filipealmeida commited on
Commit
13f22f0
1 Parent(s): a7cbcd5

Upload folder using huggingface_hub

Browse files
.gitattributes CHANGED
@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ ggml-model-Q4_0.gguf filter=lfs diff=lfs merge=lfs -text
37
+ ggml-model-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text
38
+ ggml-model-f16.gguf filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -1,3 +1,151 @@
1
  ---
2
- license: cc-by-3.0
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ license: cc-by-sa-3.0
3
+ datasets:
4
+ - VMware/open-instruct
5
+ language:
6
+ - en
7
+ library_name: transformers
8
+ pipeline_tag: text-generation
9
  ---
10
+
11
+ # Open LLama 7B v2 Open Instruct
12
+ - Model creator: [VMware](https://huggingface.co/VMware)
13
+ - Original model: [](https://huggingface.co/VMware/open-llama-7b-v2-open-instruct)
14
+
15
+ ## Description
16
+
17
+ This repo contains the GGUF model files for [Open LLama 7B v2 Open Instruct](https://huggingface.co/VMware/open-llama-7b-v2-open-instruct).
18
+
19
+ These files are compatible with [llama.cpp](https://github.com/ggerganov/llama.cpp).
20
+
21
+ # VMware/open-llama-7B-v2-open-instruct
22
+ Instruction-tuned version of the fully trained Open LLama 7B v2 model. The model is open for <b>COMMERCIAL USE</b>. <br>
23
+
24
+ - This model performs better on code compared to v1 due to the improvements made on the base model by the openlm-research team.
25
+ - The instruction model is trained on an improved instruction tuning dataset compared to v1
26
+
27
+ **NOTE**: The model was trained using the Alpaca prompt template <br>
28
+ **NOTE**: Fast tokenizer results in incorrect encoding, set the ```use_fast = False``` parameter, when instantiating the tokenizer
29
+
30
+
31
+ ## License
32
+ - CC BY-SA-3.0 **(Commercially Viable!)**
33
+ - Base Language Model ([openlm-research/open_llama_v2_7b](https://huggingface.co/openlm-research/open_llama_v2_7b)) is under apache-2.0
34
+ - Fine-Tuning Dataset ([VMware/open-instruct](https://huggingface.co/datasets/VMware/open-instruct)) is under cc-by-sa-3.0
35
+
36
+ ## Datasets used for Fine-Tuning
37
+
38
+ ### Open-instruct
39
+
40
+ **Open-instruct-v1**
41
+ - Mosaic/Dolly-HHRLHF + filtered OASST1 - cc by 3.0
42
+
43
+ **Subset of COT SUBMIX (FROM FLAN V2) Zeroshot examples**
44
+ - ESNLI - MIT
45
+ - ECQA - CDLA 1.0 - Sharing
46
+ - Strategy - MIT
47
+ - CREAK - MIT
48
+ - gsmk8 - MIT
49
+ - aqua - MIT
50
+ - qasc - Apache 2.0
51
+
52
+
53
+ ## Nomenclature
54
+
55
+ - Model : Open-llama-v2
56
+ - Model Size: 7B parameters
57
+ - Dataset: Open-instruct
58
+
59
+
60
+ ## Use in Transformers
61
+
62
+ ```
63
+ import os
64
+ import torch
65
+ from transformers import AutoModelForCausalLM, AutoTokenizer
66
+
67
+ model_name = 'VMware/open-llama-7b-v2-open-instruct'
68
+
69
+
70
+ tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=False)
71
+
72
+ model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, device_map='sequential')
73
+
74
+ prompt_template = "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n{instruction}\n\n### Response:"
75
+
76
+ prompt = """What is attention mechanism of a transformer model?
77
+ Write a python code to illustrate how attention works within a transformer model using numpy library. Donot use pytorch or tensorflow."""
78
+
79
+
80
+ inputt = prompt_template.format(instruction= prompt)
81
+ input_ids = tokenizer(inputt, return_tensors="pt").input_ids.to("cuda")
82
+
83
+ output1 = model.generate(input_ids, max_length=512)
84
+ input_length = input_ids.shape[1]
85
+ output1 = output1[:, input_length:]
86
+ output = tokenizer.decode(output1[0])
87
+
88
+ print(output)
89
+
90
+ ```
91
+
92
+
93
+ ### Output
94
+
95
+ Sure, I can help you with that!
96
+
97
+ Attention mechanisms in transformer models are typically implemented using the attention mechanism in the self-attention layer. Self-attention allows the model to focus on different parts of the input sequence when processing it. This is achieved by computing a set of attention weights, which are used to weigh the contribution of each input element to the output.
98
+
99
+ Here's an example code using NumPy to illustrate how attention works in a transformer model:
100
+
101
+ ```python
102
+ import numpy as np
103
+
104
+ def attention_weights(query, key, value, mask):
105
+ # Query, key, and value are input tensors. Mask is a tensor of zeros and ones that represents the attention mask.
106
+ # It is used to prevent the model from attending to certain positions in the input sequence if they are not relevant.
107
+ # The attention weights are the element-wise product of the query, key, and mask tensors.
108
+ # The result is a tensor of the same shape as the query tensor.
109
+
110
+ # Compute the dot product between the query tensor and the key tensor
111
+ dot = np.matmul(query, key)
112
+
113
+ # Compute the element-wise softmax of the dot product tensor
114
+ exp_dot = np.exp(dot)
115
+
116
+ # Multiply the dot product and the softmax of the dot product tensors
117
+ weights = dot * exp_dot
118
+
119
+ # Return the attention weights as a NumPy tensor
120
+ return weights
121
+
122
+ # Define the input sequence
123
+ query = np.array([[0.1, 0.2, 0.3], [0.4, 0.5, 0.6]])
124
+ key = np.array([[0.1, 0.2], [0.3, 0.4]])
125
+ value = np.array([[0.1, 0.2, 0.3], [0.4, 0.5, 0.6]])
126
+ mask = np.array([[False, True, True], [False, True, True]])
127
+
128
+ # Compute the attention weights
129
+ weights = attention_weights(query, key, value, mask)
130
+
131
+ # Print the attention weights
132
+ print(weights)
133
+ ```
134
+
135
+ In this example, the `attention_weights` function takes as input the query tensor, key tensor, value tensor, and mask tensor. It computes the dot product between the query and key tensors using the `np.matmul` function, and then applies a softmax function using the `np.exp` function to the element-wise dot product tensor. It then multiplies the dot product and softmax tensors using the `np.matmul` function, and returns the result as a NumPy tensor.
136
+
137
+ The `query`, `key`, and `value` tensors represent the input sequence to the transformer model. The `mask` tensor represents the attention mask, which is used to prevent the model from attending to certain positions in the input sequence if they are not relevant.
138
+
139
+ The output of the `attention_weights` function is a NumPy tensor that represents the attention weights for the input sequence. These weights are used by the transformer model to weigh the contribution of each input element to the output.
140
+
141
+ I hope this helps!</s>
142
+ <hr>
143
+
144
+
145
+ ## Finetuning details
146
+ The finetuning scripts will be available in our [RAIL Github Repository](https://github.com/vmware-labs/research-and-development-artificial-intelligence-lab/tree/main/instruction-tuning)
147
+
148
+
149
+ ## Evaluation
150
+
151
+ **TODO**
ggml-model-Q4_0.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:15709803fa2158f52687213188651e4d64a290101a4707339a0550dff4e64212
3
+ size 3825818912
ggml-model-Q8_0.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:243c305cdd04816039a5f20be4417b257c37fa4ef9f8ab6505b79305d463de22
3
+ size 7161101600
ggml-model-f16.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c35cfd775e365a800da010e01b379b86840c727d370114e2c87fa98da8c09e08
3
+ size 13478116608