Create README.md

README.md

@@ -0,0 +1,43 @@
+---
+license: mit
+train: false
+inference: false
+pipeline_tag: text-generation
+---
+This is a version of the <a href="https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B">DeepSeek-R1-Distill-Qwen-1.5B</a> model re-distilled for better performance.
+
+## Performance
+
+| Models            | <a href="https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B">DeepSeek-R1-Distill-Qwen-1.5B</a> | DeepSeek-R1-ReDistill-Qwen-1.5B-v1.1 | 
+|:-------------------:|:--------:|:----------------:|
+| ARC (25-shot)      | 40.96 | <b>41.3</b>  | 
+| HellaSwag (10-shot)| 44    | <b>45.22</b> |
+| MMLU (5-shot)      | 39.27 | <b>42.01</b> | 
+| TruthfulQA-MC2     | 45.17 | <b>46.64</b> | 
+| Winogrande (5-shot)| 55.49 | <b>56.75</b> | 
+| GSM8K (5-shot)     | 69.9  | <b>73.24</b> | 
+| Average            | 49.13 | <b>50.86</b> | 
+
+| Models            | <a href="https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B">DeepSeek-R1-Distill-Qwen-1.5B</a> | DeepSeek-R1-ReDistill-Qwen-1.5B-v1.1 | 
+|:-------------------:|:--------:|:----------------:|
+| GPQA (0-shot)     | 26.96 | <b>27.8</b>  | 
+| MMLU PRO (5-shot) | 16.74 | <b>19.44</b> | 
+| MUSR (0-shot)     | 35.93 | <b>35.94</b> | 
+| BBH (3-shot)      | 35.12 | 35.11 | 
+| Average           |  |  | 
+
+## Usage
+```Python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+compute_dtype = torch.bfloat16
+device   = 'cuda'
+model_id = "mobiuslabsgmbh/DeepSeek-R1-ReDistill-Qwen-1.5B-v1.1"
+
+model     = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=compute_dtype, attn_implementation=attn_implementation, device_map=device)
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+
+chat    = tokenizer.apply_chat_template([{"role":"user", "content":"What is 1.5+102.2?"}], tokenize=True, add_generation_prompt=True, return_tensors="pt")
+outputs = model_student.generate(chat.to(device), max_new_tokens=1024, do_sample=True) 
+print(tokenizer.decode(outputs[0]))
+```
+