Update README.md

README.md

@@ -1,3 +1,44 @@
----
-license: mit
----
+---
+license: mit
+---
+
+# TinyLlama-NoPE-1.1B
+
+NoPE is a transformer model without positional encoding.
+
+The model is trained following TinyLlama code base (https://github.com/jzhang38/TinyLlama)
+
+## Usage
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers.models.llama import modeling_llama
+
+
+def nope_monkey_patch(q, k, cos, sin, position_ids, unsqueeze_dim=1):
+    return q, k
+
+
+modeling_llama.apply_rotary_pos_emb = nope_monkey_patch
+
+model_path = "AntNLP/TinyLlama-NoPE-1.1B"
+tokenizer = AutoTokenizer.from_pretrained(model_path)
+model = AutoModelForCausalLM.from_pretrained(model_path).cuda()
+
+input_ids = tokenizer("Hello, TinyLlama-NoPE", return_tensors="pt").input_ids.cuda()
+output = model.generate(input_ids, do_sample=True, max_length=50)
+print(tokenizer.decode(output[0], skip_special_tokens=True))
+```
+
+## Citation
+
+```
+@misc{wang2024length,
+      title={Length Generalization of Causal Transformers without Position Encoding}, 
+      author={Jie Wang and Tao Ji and Yuanbin Wu and Hang Yan and Tao Gui and Qi Zhang and Xuanjing Huang and Xiaoling Wang},
+      year={2024},
+      eprint={2404.12224},
+      archivePrefix={arXiv},
+      primaryClass={cs.CL}
+}
+```