Upload README.md with huggingface_hub

README.md

@@ -1,80 +1,35 @@
 ---
-base_model: Qwen/Qwen2.5-0.5B
-library_name: transformers
-model_name: qwen2.5-0.5b-gsm8k-sft
 tags:
-- generated_from_trainer
-- trl
-- hf_jobs
-- sft
-- ml-intern
-licence: license
+  - sft
+  - gsm8k
+  - math
+base_model: Qwen/Qwen2.5-0.5B
 ---
 
-# Model Card for qwen2.5-0.5b-gsm8k-sft
-
-This model is a fine-tuned version of [Qwen/Qwen2.5-0.5B](https://huggingface.co/Qwen/Qwen2.5-0.5B).
-It has been trained using [TRL](https://github.com/huggingface/trl).
-
-## Quick start
-
-```python
-from transformers import pipeline
-
-question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
-generator = pipeline("text-generation", model="pngwn/qwen2.5-0.5b-gsm8k-sft", device="cuda")
-output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
-print(output["generated_text"])
-```
-
-## Training procedure
-
- 
-
-
-
-This model was trained with SFT.
-
-### Framework versions
-
-- TRL: 1.5.1
-- Transformers: 5.10.2
-- Pytorch: 2.12.0
-- Datasets: 5.0.0
-- Tokenizers: 0.22.2
-
-## Citations
-
-
-
-Cite TRL as:
-    
-```bibtex
-@software{vonwerra2020trl,
-  title   = {{TRL: Transformers Reinforcement Learning}},
-  author  = {von Werra, Leandro and Belkada, Younes and Tunstall, Lewis and Beeching, Edward and Thrush, Tristan and Lambert, Nathan and Huang, Shengyi and Rasul, Kashif and Gallouédec, Quentin},
-  license = {Apache-2.0},
-  url     = {https://github.com/huggingface/trl},
-  year    = {2020}
-}
-```
+# Qwen2.5-0.5B GSM8K SFT
 
-<!-- ml-intern-provenance -->
-## Generated by ML Intern
+Supervised fine-tuned model for grade-school math reasoning on GSM8K.
 
-This model repository was generated by [ML Intern](https://github.com/huggingface/ml-intern), an agent for machine learning research and development on the Hugging Face Hub.
+## Results
 
-- Try ML Intern: https://smolagents-ml-intern.hf.space
-- Source code: https://github.com/huggingface/ml-intern
+| Model | GSM8K test exact-match accuracy | N eval |
+|-------|-----------------------------------|--------|
+| Base (Qwen/Qwen2.5-0.5B) | 0.0008 (1/1319) | 1319 |
+| Tuned (pngwn/qwen2.5-0.5b-gsm8k-sft) | 0.3472 (458/1319) | 1319 |
 
-## Usage
+## Training details
 
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
+- **Dataset:** openai/gsm8k (main config)
+- **Train split:** 7473 samples
+- **Test split:** 1319 samples
+- **Epochs:** 3
+- **Learning rate:** 2e-5
+- **Batch size:** 4 per device
+- **Gradient accumulation:** 4
+- **Max sequence length:** 1024
+- **Decoding:** greedy (do_sample=False, max_new_tokens=256)
+- **Answer extraction:** regex `####\s*(-?\d+(?:,\d+)*(?:\.\d+)?)`
 
-model_id = 'pngwn/qwen2.5-0.5b-gsm8k-sft'
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(model_id)
-```
+## Eval script
 
-For non-causal architectures, replace `AutoModelForCausalLM` with the appropriate `AutoModel` class.
+The exact eval script used for both baseline and tuned evaluation is included in this repository as `eval_gsm8k.py`.