README.md

---
library_name: transformers
license: "mit"
tags:
  - protein-language-model
  - fastplms
---

<!-- Generated from src/fastplms/models.toml. Do not edit. -->

# ESM++ Large

## Model overview

`Synthyra/ESMplusplus_large` packages the `biohub/ESMC-600M` checkpoint with
the FastPLMs runtime for Hugging Face Transformers. It accepts amino-acid
sequences tokenized to residue IDs.

The repository uses the standard Transformers loading interface with
`trust_remote_code=True`. See Technical details for each registered class and
whether its weights come from the checkpoint.

The sequence- and token-classification classes reuse the pretrained backbone,
but their task heads are newly initialized. Fine-tune those heads before
interpreting their logits as predictions.

## Install and platform requirements

Install the direct dependencies published with this model:

```bash
python -m pip install -r \
  "https://huggingface.co/Synthyra/ESMplusplus_large/resolve/main/requirements.txt"
```

The FastPLMs implementation itself is embedded in the model repository.
Transformers loads it through `trust_remote_code=True`.

This model requires Python 3.11-3.14, PyTorch 2.13, and Transformers 5.13.

The artifact requirements include the FlashAttention loader dependency.
FlashAttention also requires compatible CUDA hardware and BF16 execution.

The Hub quick start needs network access for the first download. For an
air-gapped run, build the manifest-pinned local artifact first and use the
offline example.

## Quick start

```python
from transformers import AutoModel

model_id = "Synthyra/ESMplusplus_large"
model = AutoModel.from_pretrained(
    model_id,
    trust_remote_code=True,
    attn_implementation="sdpa",
).eval()
```

For offline validation, replace `model_id` with the manifest-built
`dist/hub/ESMplusplus_large` path. Pass `local_files_only=True`.

## Attention backends

The quick start uses `sdpa`.

Available backends are `eager`, `sdpa`, `flex_attention`, `flash_attention_2`,
`flash_attention_3`. Requesting an unavailable backend raises instead of
silently changing implementation.

`output_attentions=True` can use the documented one-call eager fallback to
materialize attention tensors. The configured backend does not change.

## Tokenization and forward inference

Load the tokenizer from the same artifact as the model. The attention mask
shows padding explicitly:

```python
import torch
from transformers import AutoTokenizer

model_id = "Synthyra/ESMplusplus_large"
tokenizer = AutoTokenizer.from_pretrained(
    model_id,
    trust_remote_code=True,
)
batch = tokenizer(
    ["MSTNPKPQRKTKRNT", "MKTIIALSYIFCLVFA"],
    padding=True,
    return_tensors="pt",
)

with torch.inference_mode():
    output = model(**batch)

print(output.last_hidden_state.shape)
```

## Dataset embeddings

The shared embedding mixin keeps input order and biological-position masking.
It accepts sequences, identified records, mappings, or a FASTA path:

```python
pooled = model.embed_dataset(
    ["MSTNPKPQRKTKRNT", "MKTIIALSYIFCLVFA"],
    batch_size=2,
    pooling=("mean", "std"),
)
residues = model.embed_dataset(
    ["MSTNPKPQRKTKRNT"],
    full_embeddings=True,
)
print(pooled[0].tensor.shape)   # (2 * d,)
print(residues[0].tensor.shape) # (l, d)
```

Set `output` and `format="safetensors"` or `"sqlite"` for transactional,
bounded-memory storage. Resume checks input order, model state, tokenizer
policy, backend, dtype, and pooling configuration before it appends data.

## Downstream prediction

The sequence and token prediction AutoClasses use the checkpoint backbone and
create a new, untrained `classifier`. Sequence labels have shape `(b,)`.
Residue labels have shape `(b, l)` and use `-100` outside biological positions.

```python
import torch
from transformers import AutoTokenizer
from transformers import (
    AutoModelForSequenceClassification,
    AutoModelForTokenClassification,
)

model_id = "Synthyra/ESMplusplus_large"
sequence_model = AutoModelForSequenceClassification.from_pretrained(
    model_id, num_labels=2, trust_remote_code=True
).eval()
token_model = AutoModelForTokenClassification.from_pretrained(
    model_id, num_labels=3, trust_remote_code=True
).eval()
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
sequences = ["MSTNPKPQRKTKRNT", "MKTIIALSYIFCLVFA"]
batch = tokenizer(sequences, padding=True, return_tensors="pt")
biological = batch["attention_mask"].bool()
for special_id in tokenizer.all_special_ids:
    biological &= batch["input_ids"].ne(special_id)

sequence_labels = torch.zeros(len(sequences), dtype=torch.long)
token_labels = torch.full_like(batch["input_ids"], -100)
token_labels[biological] = 0

with torch.inference_mode():
    sequence_output = sequence_model(**batch, labels=sequence_labels)
    token_output = token_model(**batch, labels=token_labels)
print(sequence_output.logits.shape)  # (b, 2)
print(token_output.logits.shape)     # (b, l, 3)
```

## PEFT fine-tuning

Install the training dependencies. Then attach LoRA to the loaded checkpoint:

```bash
python -m pip install "datasets>=4.8,<5" "peft>=0.19,<0.20"
```

```python
from peft import LoraConfig, TaskType, get_peft_model

peft_model = get_peft_model(
    sequence_model,
    LoraConfig(
        task_type=TaskType.SEQ_CLS,
        r=8,
        lora_alpha=16,
        target_modules="all-linear",
        modules_to_save=["classifier"],
    ),
)
```

This checkpoint advertises a classification head. Save the separately trained
`classifier` with the adapter.
All FastPLMs checkpoints follow the Transformers `PreTrainedModel` contract and
can use PEFT. The ESM2-specific shipped CLI is an example, not a
support boundary. Record the target modules, base revision, data identity, and
trainable parameter scope.

## Test-time training

TTT samples masked views of one protein and updates only injected low-rank
adapters. Base checkpoint weights stay frozen:

```python
from transformers import AutoModelForMaskedLM

ttt_model = AutoModelForMaskedLM.from_pretrained(
    "Synthyra/ESMplusplus_large",
    trust_remote_code=True,
)
metrics = ttt_model.ttt(
    seq="MSTNPKPQRKTKRNT",
    ttt_config={"steps": 3, "batch_size": 1, "seed": 7},
)
ttt_model.save_pretrained("adapted", safe_serialization=True)
ttt_model.ttt_reset()
print(metrics)
```

Saved adapters retain their deterministic reset state. TTT adds latency and
memory, can worsen an output, and does not show biological function.

## ESMC behavior

This artifact provides the Biohub ESMC sequence encoder and masked-language-
model head through Transformers. ESMFold2 also uses this language-model family.
SDPA is the default and gives the highest numerical fidelity. Flex Attention and
FlashAttention 3 are supported non-experimental backends. Their BF16 arithmetic
can differ numerically from SDPA. These differences give diagnostic warnings,
not strict-parity failures. Dispatch, masks, finite outputs, shapes, and large
biological disagreements remain hard gates.

The current GH200/aarch64 release environment validates eager, SDPA, and Flex.
Flash requests raise because compatible locked kernels are unavailable on this
platform.

When `sequence_id` is supplied, it controls ESMC attention groups and padding.
`attention_mask` is ignored. Values greater than or equal to zero are valid
sequence-group IDs. `-1` marks padding. Omit `sequence_id` to use
`attention_mask` for padding.

### Hidden-state sparse autoencoders

ESM++ supports hidden-state SAEs from the official
[Biohub ESMC SAE collection](https://huggingface.co/collections/biohub/esmc-saes-for-hidden-states-all-layers).
Select an SAE for this ESMC scale. Load only required layers. Then attach them
to the model:

```python
import torch
from transformers import AutoModel

sae = AutoModel.from_pretrained("biohub/ESMC-600M-sae-layer27-k64-codebook65536", device=model.device)
sae.initialize_layers([27])
model.add_sae_models([sae.layers["27"]])

with torch.inference_mode():
    output = model(**batch, normalize_sae=True)

features = output.sae_outputs["layer27"]
print(features.shape, features.layout)  # (valid_token_count, codebook_dim), sparse COO
```

SAEs run after you attach them. Use `compute_sae=False` to skip SAE work.
Outputs are detached sparse tensors with keys such as `layer{N}`. They omit
padding. The model uses `sequence_id`, then `attention_mask`, for padding.
`normalize_sae=True` uses Biohub `(features / max) * idf` normalization. SAE
computation requires `input_ids`. It rejects mask tokens because Biohub trained
the SAEs with unmasked sequences. This interface supports hidden-state SAEs
only, not MLP-output SAEs. FastPLMs does not copy SAE weights or add SAE
checkpoints to its model manifest.

### Experimental FP8 inference

The default uses checkpoint BF16 behavior. FP8 is an explicit experimental
inference option for every ESM++ scale:

```python
import torch
from transformers import AutoModel

fp8_model = AutoModel.from_pretrained(
    "Synthyra/ESMplusplus_large",
    trust_remote_code=True,
    dtype=torch.bfloat16,
).cuda().eval()
fp8_model.enable_fp8()
print(fp8_model.esmc_precision_status)

with torch.inference_mode():
    fp8_output = fp8_model(**{name: value.cuda() for name, value in batch.items()})
```

FP8 forward calls require `torch.inference_mode()`. The model pads the sequence
dimension to a multiple of 16. Transformer Engine converts supported linear
layers. The call fails if the dependency, compatible CUDA hardware, or complete
conversion set is unavailable. It does not silently use BF16. FP8 does not
claim numerical parity.

| Backend | Support | Measurement status |
| --- | --- | --- |
| `sdpa` | Recommended fidelity path | Pending release measurement |
| `eager` | Supported | Pending release measurement |
| `flash_attention_2` | Supported | Unavailable on current GH200/aarch64 lock |
| `flex_attention` | Supported, numerically divergent | Pending release measurement |
| `flash_attention_3` | Supported, numerically divergent | Unavailable on current GH200/aarch64 lock |

Detailed backend measurements, release guardrails, and the GH200 package
compatibility exception are maintained in the
[attention backend guide](https://github.com/Synthyra/FastPLMs/blob/main/docs/attention_backends.md)
and
[release evidence manifest](https://github.com/Synthyra/FastPLMs/blob/main/docs/generated/capability_evidence.md).


## Technical details

- Inputs: Amino-acid sequences tokenized to residue IDs
- Transformers classes: `AutoConfig`, `AutoModel`, `AutoModelForMaskedLM`, `AutoModelForSequenceClassification`, `AutoModelForTokenClassification`
- Checkpoint weights: `AutoConfig` = `FastPLMs extension`, `AutoModel` = `pretrained`, `AutoModelForMaskedLM` = `pretrained`, `AutoModelForSequenceClassification` = `base weights + untrained task head`, `AutoModelForTokenClassification` = `base weights + untrained task head`
- Attention backends: `eager`, `sdpa`, `flex_attention`, `flash_attention_2`, `flash_attention_3`
- Precision: `default`, `fp8` (experimental)
- BF16 execution: `static_parameters`
- Generation contract: `not_applicable`
- Dependencies: `core`
- Weight publication allowed: `true`
- Weight license status: `resolved`
- Redistributable: `true`
- Complete weight publication required: `false`

## Validation and provenance

FastPLMs pins the checkpoint, upstream source revisions, state transformation,
and required files in `models.toml`. Built artifacts record exact source
identities and conversion details in `source-record.json`.

- FastPLMs checkpoint: `Synthyra/ESMplusplus_large`
- Runtime revision: recorded separately in the built artifact and published commit
- Runtime source identities: recorded in `source-record.json`
- Official checkpoint: `biohub/ESMC-600M`
- Artifact source: `fast`
- State transform: `esmc_to_fastplms_v1`
- Pinned upstreams: `biohub-esm`, `biohub-transformers`
- Release tiers: `check`, `compliance`, `feature`, `artifact`, `benchmark`
- Unresolved required file identities: `0`

Release validation includes the `compliance` tier. Its evidence identifies the
checkpoint, backend, dtype, hardware, inputs, and reference revision.

Declared tiers compare configuration, tokenizer behavior, state, and
representative inference with the pinned reference. A nonzero unresolved count
blocks release. Metadata alone does not show that a build passed, that a backend
is faster, or that an output is biologically valid.

## License

Checkpoint terms: MIT. The Hub model-card identifier is
`mit`. The local artifact contains applicable source
licenses, notices, attribution, and conversion records. Review them before use.