jacksonkarel commited on
Commit
3894f96
1 Parent(s): 45e504f

Fix readme

Browse files
Files changed (1) hide show
  1. README.md +0 -230
README.md CHANGED
@@ -1,233 +1,3 @@
1
  ---
2
- <<<<<<< HEAD
3
- license: apache-2.0
4
- tags:
5
- - Composer
6
- - MosaicML
7
- - llm-foundry
8
- - StreamingDatasets
9
- datasets:
10
- - mc4
11
- - c4
12
- - togethercomputer/RedPajama-Data-1T
13
- - bigcode/the-stack
14
- - allenai/s2orc
15
- inference: false
16
- ---
17
-
18
- # MPT-7B
19
-
20
- MPT-7B is a decoder-style transformer pretrained from scratch on 1T tokens of English text and code.
21
- This model was trained by [MosaicML](https://www.mosaicml.com).
22
-
23
- MPT-7B is part of the family of MosaicPretrainedTransformer (MPT) models, which use a modified transformer architecture optimized for efficient training and inference.
24
-
25
- These architectural changes include performance-optimized layer implementations and the elimination of context length limits by replacing
26
- positional embeddings with Attention with Linear Biases ([ALiBi](https://arxiv.org/abs/2108.12409)).
27
- Thanks to these modifications, MPT models can be trained with high throughput efficiency and stable convergence.
28
- MPT models can also be served efficiently with both standard HuggingFace pipelines and NVIDIA's [FasterTransformer](https://github.com/NVIDIA/FasterTransformer).
29
-
30
- This model uses the MosaicML LLM codebase, which can be found in the [llm-foundry repository](https://github.com/mosaicml/llm-foundry). It was trained by MosaicML’s NLP team on the [MosaicML platform](https://www.mosaicml.com/training) for LLM pretraining, finetuning, and inference.
31
-
32
- ### How is this model different?
33
-
34
- MPT-7B is
35
-
36
- * **Licensed for the possibility of commercial use** (unlike [LLaMA](https://arxiv.org/abs/2302.13971)).
37
- * **Trained on a large amount of data** (1T tokens like [LLaMA](https://arxiv.org/abs/2302.13971) vs. 300B for [Pythia](https://github.com/EleutherAI/pythia), 300B for [OpenLLaMA](https://github.com/openlm-research/open_llama), and 800B for [StableLM](https://github.com/Stability-AI/StableLM)).
38
- * **Prepared to handle extremely long inputs** thanks to [ALiBi](https://arxiv.org/abs/2108.12409) (we finetuned [MPT-7B-StoryWriter-65k+](https://huggingface.co/mosaicml/mpt-7b-storywriter) on up to 65k inputs and can handle up to 84k vs. 2k-4k for other open source models).
39
- * **Capable of fast training and inference** (via [FlashAttention](https://arxiv.org/pdf/2205.14135.pdf) and [FasterTransformer](https://github.com/NVIDIA/FasterTransformer))
40
- * **Equipped with highly efficient open-source training code** via the [llm-foundry repository](https://github.com/mosaicml/llm-foundry)
41
-
42
- ### Models finetuned off MPT-7B:
43
-
44
- The following models are finetuned on MPT-7B:
45
-
46
- * [MPT-7B-StoryWriter-65k+](https://huggingface.co/mosaicml/mpt-7b-storywriter): a model designed to read and write fictional stories with super long context lengths.
47
- Built by finetuning MPT-7B with a context length of 65k tokens on a filtered fiction subset of the [books3 dataset](https://huggingface.co/datasets/the_pile_books3).
48
- At inference time, thanks to [ALiBi](https://arxiv.org/abs/2108.12409), MPT-7B-StoryWriter-65k+ can extrapolate even beyond 65k tokens.
49
- We demonstrate generations as long as 80k tokens on a single A100-80GB GPU in our [blogpost](www.mosaicml.com/blog/mpt-7b).
50
- * License: Apache 2.0
51
-
52
- * [MPT-7B-Instruct](https://huggingface.co/mosaicml/mpt-7b-instruct): a model for short-form instruction following.
53
- Built by finetuning MPT-7B on a [dataset](https://huggingface.co/datasets/mosaicml/dolly_hhrlhf) we also release, derived from the [Databricks Dolly-15k](https://huggingface.co/datasets/databricks/databricks-dolly-15k) and the [Anthropic Helpful and Harmless (HH-RLHF)](https://huggingface.co/datasets/Anthropic/hh-rlhf) datasets.
54
- * License: _CC-By-SA-3.0_
55
- * [Demo on Hugging Face Spaces](https://huggingface.co/spaces/mosaicml/mpt-7b-instruct)
56
-
57
- * [MPT-7B-Chat](https://huggingface.co/mosaicml/mpt-7b-chat): a chatbot-like model for dialogue generation.
58
- Built by finetuning MPT-7B on the [ShareGPT-Vicuna](https://huggingface.co/datasets/jeffwan/sharegpt_vicuna), [HC3](https://huggingface.co/datasets/Hello-SimpleAI/HC3),
59
- [Alpaca](https://huggingface.co/datasets/tatsu-lab/alpaca), [HH-RLHF](https://huggingface.co/datasets/Anthropic/hh-rlhf), and [Evol-Instruct](https://huggingface.co/datasets/victor123/evol_instruct_70k) datasets.
60
- * License: _CC-By-NC-SA-4.0_
61
- * [Demo on Hugging Face Spaces](https://huggingface.co/spaces/mosaicml/mpt-7b-chat)
62
-
63
- ## Model Date
64
-
65
- May 5, 2023
66
-
67
- ## Model License
68
-
69
- Apache-2.0
70
-
71
- ## Documentation
72
-
73
- * [Blog post: Introducing MPT-7B: A New Standard for Open-Source, Commercially Usable LLMs](https://www.mosaicml.com/blog/mpt-7b)
74
- * [Codebase (mosaicml/llm-foundry repo)](https://github.com/mosaicml/llm-foundry/)
75
- * Questions: Feel free to contact us via the [MosaicML Community Slack](https://mosaicml.me/slack)!
76
-
77
-
78
- ## How to Use
79
-
80
- This model is best used with the MosaicML [llm-foundry repository](https://github.com/mosaicml/llm-foundry) for training and finetuning.
81
-
82
- ```python
83
- import transformers
84
- model = transformers.AutoModelForCausalLM.from_pretrained(
85
- 'mosaicml/mpt-7b',
86
- trust_remote_code=True
87
- )
88
- ```
89
- Note: This model requires that `trust_remote_code=True` be passed to the `from_pretrained` method.
90
- This is because we use a custom `MPT` model architecture that is not yet part of the Hugging Face `transformers` package.
91
- `MPT` includes options for many training efficiency features such as [FlashAttention](https://arxiv.org/pdf/2205.14135.pdf), [ALiBi](https://arxiv.org/abs/2108.12409), [QK LayerNorm](https://arxiv.org/abs/2010.04245), and more.
92
-
93
- To use the optimized [triton implementation](https://github.com/openai/triton) of FlashAttention, you can load the model with `attn_impl='triton'` and move the model to `bfloat16`:
94
- ```python
95
- config = transformers.AutoConfig.from_pretrained(
96
- 'mosaicml/mpt-7b',
97
- trust_remote_code=True
98
- )
99
- config.attn_config['attn_impl'] = 'triton'
100
-
101
- model = transformers.AutoModelForCausalLM.from_pretrained(
102
- 'mosaicml/mpt-7b',
103
- config=config,
104
- torch_dtype=torch.bfloat16,
105
- trust_remote_code=True
106
- )
107
- model.to(device='cuda:0')
108
- ```
109
-
110
- Although the model was trained with a sequence length of 2048, ALiBi enables users to increase the maximum sequence length during finetuning and/or inference. For example:
111
-
112
- ```python
113
- config = transformers.AutoConfig.from_pretrained(
114
- 'mosaicml/mpt-7b',
115
- trust_remote_code=True
116
- )
117
- config.update({"max_seq_len": 4096})
118
- model = transformers.AutoModelForCausalLM.from_pretrained(
119
- 'mosaicml/mpt-7b',
120
- config=config,
121
- trust_remote_code=True
122
- )
123
- ```
124
-
125
- This model was trained with the [EleutherAI/gpt-neox-20b](https://huggingface.co/EleutherAI/gpt-neox-20b) tokenizer.
126
-
127
- ```python
128
- from transformers import AutoTokenizer
129
- tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-neox-20b")
130
- ```
131
-
132
- ## Model Description
133
-
134
- The architecture is a modification of a standard decoder-only transformer.
135
-
136
- The model has been modified from a standard transformer in the following ways:
137
- * It uses [FlashAttention](https://arxiv.org/pdf/2205.14135.pdf)
138
- * It uses [ALiBi (Attention with Linear Biases)](https://arxiv.org/abs/2108.12409) and does not use positional embeddings
139
- * It does not use biases
140
-
141
-
142
- | Hyperparameter | Value |
143
- |----------------|-------|
144
- |n_parameters | 6.7B |
145
- |n_layers | 32 |
146
- | n_heads | 32 |
147
- | d_model | 4096 |
148
- | vocab size | 50432 |
149
- | sequence length | 2048 |
150
-
151
-
152
-
153
- ## Training Data
154
-
155
- ### Streaming Datasets
156
-
157
- Data was formatted using the MosaicML [StreamingDataset](https://github.com/mosaicml/streaming) library to host our data in object storage and efficiently stream it to our compute cluster during training.
158
- StreamingDataset obviates the need to download the whole dataset before starting training, and allows instant resumption of training from any point in the dataset.
159
-
160
-
161
- ### Data Mix
162
-
163
- The model was trained for 1T tokens (with batch size 1760 and sequence length 2048). It was trained on the following data mix:
164
-
165
-
166
- | Data Source | Number of Tokens in Source | Proportion | Effective Number of Tokens | Epochs |
167
- |-------------|----------------------------|------------|----------------------------|--------|
168
- | mC4 3.1.0 - English | 417.99 B | 0.33 | 330 B | 0.14 |
169
- | C4 - English - SemDedup 80% | 100.42 B | 0.299 | 299 B | 2.98 |
170
- | RedPajama - CommonCrawl | 878.45 B | 0.1 | 100 B | 0.11 |
171
- | The Stack - Selected Languages | 463.78 B | 0.1 | 100 B | 0.22 |
172
- | RedPajama - Wikipedia - En | 4.87 B | 0.04 | 40 B | 8.21 |
173
- | The Stack - Markdown | 107.07 B | 0.035 | 35 B | 0.33 |
174
- | S2ORC | 48.85 B | 0.033 | 33 B | 0.68 |
175
- | RedPajama - Books | 26.02 B | 0.03 | 30B | 1.15 |
176
- | RedPajama - arXiv | 28.10 B | 0.019 | 19 B | 0.68 |
177
- | RedPajama - StackExchange | 20.54 B | 0.014 | 14 B |0.68 |
178
-
179
- Samples for each batch were selected from one of the datasets with the probability specified above.
180
- The examples were shuffled within each dataset, and each example was constructed from as many sequences from that dataset as were necessary to fill the 2048 sequence length.
181
-
182
- The data was tokenized using the [EleutherAI/gpt-neox-20b](https://huggingface.co/EleutherAI/gpt-neox-20b) tokenizer. This BPE tokenizer has a number of desirable characteristics,
183
- most of which are relevant for tokenizing code:
184
- (1) It was trained on a diverse mix of data that includes code (The Pile)
185
- (2) It applies consistent space delimitation, unlike the GPT2 tokenizer which tokenizes inconsistently depending on the presence of prefix spaces
186
- (3) It contains tokens for repeated space characters, which allows superior compression of text with large amounts of repeated space characters.
187
-
188
- The model vocabulary size of 50432 was set to be a multiple of 128 (as in [MEGATRON-LM](https://arxiv.org/abs/1909.08053)), model flop utilization (MFU) increased by up to four percentage points.
189
-
190
- ### Training Configuration
191
-
192
- This model was trained on 440 A100-40GBs for about 9.5 days using the [MosaicML Platform](https://www.mosaicml.com/platform).
193
- The model was trained with sharded data parallelism using [FSDP](https://pytorch.org/docs/stable/fsdp.html) and used the [LION](https://arxiv.org/abs/2302.06675) optimizer.
194
-
195
- ## Limitations and Biases
196
-
197
- _The following language is modified from [EleutherAI's GPT-NeoX-20B](https://huggingface.co/EleutherAI/gpt-neox-20b)_
198
-
199
- MPT-7B (Base) is **not** intended for deployment without finetuning.
200
- It should not be used for human-facing interactions without further guardrails and user consent.
201
-
202
- MPT-7B can produce factually incorrect output, and should not be relied on to produce factually accurate information.
203
- MPT-7B was trained on various public datasets.
204
- While great efforts have been taken to clean the pretraining data, it is possible that this model could generate lewd, biased or otherwise offensive outputs.
205
-
206
-
207
- ## MosaicML Platform
208
-
209
- If you're interested in [training](https://www.mosaicml.com/training) and [deploying](https://www.mosaicml.com/inference) your own MPT or LLMs on the MosaicML Platform, [sign up here](https://forms.mosaicml.com/demo?utm_source=huggingface&utm_medium=referral&utm_campaign=mpt-7b).
210
-
211
- ## Disclaimer
212
-
213
- The license on this model does not constitute legal advice. We are not responsible for the actions of third parties who use this model. Please cosult an attorney before using this model for commercial purposes.
214
-
215
- ## Citation
216
-
217
- Please cite this model using the following format:
218
-
219
- ```
220
- @online{MosaicML2023Introducing,
221
- author = {MosaicML NLP Team},
222
- title = {Introducing MPT-7B: A New Standard for Open-Source,
223
- ly Usable LLMs},
224
- year = {2023},
225
- url = {www.mosaicml.com/blog/mpt-7b},
226
- note = {Accessed: 2023-03-28}, % change this date
227
- urldate = {2023-03-28} % change this date
228
- }
229
- ```
230
- =======
231
  license: mit
232
  ---
233
- >>>>>>> 000ae17 (initial commit)
 
1
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2
  license: mit
3
  ---