Downloads · 30 days
0
pantomiman/qwen3.5-0.8b-grpo-musique-m5_5_s43
qwen3.5-0.8b-grpo-musique-m5_5_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 (M9.5b). Qwen3.5-0.8B trained with GRPO + a Wikipedia search tool on MuSiQue; this run uses the F1+format reward at seed 43 on Blackwell.
Downloads · 30 days
0
Access
Public
Updated Jun 3, 2026
Repo size
199 GB
Likes
2
Public
Click a slice to open those files.
.safetensors171 GB · 86%
From the Hugging Face model README
One run of the M5 reward-shape x seed ablation (M9.5b). Qwen3.5-0.8B trained with GRPO + a Wikipedia search tool on MuSiQue; this run uses the F1+format reward at seed 43 on Blackwell.
| field | value |
|---|---|
| run code | m5_5_s43 |
| reward shape | F1+format |
| seed | 43 |
| hardware | Blackwell (DP=4 colocated vLLM) |
| checkpoints | 25 (step_10..step_250, every 10 steps) |
| rollouts | 251 |
| best deployment ckpt (held-out avg EM) | step_250 (0.317 @step_250) |
| milestone | M9.5b |
Reward: reward = f1 if f1>0, else 0.1 if format ok, else 0 (non-additive 3-tier).
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_5_s43/...)