Composer MosaicML llm-foundry StreamingDatasets

MPT-7B-8k

MPT-7B-8k is a decoder-style transformer pretrained starting from MPT-7B, but updating the sequence length to 8k and training for an additional 500B tokens, resulting in a total of 1.5T tokens of text and code. This model was trained by MosaicML.

MPT-7B-8k is part of the family of Mosaic Pretrained Transformer (MPT) models, which use a modified transformer architecture optimized for efficient training and inference.

These architectural changes include performance-optimized layer implementations and the elimination of context length limits by replacing positional embeddings with Attention with Linear Biases (ALiBi). Thanks to these modifications, MPT models can be trained with high throughput efficiency and stable convergence. MPT models can also be served efficiently with both standard HuggingFace pipelines and NVIDIA's FasterTransformer.

This model uses the MosaicML LLM codebase, which can be found in the llm-foundry repository. It was trained by MosaicML’s NLP team on the MosaicML platform for LLM pretraining, finetuning, and inference.

How is this model different?

MPT-7B-8k is

Models finetuned off MPT-7B-8k:

The following models are finetuned on MPT-7B-8k:

Model Date

July 18, 2023

Model License

Apache-2.0

Documentation

How to Use

This model is best used with the MosaicML llm-foundry repository for training and finetuning.

import transformers
model = transformers.AutoModelForCausalLM.from_pretrained(
  'mosaicml/mpt-7b-8k',
  trust_remote_code=True
)

Note: This model requires that trust_remote_code=True be passed to the from_pretrained method. This is because we use a custom MPT model architecture that is not yet part of the Hugging Face transformers package. MPT includes options for many training efficiency features such as FlashAttention, ALiBi, QK LayerNorm, and more.

To use the optimized triton implementation of FlashAttention, you can load the model on GPU (cuda:0) with attn_impl='triton' and with bfloat16 precision:

import torch
import transformers

name = 'mosaicml/mpt-7b-8k'

config = transformers.AutoConfig.from_pretrained(name, trust_remote_code=True)
config.attn_config['attn_impl'] = 'triton'
config.init_device = 'cuda:0' # For fast initialization directly on GPU!

model = transformers.AutoModelForCausalLM.from_pretrained(
  name,
  config=config,
  torch_dtype=torch.bfloat16, # Load model weights in bfloat16
  trust_remote_code=True
)

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:

import transformers

name = 'mosaicml/mpt-7b-8k'

config = transformers.AutoConfig.from_pretrained(name, trust_remote_code=True)
config.max_seq_len = 10000 # (input + output) tokens can now be up to 10000

model = transformers.AutoModelForCausalLM.from_pretrained(
  name,
  config=config,
  trust_remote_code=True
)

This model was trained with the MPT-7B-8k tokenizer which is identical to the EleutherAI/gpt-neox-20b tokenizer.

from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained('mosaicml/mpt-7b-8k')

The model can then be used, for example, within a text-generation pipeline.
Note: when running Torch modules in lower precision, it is best practice to use the torch.autocast context manager.

from transformers import pipeline

with torch.autocast('cuda', dtype=torch.bfloat16):
    inputs = tokenizer('Here is a recipe for vegan banana bread:\n', return_tensors="pt").to('cuda')
    outputs = model.generate(**inputs, max_new_tokens=100)
    print(tokenizer.batch_decode(outputs, skip_special_tokens=True))

# or using the HF pipeline
pipe = pipeline('text-generation', model=model, tokenizer=tokenizer, device='cuda:0')
with torch.autocast('cuda', dtype=torch.bfloat16):
    print(
        pipe('Here is a recipe for vegan banana bread:\n',
            max_new_tokens=100,
            do_sample=True,
            use_cache=True))

Model Description

The architecture is a modification of a standard decoder-only transformer.

The model has been modified from a standard transformer in the following ways:

Hyperparameter Value
n_parameters 6.7B
n_layers 32
n_heads 32
d_model 4096
vocab size 50432
sequence length 2048

Training Data

Streaming Datasets

Data was formatted using the MosaicML StreamingDataset library to host our data in object storage and efficiently stream it to our compute cluster during training. StreamingDataset obviates the need to download the whole dataset before starting training, and allows instant resumption of training from any point in the dataset.

Data Mix

The model was trained for ___T tokens. First it was trained for 1T tokens (with batch size 1760 and sequence length 2048) on the following data mix:

Data Mix for Original 1T Tokens Used to Train MPT-7B

Data Source Number of Tokens in Source Proportion Effective Number of Tokens Epochs
mC4 3.1.0 - English 417.99 B 0.33 330 B 0.14
C4 - English - SemDedup 80% 100.42 B 0.299 299 B 2.98
RedPajama - CommonCrawl 878.45 B 0.1 100 B 0.11
The Stack - Selected Languages 463.78 B 0.1 100 B 0.22
RedPajama - Wikipedia - En 4.87 B 0.04 40 B 8.21
The Stack - Markdown 107.07 B 0.035 35 B 0.33
S2ORC 48.85 B 0.033 33 B 0.68
RedPajama - Books 26.02 B 0.03 30B 1.15
RedPajama - arXiv 28.10 B 0.019 19 B 0.68
RedPajama - StackExchange 20.54 B 0.014 14 B 0.68

Data Mix for Additional 500B Tokens Used to Further Train MPT-7B-8k

We took 80B tokens from document samples that were longer than 4096 tokens, and 120B tokens with varying document sample lengths that matched the "baseline" length distribution for a total of 200B tokens in a single dataset. We then trained MPT-7B for 500B tokens with a maximum sequence length of 8192, resulting in MPT-7B-8k. Since we trained for 500B tokens using 200B tokens, nearly every subset was trained on for exactly 2.5 epochs.

Sequence Length Distribution Number of Tokens in Source (Billion) Proportion Effective Number of Tokens (Billion) Epochs
mC4 3.1.0 - English (200+ words) - Baseline 33.60 16.80% 84.00 2.50
mC4 3.1.0 - English (200+ words) - ≥4096 tokens 23.04 11.52% 57.60 2.50
c4 - English - SemDedup 80% - Baseline 30.12 15.06% 75.30 2.50
c4 - English - SemDedup 80% - ≥4096 tokens 0.92 0.46% 2.30 2.50
RedPajama - CommonCrawl - Baseline 8.52 4.26% 21.30 2.50
RedPajama - CommonCrawl - ≥4096 tokens 12.80 6.40% 32.00 2.50
The Stack - Selected Languages - Baseline 30.00 15.00% 75.00 2.50
The Stack - Selected Languages - ≥4096 tokens 10.00 5.00% 25.00 2.50
RedPajama - Wikipedia - Baseline 3.60 1.80% 9.00 2.50
RedPajama - Wikipedia - ≥4096 tokens 1.04 0.52% 2.60 2.50
The Stack - Markdown - Baseline 4.50 2.25% 11.25 2.50
The Stack - Markdown - ≥4096 tokens 8.00 4.00% 20.00 2.50
Semantic Scholar ORC - Baseline 3.30 1.65% 8.25 2.50
Semantic Scholar ORC - ≥4096 tokens 8.00 4.00% 20.00 2.50
RedPajama - Books - Baseline 3.00 1.50% 7.50 2.50
RedPajama - Books - ≥4096 tokens 8.00 4.00% 20.00 2.50
RedPajama - arXiv - Baseline 1.92 0.96% 4.80 2.50
RedPajama - arXiv - ≥4096 tokens 5.40 2.70% 13.50 2.50
RedPajama - StackExchange - Baseline 1.44 0.72% 3.60 2.50
RedPajama - StackExchange - ≥4096 tokens 1.52 1.40% 7.00 4.60
N Training Tokens 200 100.00% 2.5 epochs * 200B = 500B tokens

Samples for each batch were selected from one of the datasets with the probability specified above. 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.

The data was tokenized using the EleutherAI/gpt-neox-20b tokenizer. This BPE tokenizer has a number of desirable characteristics, most of which are relevant for tokenizing code: (1) It was trained on a diverse mix of data that includes code (The Pile) (2) It applies consistent space delimitation, unlike the GPT2 tokenizer which tokenizes inconsistently depending on the presence of prefix spaces (3) It contains tokens for repeated space characters, which allows superior compression of text with large amounts of repeated space characters.

The model vocabulary size of 50432 was set to be a multiple of 128 (as in MEGATRON-LM), model flop utilization (MFU) increased by up to four percentage points.

Training Configuration

This model was trained on 440 A100-40GBs for about 9.5 days using the MosaicML Platform. The model was trained with sharded data parallelism using FSDP and used the LION optimizer.

Limitations and Biases

The following language is modified from EleutherAI's GPT-NeoX-20B

MPT-7B-8k is not intended for deployment without finetuning. It should not be used for human-facing interactions without further guardrails and user consent.

MPT-7B-8k can produce factually incorrect output, and should not be relied on to produce factually accurate information. MPT-7B-8k was trained on various public datasets. 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.

MosaicML Platform

If you're interested in training and deploying your own MPT or LLMs on the MosaicML Platform, sign up here.

Disclaimer

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.

Citation

Please cite this model using the following format:

@online{MosaicML2023Introducing,
    author    = {MosaicML NLP Team},
    title     = {Introducing MPT-7B: A New Standard for Open-Source,
    ly Usable LLMs},
    year      = {2023},
    url       = {www.mosaicml.com/blog/mpt-7b},
    note      = {Accessed: 2023-03-28}, % change this date
    urldate   = {2023-03-28} % change this date
}