Upload folder using huggingface_hub

README.md

@@ -1,3 +1,55 @@
----
-license: mit
----
+# QED-75M Web (ONNX)
+
+This repository contains the QED-75M language model exported to ONNX format for web deployment.
+
+## Files
+
+- `model.onnx` - The model weights in ONNX format
+- `tokenizer.json` - Tokenizer vocabulary and configuration
+- `tokenizer_config.json` - Tokenizer settings
+- `config.json` - Model architecture configuration
+
+## Usage
+
+### With ONNX Runtime (Python)
+
+```python
+from onnxruntime import InferenceSession
+import numpy as np
+
+# Load model
+session = InferenceSession("model.onnx")
+
+# Prepare input (tokenized text)
+input_ids = np.array([[1, 2, 3, ...]], dtype=np.int64)
+
+# Run inference
+outputs = session.run(None, {"input_ids": input_ids})
+logits = outputs[0]
+```
+
+### With Transformers.js (Browser/Node.js)
+
+```javascript
+import { AutoTokenizer, AutoModelForCausalLM } from '@xenova/transformers';
+
+const tokenizer = await AutoTokenizer.from_pretrained('./');
+const model = await AutoModelForCausalLM.from_pretrained('./');
+
+const inputs = await tokenizer('Hello, world!');
+const outputs = await model(inputs);
+```
+
+## Model Configuration
+
+- **Vocabulary Size**: 49,152
+- **Hidden Dimension**: 384
+- **Layers**: 32
+- **Attention Heads**: 6
+- **FFN Hidden Dimension**: 1,024
+- **Max Sequence Length**: 8,192
+- **RoPE Theta**: 10,000
+
+## License
+
+MIT

config.json

@@ -0,0 +1,25 @@
+{
+  "vocab_size": 49152,
+  "max_seq_len": 8192,
+  "d_model": 384,
+  "n_layers": 32,
+  "n_heads": 6,
+  "ffn_hidden_dim": 1024,
+  "rope_theta": 10000.0,
+  "rms_norm_eps": 1e-05,
+  "initializer_range": 0.02,
+  "dropout": 0.0,
+  "tie_word_embeddings": true,
+  "bias": false,
+  "pad_token_id": 0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "model_type": "qed",
+  "architectures": [
+    "QEDForCausalLM"
+  ],
+  "auto_map": {
+    "AutoConfig": "modeling_qed.QEDConfig",
+    "AutoModelForCausalLM": "modeling_qed.QEDForCausalLM"
+  }
+}

model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7fa884adddf6cec38f53e8254df2eccc4cbcf76cff2ea0bcc71cdbfe6353e1e2
+size 2936856

tokenizer_config.json

@@ -0,0 +1,13 @@
+{
+  "tokenizer_class": "PreTrainedTokenizerFast",
+  "vocab_size": 49152,
+  "model_max_length": 8192,
+  "pad_token": "<pad>",
+  "bos_token": "<bos>",
+  "eos_token": "<eos>",
+  "unk_token": "<unk>",
+  "pad_token_id": 0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "unk_token_id": 3
+}