Upload SVD_LlamaForCausalLM

config.json

@@ -0,0 +1,33 @@
+{
+  "_name_or_path": "huggingface_repos/llama-7b-hf-svdllm-60",
+  "architectures": [
+    "SVD_LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "auto_map": {
+    "AutoConfig": "config_llama.SVD_LlamaConfig",
+    "AutoModelForCausalLM": "modeling_svd_llama.SVD_LlamaForCausalLM"
+  },
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 2048,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 32,
+  "pad_token_id": 0,
+  "pretraining_tp": 1,
+  "ratio": 0.6,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.35.2",
+  "use_cache": true,
+  "vocab_size": 32000
+}

config_llama.py

@@ -0,0 +1,88 @@
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+LLAMA_PRETRAINED_CONFIG_ARCHIVE_MAP = {}
+
+
+class SVD_LlamaConfig(PretrainedConfig):
+    model_type = "llama"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=32000,
+        hidden_size=4096,
+        intermediate_size=11008,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=None,
+        hidden_act="silu",
+        max_position_embeddings=2048,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=1,
+        eos_token_id=2,
+        pretraining_tp=1,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        attention_bias=False,
+        ratio=1,
+        **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
+
+        # 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.pretraining_tp = pretraining_tp
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.rope_scaling = rope_scaling
+        self._rope_scaling_validation()
+        self.attention_bias = attention_bias
+        self.ratio = ratio
+
+        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,
+        )
+
+    def _rope_scaling_validation(self):
+        """
+        Validate the `rope_scaling` configuration.
+        """
+        if self.rope_scaling is None:
+            return
+
+        if not isinstance(self.rope_scaling, dict) or len(self.rope_scaling) != 2:
+            raise ValueError(
+                "`rope_scaling` must be a dictionary with with two fields, `type` and `factor`, "
+                f"got {self.rope_scaling}"
+            )
+        rope_scaling_type = self.rope_scaling.get("type", None)
+        rope_scaling_factor = self.rope_scaling.get("factor", None)
+        if rope_scaling_type is None or rope_scaling_type not in ["linear", "dynamic"]:
+            raise ValueError(
+                f"`rope_scaling`'s type field must be one of ['linear', 'dynamic'], got {rope_scaling_type}"
+            )
+        if rope_scaling_factor is None or not isinstance(rope_scaling_factor, float) or rope_scaling_factor <= 1.0:
+            raise ValueError(f"`rope_scaling`'s factor field must be a float > 1, got {rope_scaling_factor}")

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "pad_token_id": 0,
+  "transformers_version": "4.35.2"
+}

model-00001-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5aeac941297596695b4aaeb53aa1f515f505d6b1e5e1ad8f582dee223eee8717
+size 4995338024

model-00002-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bfc1d7bd946232ce04bd8abefa730099f916168551280146449308f0b2162fff
+size 3298699400

model.safetensors.index.json

@@ -0,0 +1,554 @@
+{
+  "metadata": {
+    "total_size": 8293974016
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00002-of-00002.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.mlp.down_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.mlp.down_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.gate_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.mlp.up_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.mlp.up_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.mlp.down_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.mlp.down_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.mlp.gate_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.mlp.gate_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.mlp.up_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.mlp.up_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.k_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.k_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.o_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.o_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.q_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.q_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.rotary_emb.inv_freq": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.v_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.v_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.mlp.down_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.mlp.down_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.mlp.gate_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.mlp.gate_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.mlp.up_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.mlp.up_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.k_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.k_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.o_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.o_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.q_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.q_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.rotary_emb.inv_freq": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.v_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.v_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.mlp.down_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.mlp.down_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.mlp.gate_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.mlp.gate_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.mlp.up_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.mlp.up_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.k_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.k_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.o_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.o_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.q_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.q_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.rotary_emb.inv_freq": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.v_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.v_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.mlp.down_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.mlp.down_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.mlp.gate_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.mlp.gate_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.mlp.up_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.mlp.up_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.k_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.k_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.o_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.o_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.q_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.q_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.rotary_emb.inv_freq": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.v_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.v_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.mlp.down_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.mlp.down_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.mlp.gate_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.mlp.gate_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.mlp.up_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.mlp.up_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.k_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.k_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.o_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.o_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.q_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.q_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.rotary_emb.inv_freq": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.v_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.v_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.mlp.down_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.mlp.down_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.mlp.gate_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.mlp.gate_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.mlp.up_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.mlp.up_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.k_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.k_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.o_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.o_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.q_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.q_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.rotary_emb.inv_freq": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.v_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.v_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.mlp.down_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.mlp.down_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.mlp.gate_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.mlp.gate_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.mlp.up_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.mlp.up_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.k_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.k_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.o_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.o_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.q_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.q_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.rotary_emb.inv_freq": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.v_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.v_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.mlp.down_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.mlp.down_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.mlp.gate_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.mlp.gate_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.mlp.up_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.mlp.up_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.k_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.k_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.o_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.o_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.q_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.q_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.rotary_emb.inv_freq": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.v_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.v_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.mlp.down_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.mlp.down_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.mlp.gate_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.mlp.gate_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.mlp.up_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.mlp.up_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.k_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.k_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.o_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.o_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.q_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.q_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.rotary_emb.inv_freq": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.v_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.v_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.mlp.down_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.mlp.down_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.mlp.gate_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.mlp.gate_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.mlp.up_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.mlp.up_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.k_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.k_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.o_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.o_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.q_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.q_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.rotary_emb.inv_freq": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.v_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.v_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.mlp.down_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.mlp.down_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.mlp.gate_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.mlp.gate_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.mlp.up_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.mlp.up_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.k_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.k_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.o_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.o_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.q_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.q_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.rotary_emb.inv_freq": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.v_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.v_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.mlp.down_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.mlp.down_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.mlp.gate_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.mlp.gate_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.mlp.up_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.mlp.up_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.k_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.k_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.o_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.o_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.q_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.q_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.rotary_emb.inv_freq": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.v_u_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.v_v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.down_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.down_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.gate_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.gate_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.up_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.up_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.k_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.k_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.o_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.o_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.q_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.q_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.rotary_emb.inv_freq": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.v_u_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.v_v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.norm.weight": "model-00002-of-00002.safetensors"
+  }
+}

modeling_svd_llama.py

@@ -0,0 +1,221 @@
+import math
+from typing import Optional, Tuple
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+
+from transformers.activations import ACT2FN
+from transformers.utils import logging
+from transformers import LlamaForCausalLM
+from .config_llama import SVD_LlamaConfig
+
+logger = logging.get_logger(__name__)
+
+_CONFIG_FOR_DOC = "SVD_LlamaConfig"
+
+class LlamaRMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        LlamaRMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        variance = hidden_states.to(torch.float32).pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+
+        # convert into half-precision if necessary
+        if self.weight.dtype in [torch.float16, torch.bfloat16]:
+            hidden_states = hidden_states.to(self.weight.dtype)
+
+        return self.weight * hidden_states
+
+
+class LlamaRotaryEmbedding(torch.nn.Module):
+    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None):
+        super().__init__()
+        inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2).float().to(device) / dim))
+        self.register_buffer("inv_freq", inv_freq)
+
+        # Build here to make `torch.jit.trace` work.
+        self.max_seq_len_cached = max_position_embeddings
+        t = torch.arange(self.max_seq_len_cached, device=self.inv_freq.device, dtype=self.inv_freq.dtype)
+        freqs = torch.einsum("i,j->ij", 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()[None, None, :, :], persistent=False)
+        self.register_buffer("sin_cached", emb.sin()[None, None, :, :], persistent=False)
+
+    def forward(self, x, seq_len=None):
+        # x: [bs, num_attention_heads, seq_len, head_size]
+        # This `if` block is unlikely to be run after we build sin/cos in `__init__`. Keep the logic here just in case.
+        if seq_len > self.max_seq_len_cached:
+            self.max_seq_len_cached = seq_len
+            t = torch.arange(self.max_seq_len_cached, device=x.device, dtype=self.inv_freq.dtype)
+            freqs = torch.einsum("i,j->ij", 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).to(x.device)
+            self.register_buffer("cos_cached", emb.cos()[None, None, :, :], persistent=False)
+            self.register_buffer("sin_cached", emb.sin()[None, None, :, :], persistent=False)
+        return (
+            self.cos_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
+            self.sin_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
+        )
+
+
+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)
+
+
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids):
+    gather_indices = position_ids[:, None, :, None]  # [bs, 1, seq_len, 1]
+    gather_indices = gather_indices.repeat(1, cos.shape[1], 1, cos.shape[3])
+    cos = torch.gather(cos.repeat(gather_indices.shape[0], 1, 1, 1), 2, gather_indices)
+    sin = torch.gather(sin.repeat(gather_indices.shape[0], 1, 1, 1), 2, gather_indices)
+    
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+class SVD_LlamaMLP(nn.Module):
+    def __init__(
+        self,
+        config: SVD_LlamaConfig
+    ):
+        super().__init__()
+        self.ratio = config.ratio
+        low_rank = int(config.intermediate_size * config.hidden_size * self.ratio / (config.intermediate_size + config.hidden_size))
+        self.gate_u_proj = nn.Linear(low_rank, config.intermediate_size, bias=False)
+        self.gate_v_proj = nn.Linear(config.hidden_size, low_rank, bias=False)
+        
+        self.down_u_proj = nn.Linear(low_rank, config.hidden_size, bias=False)
+        self.down_v_proj = nn.Linear(config.intermediate_size, low_rank, bias=False)
+        
+        self.up_u_proj = nn.Linear(low_rank, config.intermediate_size, bias=False)
+        self.up_v_proj = nn.Linear(config.hidden_size, low_rank, bias=False)
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, x):
+        up = self.up_u_proj(self.up_v_proj(x))
+        gate = self.gate_u_proj(self.gate_v_proj(x))
+        return self.down_u_proj(self.down_v_proj(self.act_fn(gate) * up))
+
+
+class SVD_LlamaAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: SVD_LlamaConfig):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.hidden_size // self.num_heads
+        self.max_position_embeddings = config.max_position_embeddings
+        self.ratio = config.ratio # 1 means no truncate, just keep normal attn
+
+        if (self.head_dim * self.num_heads) != self.hidden_size:
+            raise ValueError(
+                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
+                f" and `num_heads`: {self.num_heads})."
+            )
+        low_rank = int(self.hidden_size * self.ratio/2)
+        self.q_u_proj = nn.Linear(low_rank, self.num_heads * self.head_dim, bias=False)
+        self.q_v_proj = nn.Linear(self.hidden_size, low_rank, bias=False)
+
+        self.k_u_proj = nn.Linear(low_rank, self.num_heads * self.head_dim, bias=False)
+        self.k_v_proj = nn.Linear(self.hidden_size, low_rank, bias=False)
+
+        self.v_u_proj = nn.Linear(low_rank, self.num_heads * self.head_dim, bias=False)
+        self.v_v_proj = nn.Linear(self.hidden_size, low_rank, bias=False)
+
+        self.o_u_proj = nn.Linear(low_rank, self.hidden_size, bias=False)
+        self.o_v_proj = nn.Linear(self.num_heads * self.head_dim, low_rank, bias=False)
+
+        self.rotary_emb = LlamaRotaryEmbedding(self.head_dim, max_position_embeddings=self.max_position_embeddings)
+
+    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[Tuple[torch.Tensor]] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        bsz, q_len, _ = hidden_states.size()
+    
+        query_states = self.q_u_proj(self.q_v_proj(hidden_states)).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+
+        key_states = self.k_u_proj(self.k_v_proj(hidden_states)).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+
+        value_states = self.v_u_proj(self.v_v_proj(hidden_states)).view(bsz, q_len, self.num_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[-2]
+        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)
+        # [bsz, nh, t, hd]
+
+        if past_key_value is not None:
+            # reuse k, v, self_attention
+            key_states = torch.cat([past_key_value[0], key_states], dim=2)
+            value_states = torch.cat([past_key_value[1], value_states], dim=2)
+
+        past_key_value = (key_states, value_states) if use_cache else None
+
+        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
+            attn_weights = torch.max(attn_weights, torch.tensor(torch.finfo(attn_weights.dtype).min, device=attn_weights.device))
+
+        # upcast attention to fp32
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
+        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)
+        attn_output = attn_output.reshape(bsz, q_len, -1)
+
+        attn_output = self.o_u_proj(self.o_v_proj(attn_output))
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+    
+
+class SVD_LlamaForCausalLM(LlamaForCausalLM):
+    config_class = SVD_LlamaConfig
+    def __init__(self, config: SVD_LlamaConfig):
+        super().__init__(config)
+        for i in range(len(self.model.layers)):
+            self.model.layers[i].mlp = SVD_LlamaMLP(config=config)
+            self.model.layers[i].self_attn = SVD_LlamaAttention(config)