Upload folder using huggingface_hub

README.md

@@ -1,6 +1,5 @@
 ---
 license: apache-2.0
-language: en
 tags:
 - biology
 - genomics
@@ -8,14 +7,12 @@ tags:
 library_name: transformers
 ---
 
-# TXModel - Standalone Version
+# TXModel - Hub-Ready Version
 
-**Zero external dependencies!** This model requires only:
-- `transformers`
-- `torch`
-- `safetensors`
-
-No llmfoundry, composer, or other libraries needed!
+**Zero-hassle deployment!** Requires ONLY:
+```bash
+pip install transformers torch safetensors
+```
 
 ## 🚀 Quick Start
 
@@ -23,18 +20,17 @@ No llmfoundry, composer, or other libraries needed!
 from transformers import AutoModel
 import torch
 
-# Load model (downloads automatically from Hub)
+# Load from Hub (one command!)
 model = AutoModel.from_pretrained(
-    "your-username/tx-model-standalone",
+    "your-username/model-name",
     trust_remote_code=True
 )
 
-# Prepare inputs
+# Use immediately
 genes = torch.randint(0, 100, (2, 10))
 values = torch.rand(2, 10)
 masks = torch.ones(2, 10).bool()
 
-# Inference
 model.eval()
 with torch.no_grad():
     output = model(genes=genes, values=values, gen_masks=masks)
@@ -42,116 +38,68 @@ with torch.no_grad():
 print(output.last_hidden_state.shape)  # [2, 10, d_model]
 ```
 
+## ✨ Features
+
+- ✅ **Single file** - all code in `modeling.py`
+- ✅ **Zero dependencies** (except transformers + torch)
+- ✅ **Works with AutoModel** out of the box
+- ✅ **No import errors** - everything self-contained
+
 ## 📦 Installation
 
 ```bash
 pip install transformers torch safetensors
 ```
 
-That's it! No other dependencies required.
+That's it!
 
 ## 🎯 Usage
 
-The model works exactly like any other HuggingFace model:
+### Basic Inference
 
 ```python
 from transformers import AutoModel
 
-# Load from Hub
 model = AutoModel.from_pretrained(
-    "your-username/tx-model-standalone",
+    "your-username/model-name",
     trust_remote_code=True
 )
 
-# Or load locally
-model = AutoModel.from_pretrained(
-    "./path/to/model",
-    trust_remote_code=True
-)
-
-# Move to GPU
+# Move to GPU if available
 device = "cuda" if torch.cuda.is_available() else "cpu"
 model = model.to(device)
-model.eval()
-
-# Your inference code here
-```
-
-## ⚡ Features
-
-- ✅ **Zero external dependencies** (only transformers + torch)
-- ✅ **Works with AutoModel** out of the box
-- ✅ **Hub-ready** - upload and share easily
-- ✅ **Same architecture** as original model
-- ✅ **Full compatibility** with existing weights
-
-## 📊 Model Details
-
-| Property | Value |
-|----------|-------|
-| Parameters | ~70M |
-| Architecture | Transformer Encoder |
-| Hidden Size | 512 |
-| Layers | 12 |
-| Attention Heads | 8 |
-
-## 🔧 Advanced Usage
-
-### Accessing Model Internals
-
-```python
-# Access the TXModel directly
-tx_model = model.tx_model
-
-# Get cell embeddings
-output = model(genes, values, masks)
-cell_emb = output.last_hidden_state[:, 0, :]  # CLS token
-
-# Get gene embeddings
-tx_output = tx_model(genes, values, masks, key_padding_mask=~genes.eq(0))
-gene_embs = tx_output["gene_embeddings"]  # If return_gene_embeddings=True
 ```
 
 ### Batch Processing
 
 ```python
-from torch.utils.data import DataLoader
-
-# Your dataloader
-dataloader = DataLoader(dataset, batch_size=32)
-
-results = []
-for batch in dataloader:
-    with torch.no_grad():
-        output = model(
-            genes=batch['genes'],
-            values=batch['values'],
-            gen_masks=batch['masks']
-        )
-        results.append(output.last_hidden_state)
+# Your data
+batch = {
+    'genes': torch.randint(0, 1000, (32, 100)),
+    'values': torch.rand(32, 100),
+    'masks': torch.ones(32, 100).bool()
+}
+
+# Process
+model.eval()
+with torch.no_grad():
+    output = model(**batch)
 ```
 
-## 🆚 vs Original Version
+## 📊 Model Details
 
-This standalone version:
-- ✅ Removes dependencies on llmfoundry and composer
-- ✅ Uses only PyTorch and Transformers components
-- ✅ Works with standard HuggingFace tools
-- ✅ Maintains same model architecture and weights
-- ✅ Easier to install and deploy
+- **Parameters**: ~70M
+- **Architecture**: Transformer Encoder
+- **Hidden Size**: 512
+- **Layers**: 12
+- **Heads**: 8
 
 ## 📝 Citation
 
-If you use this model, please cite the original work:
-
 ```bibtex
 @article{tahoe2024,
-  title={Tahoe-x1: Foundation Model for Genomics},
+  title={Tahoe-x1},
   author={...},
   year={2024}
 }
 ```
-
-## 📄 License
-
-Apache 2.0

config.json

@@ -50,7 +50,6 @@
     "query_activation": "sigmoid",
     "scaled_dot_product": true
   },
-  "chemical_encoder_config": null,
   "auto_map": {
     "AutoConfig": "modeling.TXConfig",
     "AutoModel": "modeling.TXModelForHF",

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:217637af5a4d12f3fe2d2648fb9d4d1404b53eea587336c62cfcfbfb26088efd
-size 284008108
+oid sha256:305a80c5941f512526a35c05b8e0f6d3dc930fdb01616546f1494ed76961600b
+size 284014476

tokenizer_config.json

@@ -1,5 +1,4 @@
 {
   "tokenizer_class": "PreTrainedTokenizerFast",
-  "model_max_length": 1000000000000000019884624838656,
   "vocab_size": 62720
 }