refactor: remove unused safetensors files and add new configurations for EMG models

DB5/DB5_finetune_5sec.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3f155360b53600d266b12f5aff473094971a551c3f1e828e88dde83a0cd076a
+size 14927820

DB5/config.json

@@ -0,0 +1,16 @@
+{
+  "_target_": "models.TinyMyo.TinyMyo",
+  "img_size": 1000,
+  "patch_size": 20,
+  "in_chans": 16,
+  "embed_dim": 192,
+  "n_layer": 8,
+  "n_head": 3,
+  "mlp_ratio": 4,
+  "qkv_bias": true,
+  "attn_drop": 0.1,
+  "proj_drop": 0.1,
+  "drop_path": 0.1,
+  "num_classes": 53,
+  "task": "classification"
+}

DB8/DB8_finetune_500ms.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b4ee6743c584e3e1d0791be95ab9beb5b3937d20484551c22fc249b51f7dcc4b
+size 17432748

DB8/config.json

@@ -0,0 +1,16 @@
+{
+  "_target_": "models.TinyMyo.TinyMyo",
+  "img_size": 1000,
+  "patch_size": 20,
+  "in_chans": 16,
+  "embed_dim": 192,
+  "n_layer": 8,
+  "n_head": 3,
+  "mlp_ratio": 4,
+  "qkv_bias": true,
+  "attn_drop": 0.1,
+  "proj_drop": 0.1,
+  "drop_path": 0.1,
+  "num_classes": 5,
+  "task": "regression"
+}

EPN612/EPN_finetune_5sec.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:07593aa6579a99e3c9035db75fe697f7ce133bc9c3aa1ba64935a646c7b6a349
+size 14313224

EPN612/config.json

@@ -0,0 +1,16 @@
+{
+  "_target_": "models.TinyMyo.TinyMyo",
+  "img_size": 1000,
+  "patch_size": 20,
+  "in_chans": 8,
+  "embed_dim": 192,
+  "n_layer": 8,
+  "n_head": 3,
+  "mlp_ratio": 4,
+  "qkv_bias": true,
+  "attn_drop": 0.1,
+  "proj_drop": 0.1,
+  "drop_path": 0.1,
+  "num_classes": 6,
+  "task": "classification"
+}

README.md

@@ -1,170 +1,62 @@
----
-license: cc-by-nd-4.0
-language:
-- en
-model-index:
-- name: TinyMyo
-  results:
-  - task:
-      type: gesture-classification
-    dataset:
-      type: ninapro_db5
-      name: Ninapro DB5
-    metrics:
-    - name: acc@1
-      type: acc@1
-      value: 0.8941
-      verified: false
-    - name: f1
-      type: f1
-      value: 0.7797
-      verified: false
-  - task:
-      type: gesture-classification
-    dataset:
-      type: epn612
-      name: EPN-612
-    metrics:
-    - name: acc@1
-      type: acc@1
-      value: 0.9674
-      verified: false
-    - name: f1
-      type: f1
-      value: 0.9674
-      verified: false
-  - task:
-      type: gesture-classification
-    dataset:
-      type: uci_emg
-      name: UCI-EMG
-    metrics:
-    - name: acc@1
-      type: acc@1
-      value: 0.9756
-      verified: false
-    - name: f1
-      type: f1
-      value: 0.9755
-      verified: false
-  - task:
-      type: gesture-classification
-    dataset:
-      type: gni_meta_rl
-      name: Generic Neuromotor Interface (Discrete Gesture)
-    metrics:
-    - name: CLER
-      type: classification-error-rate
-      value: 0.153
-      verified: false
-  - task:
-      type: kinematic-regression
-    dataset:
-      type: ninapro_db8
-      name: Ninapro DB8
-    metrics:
-    - name: MAE
-      type: mean-absolute-error
-      value: 8.77
-      verified: false
-    - name: RMSE
-      type: root-mean-square-error
-      value: 13.35
-      verified: false
-    - name: R2
-      type: r2
-      value: 0.62
-      verified: false
-  - task:
-      type: speech-synthesis
-    dataset:
-      type: gaddy_silent_speech
-      name: Gaddy Silent Speech (MFCC to Audio)
-    metrics:
-    - name: WER
-      type: word-error-rate
-      value: 0.3354
-      verified: false
-  - task:
-      type: speech-recognition
-    dataset:
-      type: gaddy_silent_speech
-      name: Gaddy Silent Speech (EMG to Text)
-    metrics:
-    - name: WER
-      type: word-error-rate
-      value: 0.3395
-      verified: false
-tags:
-- emg
-- bio-signals
-- foundation-model
----
+# TinyMyo: Tiny Foundation Model for EMG Signal Processing
 
 <div align="center">
   <img src="https://raw.githubusercontent.com/MatteoFasulo/BioFoundation/refs/heads/TinyMyo/docs/model/logo/TinyMyo_logo.png" alt="TinyMyo Logo" width="400" />
-  <h1>TinyMyo: a Tiny Foundation Model for Flexible EMG Signal Processing at the Edge</h1>
 </div>
+
 <p align="center">
   <a href="https://github.com/pulp-bio/BioFoundation"><img src ="https://img.shields.io/github/stars/pulp-bio/BioFoundation?color=ccf" alt="Github"></a>
   <a href="https://creativecommons.org/licenses/by-nd/4.0/"><img src="https://img.shields.io/badge/License-CC_BY--ND_4.0-lightgrey.svg" alt="License"></a>
   <a href="https://arxiv.org/abs/2512.15729"><img src="https://img.shields.io/badge/arXiv-2512.15729-b31b1b.svg" alt="Paper"></a>
 </p>
 
-**TinyMyo** is a **3.6M-parameter** Transformer foundation model for surface EMG (sEMG), optimized for ultra-low-power edge deployment (GAP9 MCU). It demonstrates state-of-the-art performance across gesture classification, kinematic regression, and speech synthesis.
-
----
+## 📖 Overview
+**TinyMyo** is a lightweight (3.6M parameters), Transformer-based foundation model designed specifically for surface electromyography (sEMG) signal processing. Unlike large-scale models, TinyMyo is purpose-built for **ultra-low-power edge deployment**, enabling real-time motor intent decoding, neuromuscular assessment, and human-machine interaction directly on microcontrollers like the GAP9.
 
-## 🚀 Quick Start
+## 🚀 Key Highlights
+*   **Generalist Foundation:** Pre-trained on a massive, heterogeneous corpus of >480 GB of EMG data (NinaPro DB6/7, EMG2Pose) using self-supervised masked reconstruction.
+*   **Edge-Ready:** The first EMG foundation model demonstrated on an ultra-low-power MCU (GAP9).
+*   **Highly Efficient:** Just 3.6M parameters, ensuring low latency and high energy efficiency (44.91 mJ per inference).
+*   **Versatile:** Achieves state-of-the-art (SoA) performance across hand gesture classification, kinematic regression, and speech processing.
 
-TinyMyo is built as a specialized model within the [BioFoundation](https://github.com/pulp-bio/BioFoundation) framework.
+## 🧠 Model Architecture
+*   **Core:** 8-layer bidirectional Transformer encoder.
+*   **Embeddings:** 192-dimensional latent space with 3 attention heads.
+*   **Tokenization:** Channel-independent patching (20 samples per patch) utilizing Rotary Position Embeddings (RoPE) to preserve temporal alignment across channels.
+*   **Deployment:** Optimized via multi-level tiling and INT8 quantization for execution on resource-constrained hardware.
 
-### 1. Requirements
-- **Preprocessing:** Dependencies for data scripts are in `scripts/requirements.txt`.
-- **BioFoundation:** Full framework requirements for training/inference are in the [GitHub repository](https://github.com/pulp-bio/BioFoundation/blob/main/requirements.txt).
+## 📊 Performance Benchmarks
 
-### 2. Preprocessing
-Process raw datasets into HDF5 format:
-```bash
-python scripts/db5.py --data_dir $DATA_PATH/raw/ --save_dir $DATA_PATH/h5/ --seq_len 200 --stride 50
-```
-*See [scripts/README.md](scripts/README.md) for all dataset commands.*
-
-### 3. Fine-tuning
-```bash
-python run_train.py +experiment=TinyMyo_finetune pretrained_safetensors_path=/path/to/base.safetensors
-```
-
----
-
-## 🧠 Architecture & Pretraining
-- **Core:** 8-layer Transformer encoder (192-dim embeddings, 3 heads).
-- **Tokenization:** Channel-independent patching (20 samples/patch) with RoPE.
-- **Data:** Pretrained on >480 GB of EMG (NinaPro DB6/7, EMG2Pose).
-- **Specs:** 3.6M parameters, 4.0 GFLOPs.
+| Task | Dataset | Metric | TinyMyo Result |
+| :--- | :--- | :--- | :--- |
+| **Gesture Classification** | NinaPro DB5 | Accuracy | **89.41%** |
+| **Gesture Classification** | EPN-612 | Accuracy | **96.74%** |
+| **Gesture Classification** | UCI EMG | Accuracy | **97.56%** |
+| **Kinematic Regression** | NinaPro DB8 | MAE | **8.77°** |
+| **Speech Synthesis** | Gaddy | WER | **33.54%** |
+| **Speech Recognition** | Gaddy | WER | **33.95%** |
 
-## 🎯 Benchmarks
+## ⚡ Deployment (GAP9 MCU)
+TinyMyo bridges the gap between high-performance deep learning and wearable constraints:
+*   **Inference Time:** 0.785 s
+*   **Energy Consumption:** 44.91 mJ
+*   **Power Envelope:** 57.18 mW
 
-| Task | Dataset | Metric | TinyMyo |
-| :--- | :--- | :--- | :--- |
-| **Gesture** | NinaPro DB5 | Accuracy | **89.41%** |
-| **Gesture** | EPN-612 | Accuracy | **96.74%** |
-| **Gesture** | UCI EMG | Accuracy | **97.56%** |
-| **Regression**| NinaPro DB8 | MAE | **8.77°** |
-| **Speech** | Gaddy (Speech Synthesis) | WER | **33.54%** |
-| **Speech** | Gaddy (Speech Recognition) | WER | **33.95%** |
+## 🛠️ Getting Started
+TinyMyo is part of the [BioFoundation](https://github.com/pulp-bio/BioFoundation) ecosystem.
 
----
+### Prerequisites
+Install the required dependencies from the [BioFoundation repository](https://github.com/pulp-bio/BioFoundation).
 
-## ⚡ Edge Performance (GAP9 MCU)
-- **Inference:** 0.785 s
-- **Energy:** 44.91 mJ
-- **Power:** 57.18 mW
+### Fine-tuning
+You can easily fine-tune the pre-trained weights for your specific task:
 
----
+```bash
+python run_train.py +experiment=TinyMyo_finetune pretrained_safetensors_path={*.safetensors}
+```
 
 ## 📜 License & Citation
-Weights are licensed under **CC BY-ND 4.0**. See [LICENSE](LICENSE) for details.
+This model is licensed under **CC BY-ND 4.0**. If you find TinyMyo useful in your research, please cite our paper:
 
 ```bibtex
 @misc{fasulo2026tinymyotinyfoundationmodel,
@@ -173,7 +65,6 @@ Weights are licensed under **CC BY-ND 4.0**. See [LICENSE](LICENSE) for details.
       year={2026},
       eprint={2512.15729},
       archivePrefix={arXiv},
-      primaryClass={eess.SP},
-      url={https://arxiv.org/abs/2512.15729},
+      primaryClass={eess.SP}
 }
 ```

UCI_EMG/UCI_finetune_5sec.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f4349d063bc4d8a04bcdb70b919c6dcf2b6c809075d9ca1d849adce6feb4ebe
+size 14313224

UCI_EMG/config.json

@@ -0,0 +1,16 @@
+{
+  "_target_": "models.TinyMyo.TinyMyo",
+  "img_size": 1000,
+  "patch_size": 20,
+  "in_chans": 8,
+  "embed_dim": 192,
+  "n_layer": 8,
+  "n_head": 3,
+  "mlp_ratio": 4,
+  "qkv_bias": true,
+  "attn_drop": 0.1,
+  "proj_drop": 0.1,
+  "drop_path": 0.1,
+  "num_classes": 6,
+  "task": "classification"
+}

ckpt_to_safetensor.py

@@ -0,0 +1,77 @@
+import argparse
+import json
+import os
+
+import torch
+from omegaconf import OmegaConf
+from safetensors.torch import load_file, save_file
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Convert a PyTorch Lightning checkpoint to a safetensors file."
+    )
+    parser.add_argument("ckpt_path", type=str, help="Path to .ckpt file.")
+    parser.add_argument(
+        "--exclude_keys", type=str, nargs="*", default=[], help="Keys to exclude."
+    )
+    parser.add_argument(
+        "--strip_prefix",
+        type=str,
+        default=None,
+        help="Prefix to remove from keys (e.g., 'model.').",
+    )
+    parser.add_argument(
+        "--verbose", action="store_true", help="Print keys being saved."
+    )
+
+    args = parser.parse_args()
+
+    # Load checkpoint
+    ckpt = torch.load(args.ckpt_path, map_location="cpu", weights_only=False)
+    state_dict = ckpt["state_dict"]
+    hparams = ckpt["hyper_parameters"]
+
+    # Process: Exclude keys and strip prefixes
+    parameters = {}
+    for k, v in state_dict.items():
+        if any(k.startswith(excl) for excl in args.exclude_keys):
+            continue
+
+        new_key = k
+        if args.strip_prefix and k.startswith(f"{args.strip_prefix}."):
+            new_key = k.replace(f"{args.strip_prefix}.", "", 1)
+
+        parameters[new_key] = v
+
+    if args.verbose:
+        print("The following keys will be saved:")
+        for key in parameters.keys():
+            print(f" - {key}")
+
+    # Save safetensors
+    output_path = args.ckpt_path.replace(".ckpt", ".safetensors")
+    save_file(parameters, output_path)
+    print(f"Safetensors file saved to {output_path}")
+
+    # Export config.json
+    hparams_dict = OmegaConf.to_container(hparams, resolve=False)
+
+    # We only save the 'model' key to keep the config clean
+    config_data = hparams_dict.get("model", hparams_dict)
+
+    config_path = os.path.join(os.path.dirname(output_path), "config.json")
+    with open(config_path, "w", encoding="utf-8") as f:
+        json.dump(config_data, f, indent=2)
+    print(f"Configuration saved to {config_path}")
+
+    # Verification
+    try:
+        loaded_params = load_file(output_path)
+        assert len(parameters) == len(loaded_params), "Mismatch in parameter count!"
+        for k in parameters:
+            # We verify the tensor shape and dtype, as torch.equal can be slow/strict
+            assert parameters[k].shape == loaded_params[k].shape, f"Shape mismatch: {k}"
+            assert parameters[k].dtype == loaded_params[k].dtype, f"Dtype mismatch: {k}"
+        print("Verification successful: File is valid.")
+    except Exception as e:
+        print(f"Verification failed: {e}")

pretraining/TinyMyo.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:33c44fb4db05b9673227400b1fcf89e4dee03e0bfaf1573772fc760a6c2287f6
+size 14291784

pretraining/config.json

@@ -0,0 +1,16 @@
+{
+  "_target_": "models.TinyMyo.TinyMyo",
+  "img_size": 1000,
+  "patch_size": 20,
+  "in_chans": 16,
+  "embed_dim": 192,
+  "n_layer": 8,
+  "n_head": 3,
+  "mlp_ratio": 4,
+  "qkv_bias": true,
+  "attn_drop": 0.1,
+  "proj_drop": 0.1,
+  "drop_path": 0.0,
+  "num_classes": 0,
+  "task": "pretraining"
+}

scripts/avespeech.py

@@ -0,0 +1,37 @@
+import argparse
+import os
+from huggingface_hub import snapshot_download
+
+
+def download_emg_only(save_dir: str):
+    """
+    Download only EMG folders from the AVE-Speech dataset.
+    Requires HF authentication if the dataset is gated.
+    """
+
+    repo_id = "MML-Group/AVE-Speech"
+
+    # Patterns to include only EMG folders across splits
+    allow_patterns = [
+        "Train/EMG/**",
+        "Dev/EMG/**",
+        "Test/EMG/**",
+    ]
+
+    snapshot_download(
+        repo_id=repo_id,
+        repo_type="dataset",
+        local_dir=save_dir,
+        allow_patterns=allow_patterns,
+        resume_download=True,
+    )
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Download AVE-Speech EMG data only")
+    parser.add_argument("--save_dir", type=str, required=True)
+    args = parser.parse_args()
+
+    os.makedirs(args.save_dir, exist_ok=True)
+
+    download_emg_only(args.save_dir)