Skip to content

Latest commit

 

History

History
255 lines (183 loc) · 10.3 KB

README.md

File metadata and controls

255 lines (183 loc) · 10.3 KB

ESM-2 Optimized with NVIDIA TransformerEngine

This folder contains source code and tests for an ESM-2 model that inherits from the transformers PreTrainedModel class and uses TransformerEngine layers. Users don't need to install this package directly, but can load the model directly from HuggingFace Hub using the standard transformers API. For more information, refer to Inference Examples.

Feature support

The ESM-2 implementation natively supports the following TransformerEngine-provided optimizations:

Feature Support
FP8 ✅ Supported on compute capacity 9.0 and above (Hopper+)
MXFP8 ✅ Supported on compute capacity 10.0 and 10.3 (Blackwell), 12.0 support pending
Sequence Packing / THD input format ✅ Supported
FP8 with THD input format ✅ Supported where FP8 is supported
Import from HuggingFace checkpoints ✅ Supported
Export to HuggingFace checkpoints ✅ Supported

Refer to BioNemo Recipes for more details on how to use these features to accelerate model training and inference.

Links to HF checkpoints

Pre-trained ESM-2 models converted from the original Facebook weights are available on HuggingFace as part of the NVIDIA BioNeMo collection on the HuggingFace Hub:

Available Models:

Runtime Requirements

We recommend using the latest NVIDIA PyTorch container for optimal performance and compatibility. Refer to the provided Dockerfile for details.

Inference Examples

Quick start example using HuggingFace transformers:

from transformers import AutoModel, AutoTokenizer

model = AutoModel.from_pretrained("nvidia/esm2_t6_8M_UR50D")
tokenizer = AutoTokenizer.from_pretrained("nvidia/esm2_t6_8M_UR50D")

gfp_P42212 = (
    "MSKGEELFTGVVPILVELDGDVNGHKFSVSGEGEGDATYGKLTLKFICTTGKLPVPWPTL"
    "VTTFSYGVQCFSRYPDHMKQHDFFKSAMPEGYVQERTIFFKDDGNYKTRAEVKFEGDTLV"
    "NRIELKGIDFKEDGNILGHKLEYNYNSHNVYIMADKQKNGIKVNFKIRHNIEDGSVQLAD"
    "HYQQNTPIGDGPVLLPDNHYLSTQSALSKDPNEKRDHMVLLEFVTAAGITHGMDELYK"
)

inputs = tokenizer(gfp_P42212, return_tensors="pt")
output = model(**inputs)

Recipe Links

Training recipes are available in the bionemo-recipes/recipes/ directory:

Running with Low Precision (FP8/FP4)

The TE-optimized ESM-2 model supports per-layer quantization via two mechanisms: a config-level layer_precision list that declares which layers use which precision, and constructor-level recipe objects (fp8_recipe, fp4_recipe) that control the quantization behaviour.

Configuration: layer_precision

NVEsmConfig.layer_precision is a list of length num_hidden_layers where each element is "fp8", "fp4", or None (BF16 fallback). When set, it controls the te.autocast context used for each transformer layer during both initialization and forward pass.

from modeling_esm_te import NVEsmConfig, NVEsmForMaskedLM

# All layers in FP8
config = NVEsmConfig.from_pretrained(
    "nvidia/esm2_t6_8M_UR50D",
    layer_precision=["fp8"] * 6,
)

If you pass an fp8_recipe to the model constructor without setting layer_precision, it defaults to ["fp8"] * num_hidden_layers (all layers FP8). You can also mix precisions, for example running most layers in FP8 but keeping the first and last layers in BF16:

layer_precision = [None] + ["fp8"] * 4 + [None]
config = NVEsmConfig.from_pretrained(
    "nvidia/esm2_t6_8M_UR50D",
    layer_precision=layer_precision,
)

Constructor arguments: fp8_recipe and fp4_recipe

The model classes (NVEsmModel, NVEsmForMaskedLM, NVEsmForTokenClassification) accept fp8_recipe and fp4_recipe keyword arguments. These are transformer_engine.common.recipe.Recipe objects that configure the quantization algorithm (e.g., delayed scaling, block scaling, MXFP8).

import transformer_engine.common.recipe as te_recipe

from modeling_esm_te import NVEsmConfig, NVEsmForMaskedLM

fp8_recipe = te_recipe.DelayedScaling()

config = NVEsmConfig.from_pretrained(
    "nvidia/esm2_t6_8M_UR50D",
    layer_precision=["fp8"] * 6,
)
model = NVEsmForMaskedLM(config, fp8_recipe=fp8_recipe)

For FP4 (NVFP4) quantization, pass an fp4_recipe instead and set the corresponding layers to "fp4" in layer_precision:

fp4_recipe = te_recipe.NVFP4BlockScaling()

config = NVEsmConfig.from_pretrained(
    "nvidia/esm2_t6_8M_UR50D",
    layer_precision=["fp4"] * 6,
)
model = NVEsmForMaskedLM(config, fp4_recipe=fp4_recipe)

You can also mix FP8 and FP4 layers by providing both recipes and a mixed layer_precision list.

Quantized model initialization: use_quantized_model_init

When use_quantized_model_init=True is set in the config, layers are created inside a te.quantized_model_init context. This tells TransformerEngine to initialize weights directly in the target quantized format, avoiding a separate quantization step after initialization.

config = NVEsmConfig.from_pretrained(
    "nvidia/esm2_t6_8M_UR50D",
    layer_precision=["fp4"] * 6,
    use_quantized_model_init=True,
)
model = NVEsmForMaskedLM(config, fp4_recipe=te_recipe.NVFP4BlockScaling())

Notes

  • The lm_head (and dense projection in NVEsmLMHead) always runs in higher precision (te.autocast(enabled=False)) regardless of layer_precision, to avoid numerical instability in the output logits.
  • FP8 requires compute capability 9.0+ (Hopper). MXFP8 requires compute capability 10.0+ (Blackwell).
  • If an fp8_recipe is provided without layer_precision, all layers default to FP8. Providing both fp8_recipe and fp4_recipe without layer_precision raises a RuntimeError.
  • An FP4 layer requires an fp4_recipe; omitting it raises a RuntimeError.

Converting Between Model Formats

This section explains how to convert between Hugging Face Transformers and Transformer Engine (TE) ESM2 model formats. The process demonstrates bidirectional conversion: from Transformers to TE format for optimized inference, and back to Hugging Face Transformers format for sharing and deployment. The workflow involves several key steps:

Converting from HF Transformers to TE

from transformers import AutoModelForMaskedLM

from convert import convert_esm_hf_to_te

hf_model = AutoModelForMaskedLM.from_pretrained("facebook/esm2_t6_8M_UR50D")
te_model = convert_esm_hf_to_te(hf_model)
te_model.save_pretrained("/path/to/te_checkpoint")

This loads the pre-trained ESM2 model that will serve as our reference for comparison.

Converting from TE back to HF Transformers

from convert import convert_esm_te_to_hf
from modeling_esm_te import NVEsmForMaskedLM

te_model = NVEsmForMaskedLM.from_pretrained("/path/to/te_checkpoint")
hf_model = convert_esm_te_to_hf(te_model)
hf_model.save_pretrained("/path/to/hf_checkpoint")

Loading and Testing the Exported Model

Load the exported model and perform validation:

from transformers import AutoTokenizer

model_hf_exported = AutoModelForMaskedLM.from_pretrained("/path/to/hf_checkpoint")
tokenizer = AutoTokenizer.from_pretrained("facebook/esm2_t6_8M_UR50D")

Validating Converted Models

To validate the converted models, refer to the commands in Inference Examples above to load and test both the original and converted models to ensure loss and logit values are similar. Additionally, refer to the golden value tests in test_modeling_esm_te.py and test_export.py.

Developer Guide

Running tests

To run tests locally, run recipes_local_test.py from the repository root with the model directory as an argument.

./ci/scripts/recipes_local_test.py bionemo-recipes/models/esm2/

Development container

To use the provided devcontainer, use "Dev Containers: Reopen in Container" from the VSCode menu, and choose the "BioNeMo Recipes Dev Container" option. To run the tests inside the container, first install the dependencies with pip install -r requirements.txt, then run pytest -v . in the model directory.

Deploying converted checkpoints to HuggingFace Hub

First, generate converted ESM-2 checkpoints from existing HuggingFace transformers checkpoints:

mkdir -p checkpoint_export
docker build -t esm2 .
docker run --rm -it --gpus all \
  -v $PWD/checkpoint_export/:/workspace/bionemo/checkpoint_export \
  -v $HOME/.cache/huggingface/:/root/.cache/huggingface \
  esm2 python export.py

Now deploy the converted checkpoints to the HuggingFace Hub by running the following command for each model:

huggingface-cli upload nvidia/${MODEL_NAME} $PWD/checkpoint_export/${MODEL_NAME}

You can also upload all models at once with:

cd checkpoint_export
for dir in */; do hf upload --repo-type model nvidia/$(basename "$dir") "$dir/"; done