Downloads · 30 days
0
pantomiman/qwen3.5-0.8b-grpo-musique-m5_6_s43
qwen3.5-0.8b-grpo-musique-m5_6_s43 is a machine learning model from pantomiman. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
One run of the M5 reward-shape x seed ablation (M9.5f). Qwen3.5-0.8B trained with GRPO + a Wikipedia search tool on MuSiQue; this run uses the EM-only reward at seed 43 on Blackwell.
Downloads · 30 days
0
Access
Public
Updated Jun 3, 2026
Repo size
192 GB
Likes
0
Public
Click a slice to open those files.
.safetensors164 GB · 85%
From the Hugging Face model README
One run of the M5 reward-shape x seed ablation (M9.5f). Qwen3.5-0.8B trained with GRPO + a Wikipedia search tool on MuSiQue; this run uses the EM-only reward at seed 43 on Blackwell.
| field | value |
|---|---|
| run code | m5_6_s43 |
| reward shape | EM-only |
| seed | 43 |
| hardware | Blackwell (DP=4 colocated vLLM) |
| checkpoints | 24 (step_10..step_240, every 10 steps) |
| rollouts | 247 |
| best deployment ckpt (held-out avg EM) | step_120 (0.264 @step_120) |
| milestone | M9.5f |
Reward: reward = 1.0 if exact_match else 0.0 (Search-R1 paper-faithful).
step_<N>/policy/weights/model/consolidated/model-*.safetensors # deployable consolidated model
step_<N>/policy/tokenizer/ # tokenizer (or backfill from Qwen/Qwen3.5-0.8B)
rollouts/train_data_step<N>.jsonl # per-step training rollouts
config_snapshot.yaml # GRPO config for this run
pantomiman/qwen3.5-0.8b-grpo-musique-m5pantomiman/reason-over-search-eval-m5 (m5_6_s43/...)