Skip to content
ModelVerifiedExternal
Download

Qwen3.5 0.8b Grpo Musique H200 A4 Seed42

license: mit tags: retrieval-augmented-generation grpo search-r1 qwen3.5 basemodel: Qwen/Qwen3.5-0.8B prettyname: "M5 GRPO MuSiQue m51s42 (F1-only, seed 42)"

Sourcehuggingface.co/pantomiman/qwen3.5-0.8b-grpo-musique-h200-a4-seed42/tree/ff13466865d79675aaf4d5d8435ca9a96a22580c
KindOtherBase modelQwen/Qwen3.5-0.8BVersionPublisher@pantomimanCgrade

Model source

Kind
Other
File size
3.18 GB
Base model
Qwen/Qwen3.5-0.8B
Version
vmodel-00001-of-00001
Updated
Oct 8, 2026
Publisher
@pantomiman
Parameters
8B
Source
Hugging Face

Source description

Qwen3.5-0.8B GRPO on MuSiQue: run m5_1_s42 (F1-only, seed 42)

One run of the M5 reward-shape x seed ablation (M9). Qwen3.5-0.8B trained with GRPO + a Wikipedia search tool on MuSiQue; this run uses the F1-only reward at seed 42 on H200.

fieldvalue
run codem5_1_s42
reward shapeF1-only
seed42
hardwareH200 (DP=4 colocated vLLM)
checkpoints18 (step_10..step_180, every 10 steps)
rollouts181
best deployment ckpt (held-out avg EM)step_110 (0.313 @step_110)
milestoneM9

Reward: reward = f1(answer, gold); no floor, no format gate.

Repo structure (uniform across the 9 M5 runs)

step_ /policy/weights/model/consolidated/model-*.safetensors # deployable consolidated model
step_ /policy/tokenizer/ # tokenizer (or backfill from Qwen/Qwen3.5-0.8B)
rollouts/train_data_step .jsonl # per-step training rollouts
config_snapshot.yaml # GRPO config for this run

Family

Sources

1 source
Hugging FacePrimaryVersion ff13466865d79675aaf4d5d8435ca9a96a22580c
huggingface.co/pantomiman/qwen3.5-0.8b-grpo-musique-h200-a4-seed42/tree/ff13466865d79675aaf4d5d8435ca9a96a22580c

Verified Sep 14

Model artifacts

1 artifact

step_10/policy/weights/model/consolidated/model-00001-of-00001.safetensors

safetensors · 3.18 GB · SHA-256 9658a10948e4…be0a · Hugging Face

Download

Source excerpts

3 excerpts
ReadmeJul 11huggingface.co/pantomiman/qwen3.5-0.8b-grpo-musique-m5_1_s42

--- license: mit tags: - retrieval-augmented-generation - grpo - search-r1 - qwen3.5 base_model: Qwen/Qwen3.5-0.8B pretty_name: "M5 GRPO MuSiQue m5_1_s42 (F1-only, seed 42)" --- # Qwen3.5-0.8B GRPO on MuSiQue: run m5_1_s42 (F1-only, seed 42) One run of the **M5 reward-shape x seed ablation** (M9). Qwen3.5-0.8B trained with GRPO + a Wikipedia search tool on MuSiQue; this run uses the **F1-only** reward at **seed 42** on **H200**. | field | value | |---|---| | run code | `m5_1_s42` | | reward shape | F1-only | | seed | 42 | | hardware | H200 (DP=4 colocated vLLM) | | checkpoints | 18 (step_10..step_180, every 10 steps) | | rollouts | 181 | | best deployment ckpt (held-out avg EM) | **step_110** (0.313 @step_110) | | milestone | M9 | **Reward**: `reward = f1(answer, gold)`; no floor, no format gate. ## Repo structure (uniform across the 9 M5 runs) ``` 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 ``` ## Family - Index + full 9-run matrix: [`pantomiman/qwen3.5-0.8b-grpo-musique-m5`](https://huggingface.co/datasets/pantomiman/qwen3.5-0.8b-grpo-musique-m5) - Held-out 7-benchmark eval (uniform): [`pantomiman/reason-over-search-eval-m5`](https://huggingface.co/datasets/pantomiman/reason-over-search-eval-m5) (`m5_1_s42/...`) - Collection: see the index repo. Recipe varies only by reward shape (m5_1=F1-only, m5_5=F1+format, m5_6=EM-only) x seed (42/43/44).

NoteJul 11huggingface.co/pantomiman/qwen3.5-0.8b-grpo-musique-m5_1_s42

Source context: 0 downloads · 0 likes · Repo pantomiman/qwen3.5-0.8b-grpo-musique-m5_1_s42

NoteMay 17huggingface.co/api/models/pantomiman/qwen3.5-0.8b-grpo-musique-h200-a4-seed42

Source context: 0 downloads · 0 likes · Pipeline text-generation · Repo pantomiman/qwen3.5-0.8b-grpo-musique-h200-a4-seed42