File size: 11,172 Bytes
fb0b1c8
 
3dafee6
 
fb0b1c8
3dafee6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2afa76c
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: apache-2.0
datasets:
- togethercomputer/RedPajama-Data-1T
---

# OpenLLaMA: An Open Reproduction of LLaMA


In this repo, we present a permissively licensed open source reproduction of Meta AI's [LLaMA](https://ai.facebook.com/blog/large-language-model-llama-meta-ai/) large language model. We are releasing a 7B and 3B model trained on 1T tokens, as well as the preview of a 13B model trained on 600B tokens. We provide PyTorch and JAX weights of pre-trained OpenLLaMA models, as well as evaluation results and comparison against the original LLaMA models. Please see the [project homepage of OpenLLaMA](https://github.com/openlm-research/open_llama) for more details.


## Weights Release, License and Usage

We release the weights in two formats: an EasyLM format to be use with our [EasyLM framework](https://github.com/young-geng/EasyLM), and a PyTorch format to be used with the [Hugging Face transformers](https://huggingface.co/docs/transformers/index) library. Both our training framework EasyLM and the checkpoint weights are licensed permissively under the Apache 2.0 license.

### Loading the Weights with Hugging Face Transformers
Preview checkpoints can be directly loaded from Hugging Face Hub. **Please note that it is advised to avoid using the Hugging Face fast tokenizer for now, as we’ve observed that the auto-converted fast tokenizer sometimes gives incorrect tokenizations.** This can be achieved by directly using the `LlamaTokenizer` class, or passing in the `use_fast=False` option for the `AutoTokenizer` class. See the following example for usage.

```python
import torch
from transformers import LlamaTokenizer, LlamaForCausalLM

model_path = 'openlm-research/open_llama_3b'
# model_path = 'openlm-research/open_llama_7b'

tokenizer = LlamaTokenizer.from_pretrained(model_path)
model = LlamaForCausalLM.from_pretrained(
    model_path, torch_dtype=torch.float16, device_map='auto',
)

prompt = 'Q: What is the largest animal?\nA:'
input_ids = tokenizer(prompt, return_tensors="pt").input_ids

generation_output = model.generate(
    input_ids=input_ids, max_new_tokens=32
)
print(tokenizer.decode(generation_output[0]))
```

For more advanced usage, please follow the [transformers LLaMA documentation](https://huggingface.co/docs/transformers/main/model_doc/llama).

### Evaluating with LM-Eval-Harness
The model can be evaluated with [lm-eval-harness](https://github.com/EleutherAI/lm-evaluation-harness). However, due to the aforementioned tokenizer issue, we need to avoid using the fast tokenizer to obtain the correct results. This can be achieved by passing in `use_fast=False` to [this part of lm-eval-harness](https://github.com/EleutherAI/lm-evaluation-harness/blob/4b701e228768052cfae9043dca13e82052ca5eea/lm_eval/models/huggingface.py#LL313C9-L316C10), as shown in the example below:

```python
tokenizer = self.AUTO_TOKENIZER_CLASS.from_pretrained(
    pretrained if tokenizer is None else tokenizer,
    revision=revision + ("/" + subfolder if subfolder is not None else ""),
    use_fast=False
)
```

### Loading the Weights with EasyLM

For using the weights in our EasyLM framework, please refer to the [LLaMA documentation of EasyLM](https://github.com/young-geng/EasyLM/blob/main/docs/llama.md). Note that unlike the original LLaMA model, our OpenLLaMA tokenizer and weights are trained completely from scratch so it is no longer needed to obtain the original LLaMA tokenizer and weights. Note that we use BOS (beginning of sentence) token (id=1) during training, so it is best to prepend this token for best performance during few-shot evaluation.



## Dataset and Training

We train our models on the [RedPajama](https://www.together.xyz/blog/redpajama) dataset released by [Together](https://www.together.xyz/), which is a reproduction of the LLaMA training dataset containing over 1.2 trillion tokens. We follow the exactly same preprocessing steps and training hyperparameters as the original LLaMA paper, including model architecture, context length, training steps, learning rate schedule, and optimizer.  The only difference between our setting and the original one is the dataset used: OpenLLaMA employs the RedPajama dataset rather than the one utilized by the original LLaMA.

We train the models on cloud TPU-v4s using [EasyLM](https://github.com/young-geng/EasyLM), a JAX based training pipeline we developed for training and fine-tuning large language models. We employ a combination of normal data parallelism and [fully sharded data parallelism (also know as ZeRO stage 3)](https://engineering.fb.com/2021/07/15/open-source/fsdp/) to balance the training throughput and memory usage. Overall we reach a throughput of over 2200 tokens / second / TPU-v4 chip for our 7B model.


## 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**        | GPT-J 6B | LLaMA 7B | OpenLLaMA 7B | OpenLLaMA 3B | OpenLLaMA 13B 600BT |
| ---------------------- | -------- | -------- | ------------ | ------------ | ------------------- |
| anli_r1/acc            | 0.32     | 0.35     | 0.33         | 0.33         | 0.33                |
| anli_r2/acc            | 0.34     | 0.34     | 0.36         | 0.32         | 0.35                |
| anli_r3/acc            | 0.35     | 0.37     | 0.38         | 0.35         | 0.38                |
| arc_challenge/acc      | 0.34     | 0.39     | 0.37         | 0.34         | 0.39                |
| arc_challenge/acc_norm | 0.37     | 0.41     | 0.38         | 0.37         | 0.42                |
| arc_easy/acc           | 0.67     | 0.68     | 0.72         | 0.69         | 0.74                |
| arc_easy/acc_norm      | 0.62     | 0.52     | 0.68         | 0.65         | 0.70                |
| ddboolq/acc            | 0.50     | 0.56     | 0.53         | 0.49         | 0.71                |
| hellaswag/acc          | 0.36     | 0.36     | 0.63         | 0.43         | 0.54                |
| hellaswag/acc_norm     | 0.66     | 0.73     | 0.72         | 0.67         | 0.73                |
| openbookqa/acc         | 0.29     | 0.29     | 0.30         | 0.27         | 0.30                |
| openbookqa/acc_norm    | 0.38     | 0.41     | 0.40         | 0.40         | 0.41                |
| piqa/acc               | 0.75     | 0.78     | 0.76         | 0.75         | 0.77                |
| piqa/acc_norm          | 0.76     | 0.78     | 0.77         | 0.76         | 0.78                |
| record/em              | 0.88     | 0.91     | 0.89         | 0.88         | 0.90                |
| record/f1              | 0.89     | 0.91     | 0.90         | 0.89         | 0.90                |
| rte/acc                | 0.54     | 0.56     | 0.60         | 0.58         | 0.65                |
| truthfulqa_mc/mc1      | 0.20     | 0.21     | 0.23         | 0.22         | 0.22                |
| truthfulqa_mc/mc2      | 0.36     | 0.34     | 0.35         | 0.35         | 0.35                |
| wic/acc                | 0.50     | 0.50     | 0.51         | 0.48         | 0.49                |
| winogrande/acc         | 0.64     | 0.68     | 0.67         | 0.62         | 0.67                |
| Average                | 0.51     | 0.53     | 0.55         | 0.52         | 0.56                |


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.


## Contact

We would love to get feedback from the community. If you have any questions, please open an issue or contact us.

OpenLLaMA is developed by:
[Xinyang Geng](https://young-geng.xyz/)* and [Hao Liu](https://www.haoliu.site/)* from Berkeley AI Research.
*Equal Contribution



## Acknowledgment

We thank the [Google TPU Research Cloud](https://sites.research.google/trc/about/) program for providing part of the computation resources. We’d like to specially thank Jonathan Caton from TPU Research Cloud for helping us organizing compute resources, Rafi Witten from the Google Cloud team and James Bradbury from the Google JAX team for helping us optimizing our training throughput. We’d also want to thank Charlie Snell, Gautier Izacard, Eric Wallace, Lianmin Zheng and our user community for the discussions and feedback.

The OpenLLaMA 13B model is trained in collaboration with [Stability AI](https://stability.ai/), and we thank Stability AI for providing the computation resources. We’d like to especially thank David Ha and Shivanshu Purohit for the coordinating the logistics and providing engineering support.


## Reference

If you found OpenLLaMA useful in your research or applications, please cite using the following BibTeX:
```
@software{openlm2023openllama,
  author = {Geng, Xinyang and Liu, Hao},
  title = {OpenLLaMA: An Open Reproduction of LLaMA},
  month = May,
  year = 2023,
  url = {https://github.com/openlm-research/open_llama}
}
```
```
@software{together2023redpajama,
  author = {Together Computer},
  title = {RedPajama-Data: An Open Source Recipe to Reproduce LLaMA training dataset},
  month = April,
  year = 2023,
  url = {https://github.com/togethercomputer/RedPajama-Data}
}
```
```
@article{touvron2023llama,
  title={Llama: Open and efficient foundation language models},
  author={Touvron, Hugo and Lavril, Thibaut and Izacard, Gautier and Martinet, Xavier and Lachaux, Marie-Anne and Lacroix, Timoth{\'e}e and Rozi{\`e}re, Baptiste and Goyal, Naman and Hambro, Eric and Azhar, Faisal and others},
  journal={arXiv preprint arXiv:2302.13971},
  year={2023}
}
```






# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_openlm-research__open_llama_3b)

| Metric                | Value                     |
|-----------------------|---------------------------|
| Avg.                  | 33.52   |
| ARC (25-shot)         | 39.85          |
| HellaSwag (10-shot)   | 62.65    |
| MMLU (5-shot)         | 26.94         |
| TruthfulQA (0-shot)   | 34.97   |
| Winogrande (5-shot)   | 64.72   |
| GSM8K (5-shot)        | 0.45        |
| DROP (3-shot)         | 5.06         |