Downloads · 30 days
8
28% of all-time downloads
weixu-zhang/qwen35-2b-grpo-phase3e-8lang3k-step100-promptv4
qwen35-2b-grpo-phase3e-8lang3k-step100-promptv4 is a machine learning model from weixu-zhang. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
This is the merged Hugging Face export for the interrupted VisRL Qwen3.5-2B retrain run:
Downloads · 30 days
8
28% of all-time downloads
All-time downloads
29
Public
Parameters
2.7B
5.5 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors5.4 GB · 99%
From the Hugging Face model README
This is the merged Hugging Face export for the interrupted VisRL Qwen3.5-2B retrain run:
global_step_100qwen35_2b_retrain_20260525_promptv4_compact_raw_vllm_n2_r1152_b32_8gpu_singlepass_frombaseThe run was intentionally frozen at step 100 on 2026-05-25 because the trainer
had progressed past the latest saved checkpoint but became unhealthy: GPU
utilization stayed at 0% while reward/Jupyter rendering workers were timing out.
The step 100 checkpoint was kept as the clean resumable point and exported for
evaluation. Raw FSDP resume shards are mirrored separately in the private
weixu-zhang/visrl-resume-checkpoints dataset.
This checkpoint is an early pilot candidate, not the final 2B paper checkpoint. Use the paper evaluation CSVs in the VisRL repository for current measured quality.