading Problems
+If you encounter the following error:
+```
+import vllm._C  # noqa
+ModuleNotFoundError: No module named 'vllm._C'
+```
+
+Or
+
+```
+MiniMax-M1 model is not currently supported
+```
+
+We provide two solutions:
+
+#### Solution 1: Copy Dependency Files
+```bash
+cd <working directory>
+git clone https://github.com/vllm-project/vllm.git
+cd vllm
+cp /usr/local/lib/python3.12/dist-packages/vllm/*.so vllm 
+cp -r /usr/local/lib/python3.12/dist-packages/vllm/vllm_flash_attn/* vllm/vllm_flash_attn
+```
+
+#### Solution 2: Install from Source
+```bash
+cd <working directory>
+git clone https://github.com/vllm-project/vllm.git
+
+cd vllm/
+pip install -e .
+```
+
+## 📮 Getting Support
+
+If you encounter any issues while deploying MiniMax-M1 model:
+- Please check our official documentation
+- Contact our technical support team through official channels
+- Submit an [Issue](https://github.com/MiniMax-AI/MiniMax-M1/issues) on our GitHub repository
+
+We will continuously optimize the deployment experience of this model and welcome your feedback!

vllm_deployment_guide_cn.md

@@ -0,0 +1,161 @@
+# 🚀 MiniMax 模型 vLLM 部署指南
+
+## 📖 简介
+
+我们推荐使用 [vLLM](https://docs.vllm.ai/en/latest/) 来部署 [MiniMax-M1](https://huggingface.co/MiniMaxAI/MiniMax-M1-40k) 模型。经过我们的测试，vLLM 在部署这个模型时表现出色，具有以下特点：
+
+- 🔥 卓越的服务吞吐量性能
+- ⚡ 高效智能的内存管理机制
+- 📦 强大的批量请求处理能力
+- ⚙️ 深度优化的底层性能
+
+MiniMax-M1 模型可在单台配备8个H800或8个H20 GPU的服务器上高效运行。在硬件配置方面，搭载8个H800 GPU的服务器可处理长达200万token的上下文输入，而配备8个H20 GPU的服务器则能够支持高达500万token的超长上下文处理能力。
+
+## 💾 获取 MiniMax 模型
+
+### MiniMax-M1 模型获取
+
+您可以从我们的官方 HuggingFace 仓库下载模型：[MiniMax-M1](https://huggingface.co/MiniMaxAI/MiniMax-M1-40k)
+
+下载命令：
+```
+pip install -U huggingface-hub
+huggingface-cli download MiniMaxAI/MiniMax-M1
+
+# 如果遇到网络问题，可以设置代理
+export HF_ENDPOINT=https://hf-mirror.com
+```
+
+或者使用 git 下载：
+
+```bash
+git lfs install
+git clone https://huggingface.co/MiniMaxAI/MiniMax-M1-40k
+```
+
+⚠️ **重要提示**：请确保系统已安装 [Git LFS](https://git-lfs.github.com/)，这对于完整下载模型权重文件是必需的。
+
+## 🛠️ 部署方案
+
+### 方案一：使用 Docker 部署（推荐）
+
+为确保部署环境的一致性和稳定性，我们推荐使用 Docker 进行部署。
+
+⚠️ **版本要求**：
+- MiniMax-M1 模型需要 vLLM 0.8.3 或更高版本才能获得完整支持
+
+1. 获取容器镜像：
+```bash
+docker pull vllm/vllm-openai:v0.8.3
+```
+
+2. 运行容器：
+```bash
+# 设置环境变量
+IMAGE=vllm/vllm-openai:v0.8.3
+MODEL_DIR=<模型存放路径>
+CODE_DIR=<代码路径>
+NAME=MiniMaxImage
+
+# Docker运行配置
+DOCKER_RUN_CMD="--network=host --privileged --ipc=host --ulimit memlock=-1 --shm-size=2gb --rm --gpus all --ulimit stack=67108864"
+
+# 启动容器
+sudo docker run -it \
+    -v $MODEL_DIR:$MODEL_DIR \
+    -v $CODE_DIR:$CODE_DIR \
+    --name $NAME \
+    $DOCKER_RUN_CMD \
+    $IMAGE /bin/bash
+```
+
+
+### 方案二：直接安装 vLLM
+
+如果您的环境满足以下要求：
+
+- CUDA 12.1
+- PyTorch 2.1
+
+可以直接安装 vLLM
+
+安装命令：
+```bash
+pip install vllm
+```
+
+💡 如果您使用其他环境配置，请参考 [vLLM 安装指南](https://docs.vllm.ai/en/latest/getting_started/installation.html)
+
+## 🚀 启动服务
+
+### 启动 MiniMax-M1 服务
+
+```bash
+export SAFETENSORS_FAST_GPU=1
+export VLLM_USE_V1=0
+python3 -m vllm.entrypoints.openai.api_server \
+--model <模型存放路径> \
+--tensor-parallel-size 8 \
+--trust-remote-code \
+--quantization experts_int8  \
+--max_model_len 4096 \
+--dtype bfloat16
+```
+
+### API 调用示例
+
+```bash
+curl http://localhost:8000/v1/chat/completions \
+    -H "Content-Type: application/json" \
+    -d '{
+        "model": "MiniMaxAI/MiniMax-M1",
+        "messages": [
+            {"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant."}]},
+            {"role": "user", "content": [{"type": "text", "text": "Who won the world series in 2020?"}]}
+        ]
+    }'
+```
+
+## ❗ 常见问题
+
+### 模块加载问题
+如果遇到以下错误：
+```
+import vllm._C  # noqa
+ModuleNotFoundError: No module named 'vllm._C'
+```
+
+或
+
+```
+当前并不支持 MiniMax-M1 模型
+```
+
+我们提供两种解决方案：
+
+#### 解决方案一：复制依赖文件
+```bash
+cd <工作目录>
+git clone https://github.com/vllm-project/vllm.git
+cd vllm
+cp /usr/local/lib/python3.12/dist-packages/vllm/*.so vllm 
+cp -r /usr/local/lib/python3.12/dist-packages/vllm/vllm_flash_attn/* vllm/vllm_flash_attn
+```
+
+#### 解决方案二：从源码安装
+```bash
+cd <工作目录>
+git clone https://github.com/vllm-project/vllm.git
+
+cd vllm/
+pip install -e .
+```
+
+## 📮 获取支持
+
+如果您在部署 MiniMax-M1 模型过程中遇到任何问题：
+- 请查看我们的官方文档
+- 通过官方渠道联系我们的技术支持团队
+- 在我们的 GitHub 仓库提交 [Issue](https://github.com/MiniMax-AI/MiniMax-M1/issues)
+
+我们会持续优化模型的部署体验，欢迎您的反馈！