Downloads · 30 days
0
qwen2.5-0.5b-gsm8k
saad1926q/qwen2.5-0.5b-gsm8k
qwen2.5-0.5b-gsm8k is a text generation model from saad1926q. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
Fine-tuned from Qwen/Qwen2.5-0.5B-Instruct using GRPO (Group Relative Policy Optimization) on GSM8K grade school math problems.
Access
Public
Updated Jun 21, 2026
Repo size
211 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors423 MB · 100%
At a glance
- Task
- Text Generation
- Library
- transformers
- License
- apache-2.0
- Access
- Public
- Created
- Jun 21, 2026
- Updated
- Jun 21, 2026
- SHA
- 3157d8a6
Try a prompt
Base models
Datasets
Model card
From the Hugging Face model README
Qwen2.5-0.5B-Instruct GSM8K (GRPO)
Fine-tuned from Qwen/Qwen2.5-0.5B-Instruct using GRPO (Group Relative Policy Optimization) on GSM8K grade school math problems.
Training Details
- Method: GRPO with LoRA (rank 32, alpha 64)
- Dataset: openai/gsm8k
- Reward: Exact match on final numeric answer
- Steps: 1500
- Learning rate: 5e-6
- Gradient clipping: max_norm 0.5
- Group size: 16
- Hardware: 1x RTX 4090 (colocated train + inference)
- Framework: prime-rl + verifiers
Results
The model surpasses the base model's GSM8K baseline, with eval pass@1 peaking at step 1200.
| Checkpoint | eval pass@1 |
|---|---|
| step_1200 (peak) | 53.45% |
| step_1500 (final) | 52.84% |
Note: step_1200 was not saved due to my dumb ahh not setting up checkpoint saving before starting the run.
Checkpoints
step_1225— closest to peak performancestep_1300step_1500— final checkpoint
- Task
- Text Generation
- Library
- transformers
- License
- apache-2.0
- Languages
- en
- Created
- Jun 21, 2026
- Updated
- Jun 21, 2026