Synthyra
/

FastESMFold

@@ -1,199 +1,144 @@
 ---
 library_name: transformers
-tags: []
 ---
-# Model Card for Model ID
-<!-- Provide a quick summary of what the model is/does. -->
-## Model Details
-### Model Description
-<!-- Provide a longer summary of what this model is. -->
-This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
-- **Developed by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Model type:** [More Information Needed]
-- **Language(s) (NLP):** [More Information Needed]
-- **License:** [More Information Needed]
-- **Finetuned from model [optional]:** [More Information Needed]
-### Model Sources [optional]
-<!-- Provide the basic links for the model. -->
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
-## Uses
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
-### Direct Use
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
-[More Information Needed]
-### Downstream Use [optional]
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
-[More Information Needed]
-### Out-of-Scope Use
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
-[More Information Needed]
-## Bias, Risks, and Limitations
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
-[More Information Needed]
-### Recommendations
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
-## How to Get Started with the Model
-Use the code below to get started with the model.
-[More Information Needed]
-## Training Details
-### Training Data
-<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
-[More Information Needed]
-### Training Procedure
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
-#### Preprocessing [optional]
-[More Information Needed]
-#### Training Hyperparameters
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
-#### Speeds, Sizes, Times [optional]
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
-[More Information Needed]
-## Evaluation
-<!-- This section describes the evaluation protocols and provides the results. -->
-### Testing Data, Factors & Metrics
-#### Testing Data
-<!-- This should link to a Dataset Card if possible. -->
-[More Information Needed]
-#### Factors
-<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
-[More Information Needed]
-#### Metrics
-<!-- These are the evaluation metrics being used, ideally with a description of why. -->
-[More Information Needed]
-### Results
-[More Information Needed]
-#### Summary
-## Model Examination [optional]
-<!-- Relevant interpretability work for the model goes here -->
-[More Information Needed]
-## Environmental Impact
-<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
-Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
-- **Hardware Type:** [More Information Needed]
-- **Hours used:** [More Information Needed]
-- **Cloud Provider:** [More Information Needed]
-- **Compute Region:** [More Information Needed]
-- **Carbon Emitted:** [More Information Needed]
-## Technical Specifications [optional]
-### Model Architecture and Objective
-[More Information Needed]
-### Compute Infrastructure
-[More Information Needed]
-#### Hardware
-[More Information Needed]
-#### Software
-[More Information Needed]
-## Citation [optional]
-<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
-**BibTeX:**
-[More Information Needed]
-**APA:**
-[More Information Needed]
-## Glossary [optional]
-<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
-[More Information Needed]
-## More Information [optional]
-[More Information Needed]
-## Model Card Authors [optional]
-[More Information Needed]
-## Model Card Contact
-[More Information Needed]

 ---
 library_name: transformers
+tags:
+  - protein
+  - structure-prediction
+  - esmfold
+  - test-time-training
 ---
+# NOTE
+The GitHub with the implementation and requirements.txt can be found [here](https://github.com/Synthyra/FastPLMs.git)
+# FastESMFold
+FastESMFold is a self-contained, HuggingFace-compatible reimplementation of ESMFold with **built-in Test-Time Training (TTT)** and multi-backend attention (SDPA, Flash, Flex).
+No dependency on `fair-esm`, `proteinttt`, or `openfold`. Just `transformers`, `torch`, and `einops`.
+## Key Features
+- **Always-on TTT**: Runs 10 steps of masked language model adaptation via LoRA before folding. Returns the structure with the highest pLDDT across all steps.
+- **Best structure selection**: Folds after each TTT step, tracks per-step pLDDT, returns the best.
+- **FastESM2 attention**: SDPA/Flash/Flex backends for the 3B ESM2 backbone.
+- **Self-contained LoRA**: lora_diffusion-compatible implementation (no peft dependency). `Normal(0, 1/r)` initialization, `scale=alpha`.
+- **3.5B parameters**: Full ESMFold v1 architecture (ESM2-3B backbone + folding trunk).
+## Benchmark
+Tested on 10 difficult sequences on A10G GPU:
+| Metric | Value |
+|--------|-------|
+| Mean baseline pLDDT | 0.549 |
+| Mean best TTT pLDDT | 0.637 |
+| Mean improvement | +0.088 |
+| Sequences improved >5pt | 5/10 |
+| Time per sequence | ~20-45s |
+| GPU memory peak | 18.3 GB |
+On the hardest sequence (baseline pLDDT 0.38), TTT achieves 0.72 (+34 points).
+## Use with transformers
+### Basic usage
+```python
+import torch
+from transformers import AutoModel
+model = AutoModel.from_pretrained(
+    "Synthyra/FastESMFold",
+    trust_remote_code=True,
+    torch_dtype=torch.float32,
+).cuda().eval()
+result = model.fold_protein("MKTLLILAVVAAALA...")
+print(f"pLDDT: {result['plddt']:.3f}")
+print(f"Best step: {result['best_step']}")
+print(f"Step pLDDTs: {result['step_plddts']}")
+print(f"PDB length: {len(result['pdb_string'])} chars")
+```
+### Return values
+`fold_protein(sequence)` returns a dict with:
+| Key | Type | Description |
+|-----|------|-------------|
+| `plddt` | float | Best mean pLDDT across all TTT steps |
+| `ptm` | float | Predicted TM-score from best step |
+| `pdb_string` | str | PDB format structure from best step |
+| `step_plddts` | list[float] | pLDDT at each step [baseline, s1, ..., s10] |
+| `best_step` | int | Which step produced the best structure (0=baseline) |
+### Loading from Synthyra/ESMFold-v1 with custom config
+```python
+from esmfold.modeling_fast_esmfold import FastEsmFoldConfig, FastEsmForProteinFolding
+config = FastEsmFoldConfig.from_pretrained("Synthyra/ESMFold-v1")
+config.attn_backend = "sdpa"
+config.ttt_config = {
+    "lr": 4e-4,
+    "steps": 10,
+    "ags": 4,
+    "batch_size": 4,
+    "lora_rank": 8,
+    "lora_alpha": 32.0,
+    "seed": 0,
+}
+model = FastEsmForProteinFolding.from_pretrained(
+    "Synthyra/ESMFold-v1",
+    config=config,
+    torch_dtype=torch.float32,
+).cuda().eval()
+```
+## Attention backends
+The ESM2 backbone supports multiple attention backends via `config.attn_backend`:
+| Backend | Key | Notes |
+| :--- | :--- | :--- |
+| PyTorch SDPA | `"sdpa"` | Default. Exact numerics, stable on all hardware. |
+| Flash Attention | `"kernels_flash"` | Fastest. Requires `pip install kernels`. |
+| Flex Attention | `"flex"` | Skips padding tokens via block mask. First use compiles a Triton kernel. |
+| Auto | `"auto"` | Picks best available: `kernels_flash` > `flex` > `sdpa`. |
+## TTT Configuration
+TTT parameters can be customized via `config.ttt_config`:
+| Parameter | Default | Description |
+|-----------|---------|-------------|
+| `lr` | 4e-4 | Learning rate for SGD optimizer |
+| `steps` | 10 | Number of optimizer steps |
+| `ags` | 4 | Gradient accumulation steps per optimizer step |
+| `batch_size` | 4 | Batch size for masked language model training |
+| `mask_ratio` | 0.15 | Fraction of tokens to mask |
+| `lora_rank` | 8 | LoRA rank (0 for full fine-tuning) |
+| `lora_alpha` | 32.0 | LoRA scaling factor |
+| `seed` | 0 | Random seed for reproducibility |
+## How TTT Works
+1. **Baseline fold** (step 0): Standard ESMFold prediction
+2. **LoRA injection**: Rank-8 LoRA adapters on ESM2 attention Q/K/V projections
+3. **Masked LM training**: 10 optimizer steps of BERT-style masked language modeling on the input sequence
+4. **Per-step folding**: After each optimizer step, fold the sequence and record pLDDT
+5. **Best selection**: Return the structure with the highest pLDDT
+6. **Reset**: Restore LoRA weights to initial state for the next sequence
+This is based on the ProteinTTT paper (test-time compute for protein structure prediction).
+### Citation
+If you use this implementation please cite it:
+```
+@misc {FastPLMs,
+    author       = { Hallee, Logan and Bichara, David and Gleghorn, Jason P.},
+    title        = { FastPLMs: Fast, efficient, protein language model inference from Huggingface AutoModel.},
+    year         = {2024},
+    url          = { https://huggingface.co/Synthyra/ESMplusplus_small },
+    DOI          = { 10.57967/hf/3726 },
+    publisher    = { Hugging Face }
+}
+```