File size: 10,523 Bytes
b1ffc64
7ba0cd6
 
 
 
 
 
 
 
 
 
b1ffc64
 
7ba0cd6
 
13d7a7a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7ba0cd6
 
 
52b08a9
7ba0cd6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
---
language:
- en
library_name: transformers
tags:
- gpt
- llm
- large language model
inference: false
thumbnail: >-
  https://h2o.ai/etc.clientlibs/h2o/clientlibs/clientlib-site/resources/images/favicon.ico
license: apache-2.0
---
# Model Card

** The Best 3B Model! Surpassing dolly-v2-12b **

The best 3B model on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard), with performance surpassing dolly-v2-12b

| Metric                | Value |
|-----------------------|-------|
| MMLU (5-shot)         | 27.1  |
| ARC (25-shot)         | 42.2  |
| HellaSwag (10-shot)   | 71.5  |
| TruthfulQA (0-shot)   | 36.7  |
| Avg.                  | 44.4  | 

We use state-of-the-art [Language Model Evaluation Harness](https://github.com/EleutherAI/lm-evaluation-harness) to run the benchmark tests above.



## Summary

We have fine-tuned the open-lama model and surpassed the original model in multiple evaluation subtasks, making it currently the best performing 3B model with comparable performance to llama-7b
- Base model: [openlm-research/open_llama_3b_v2](https://huggingface.co/openlm-research/open_llama_3b_v2)

## Usage

To use the model with the `transformers` library on a machine with GPUs, first make sure you have the `transformers`, `accelerate` and `torch` libraries installed.

```bash
pip install transformers==4.29.2
pip install accelerate==0.19.0
pip install torch==2.0.0
```

```python
import torch
from transformers import pipeline

generate_text = pipeline(
    model="CobraMamba/mamba-gpt-3b",
    torch_dtype="auto",
    trust_remote_code=True,
    use_fast=False,
    device_map={"": "cuda:0"},
)

res = generate_text(
    "Why is drinking water so healthy?",
    min_new_tokens=2,
    max_new_tokens=1024,
    do_sample=False,
    num_beams=1,
    temperature=float(0.3),
    repetition_penalty=float(1.2),
    renormalize_logits=True
)
print(res[0]["generated_text"])
```

You can print a sample prompt after the preprocessing step to see how it is feed to the tokenizer:

```python
print(generate_text.preprocess("Why is drinking water so healthy?")["prompt_text"])
```

```bash
<|prompt|>Why is drinking water so healthy?</s><|answer|>
```

Alternatively, you can download the mamba_gpt_pipeline.py, store it alongside your notebook, and construct the pipeline yourself from the loaded model and tokenizer. If the model and the tokenizer are fully supported in the `transformers` package, this will allow you to set `trust_remote_code=False`.

```python
import torch
from mamba_gpt_pipeline.py import MambaGPTTextGenerationPipeline
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained(
    "CobraMamba/mamba-gpt-3b",
    use_fast=False,
    padding_side="left",
    trust_remote_code=False,
)
model = AutoModelForCausalLM.from_pretrained(
    "CobraMamba/mamba-gpt-3b",
    torch_dtype="auto",
    device_map={"": "cuda:0"},
    trust_remote_code=False,
)
generate_text = MambaGPTTextGenerationPipeline(model=model, tokenizer=tokenizer)

res = generate_text(
    "Why is drinking water so healthy?",
    min_new_tokens=2,
    max_new_tokens=1024,
    do_sample=False,
    num_beams=1,
    temperature=float(0.3),
    repetition_penalty=float(1.2),
    renormalize_logits=True
)
print(res[0]["generated_text"])
```


You may also construct the pipeline from the loaded model and tokenizer yourself and consider the preprocessing steps:

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "CobraMamba/mamba-gpt-3b"  # either local folder or huggingface model name
# Important: The prompt needs to be in the same format the model was trained with.
# You can find an example prompt in the experiment logs.
prompt = "<|prompt|>How are you?</s><|answer|>"

tokenizer = AutoTokenizer.from_pretrained(
    model_name,
    use_fast=False,
    trust_remote_code=False,
)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map={"": "cuda:0"},
    trust_remote_code=False,
)
model.cuda().eval()
inputs = tokenizer(prompt, return_tensors="pt", add_special_tokens=False).to("cuda")

# generate configuration can be modified to your needs
tokens = model.generate(
    **inputs,
    min_new_tokens=2,
    max_new_tokens=1024,
    do_sample=False,
    num_beams=1,
    temperature=float(0.3),
    repetition_penalty=float(1.2),
    renormalize_logits=True
)[0]

tokens = tokens[inputs["input_ids"].shape[1]:]
answer = tokenizer.decode(tokens, skip_special_tokens=True)
print(answer)
```

## Model Architecture

```
LlamaForCausalLM(
  (model): LlamaModel(
    (embed_tokens): Embedding(32000, 4096, padding_idx=0)
    (layers): ModuleList(
      (0-31): 32 x LlamaDecoderLayer(
        (self_attn): LlamaAttention(
          (q_proj): Linear(in_features=4096, out_features=4096, bias=False)
          (k_proj): Linear(in_features=4096, out_features=4096, bias=False)
          (v_proj): Linear(in_features=4096, out_features=4096, bias=False)
          (o_proj): Linear(in_features=4096, out_features=4096, bias=False)
          (rotary_emb): LlamaRotaryEmbedding()
        )
        (mlp): LlamaMLP(
          (gate_proj): Linear(in_features=4096, out_features=11008, bias=False)
          (down_proj): Linear(in_features=11008, out_features=4096, bias=False)
          (up_proj): Linear(in_features=4096, out_features=11008, bias=False)
          (act_fn): SiLUActivation()
        )
        (input_layernorm): LlamaRMSNorm()
        (post_attention_layernorm): LlamaRMSNorm()
      )
    )
    (norm): LlamaRMSNorm()
  )
  (lm_head): Linear(in_features=4096, out_features=32000, bias=False)
)
```

## Evaluation
We evaluated OpenLLaMA on a wide range of tasks using [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness).  The LLaMA results are generated by running the original LLaMA model on the same evaluation metrics. We note that our results for the LLaMA model differ slightly from the original LLaMA paper, which we believe is a result of different evaluation protocols. Similar differences have been reported in [this issue of lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness/issues/443). Additionally, we present the results of GPT-J, a 6B parameter model trained on the [Pile](https://pile.eleuther.ai/) dataset by [EleutherAI](https://www.eleuther.ai/).

The original LLaMA model was trained for 1 trillion tokens and GPT-J was trained for 500 billion tokens.  We present the results in the table below. OpenLLaMA exhibits comparable performance to the original LLaMA and GPT-J across a majority of tasks, and outperforms them in some tasks.


| **Task/Metric**        | finetuned-GPT 3B  | OpenLLaMA 3B |
| ---------------------- | -------- | ------------ |
| anli_r1/acc            | **0.35** | 0.33         |      
| anli_r2/acc            | **0.33** | 0.32         | 
| anli_r3/acc            | 0.35     | 0.35         | 
| arc_challenge/acc      | **0.35** | 0.34         | 
| arc_challenge/acc_norm | 0.37     | 0.37         |
| arc_easy/acc           | **0.71** | 0.69         | 
| arc_easy/acc_norm      | 0.65     | 0.65         | 
| boolq/acc              | **0.72** | 0.66         | 
| hellaswag/acc          | **0.49** | 0.43         | 
| hellaswag/acc_norm     | 0.66     | **0.67**         |
| openbookqa/acc         | 0.26     | **0.27**         | 
| openbookqa/acc_norm    | 0.40     | 0.40         | 
| piqa/acc               | **0.76** | 0.75         | 
| piqa/acc_norm          | 0.76     | 0.76         | 
| record/em              | 0.88     | 0.88         | 
| record/f1              | 0.88     | **0.89**         | 
| rte/acc                | 0.55     | **0.58**         | 
| truthfulqa_mc/mc1      | **0.27** | 0.22         | 
| truthfulqa_mc/mc2      | **0.37** | 0.35         |
| wic/acc                | **0.49** | 0.48         | 
| winogrande/acc         | **0.63** | 0.62         | 
| Average                | **0.53** | 0.52         | 


We removed the task CB and WSC from our benchmark, as our model performs suspiciously well on these two tasks. We hypothesize that there could be a benchmark data contamination in the training set.

## Disclaimer

Please read this disclaimer carefully before using the large language model provided in this repository. Your use of the model signifies your agreement to the following terms and conditions.

- Biases and Offensiveness: The large language model is trained on a diverse range of internet text data, which may contain biased, racist, offensive, or otherwise inappropriate content. By using this model, you acknowledge and accept that the generated content may sometimes exhibit biases or produce content that is offensive or inappropriate. The developers of this repository do not endorse, support, or promote any such content or viewpoints.
- Limitations: The large language model is an AI-based tool and not a human. It may produce incorrect, nonsensical, or irrelevant responses. It is the user's responsibility to critically evaluate the generated content and use it at their discretion.
- Use at Your Own Risk: Users of this large language model must assume full responsibility for any consequences that may arise from their use of the tool. The developers and contributors of this repository shall not be held liable for any damages, losses, or harm resulting from the use or misuse of the provided model.
- Ethical Considerations: Users are encouraged to use the large language model responsibly and ethically. By using this model, you agree not to use it for purposes that promote hate speech, discrimination, harassment, or any form of illegal or harmful activities.
- Reporting Issues: If you encounter any biased, offensive, or otherwise inappropriate content generated by the large language model, please report it to the repository maintainers through the provided channels. Your feedback will help improve the model and mitigate potential issues.
- Changes to this Disclaimer: The developers of this repository reserve the right to modify or update this disclaimer at any time without prior notice. It is the user's responsibility to periodically review the disclaimer to stay informed about any changes.

By using the large language model provided in this repository, you agree to accept and comply with the terms and conditions outlined in this disclaimer. If you do not agree with any part of this disclaimer, you should refrain from using the model and any content generated by it.