Skip to page content
SAVRN
Search Plan a project

Open-weight model

visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean

by Zeyuan Yang Miiche/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean

visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean is an open-weight model from Zeyuan Yang. Its published files total 38.7 GB.

Parameters—
Context—
Weights29.0 GB
License—
AccessOpen weights
Monthly Downloads—

Model Card

The publisher has not written a card for this model.

Identity and Version

Repository
Miiche/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean
Publisher
Zeyuan Yang
Task
Not stated by the source
Modality
Other
Library
Not stated by the source
Parameters
Not stated by the source
Languages
Not stated by the source
Revision
3387bce225c32843a4cfb67d71cea876fca37215
First published
2026-09-20
Last updated
2026-09-22

Files and Weights

159 files, 38.7 GB in total. The weights are 7 files totalling 29.0 GB in safetensors.

Weights7 files · 29.0 GB
Configuration95 files · 179.4 MB
Tokenizer12 files · 47.6 MB
Other44 files · 9.5 GB
Repository1 file · 2.3 KB
Every file
FileTypeSizeSHA-256
global_step_180_merged/model-00001-of-00002.safetensorsWeights5.0 GB 7345b4671efb
global_step_180_merged/model-00002-of-00002.safetensorsWeights4.7 GB 361f91316c1d
global_step_270_merged/model-00001-of-00002.safetensorsWeights5.0 GB 40511aaea2c4
global_step_270_merged/model-00002-of-00002.safetensorsWeights4.7 GB e81d9a5f3edb
global_step_90_merged/model-00001-of-00003.safetensorsWeights4.3 GB 039abf314151
global_step_90_merged/model-00002-of-00003.safetensorsWeights5.0 GB ace8236ee318
global_step_90_merged/model-00003-of-00003.safetensorsWeights351.8 MB 2e73512801f1
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/blink_output.jsonConfiguration5.1 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/blink_scores.jsonConfiguration3.2 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/chartqa_output.jsonConfiguration3.0 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/chartqa_scores.jsonConfiguration191 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/hallusionbench_output.jsonConfiguration1.5 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/hallusionbench_scores.jsonConfiguration492 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/logicvista_output.jsonConfiguration2.0 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/logicvista_scores.jsonConfiguration500 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mathverse_output.jsonConfiguration11.6 MB 5db4c422fc31
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mathverse_scores.jsonConfiguration2.3 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mathvision_output.jsonConfiguration2.2 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mathvision_scores.jsonConfiguration9.0 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mathvista_output.jsonConfiguration3.0 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mathvista_scores.jsonConfiguration7.5 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmbench_output.jsonConfiguration8.0 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmbench_scores.jsonConfiguration6.4 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mme_output.jsonConfiguration2.9 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mme_scores.jsonConfiguration2.0 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmmu_output.jsonConfiguration4.2 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmmu_pro_output.jsonConfiguration196.1 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmmu_pro_scores.jsonConfiguration108 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmmu_scores.jsonConfiguration2.7 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmstar_output.jsonConfiguration3.5 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmstar_scores.jsonConfiguration3.1 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmvet_output.jsonConfiguration199.1 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmvet_scores.jsonConfiguration502 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/perceptionbench_output.jsonConfiguration11.0 MB 3df75c4b377a
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/perceptionbench_scores.jsonConfiguration1.1 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/realworldqa_output.jsonConfiguration898.2 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/realworldqa_scores.jsonConfiguration94 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/scienceqa_output.jsonConfiguration2.7 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/scienceqa_scores.jsonConfiguration359 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/visnumbench_output.jsonConfiguration4.4 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/visnumbench_scores.jsonConfiguration951 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s270-vllm/chartqa_output.jsonConfiguration4.1 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s270-vllm/chartqa_scores.jsonConfiguration193 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s270-vllm/mmvet_output.jsonConfiguration212.7 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s270-vllm/mmvet_scores.jsonConfiguration516 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s270-vllm/visnumbench_output.jsonConfiguration10.0 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s270-vllm/visnumbench_scores.jsonConfiguration954 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/blink_output.jsonConfiguration7.6 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/blink_scores.jsonConfiguration3.2 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/chartqa_output.jsonConfiguration3.8 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/chartqa_scores.jsonConfiguration193 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/hallusionbench_output.jsonConfiguration2.0 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/hallusionbench_scores.jsonConfiguration491 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/logicvista_output.jsonConfiguration3.0 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/logicvista_scores.jsonConfiguration501 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mathverse_output.jsonConfiguration17.8 MB fa0a3c5fcfa1
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mathverse_scores.jsonConfiguration2.3 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mathvision_output.jsonConfiguration3.0 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mathvision_scores.jsonConfiguration9.0 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mathvista_output.jsonConfiguration4.0 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mathvista_scores.jsonConfiguration7.5 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmbench_output.jsonConfiguration9.7 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmbench_scores.jsonConfiguration6.4 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mme_output.jsonConfiguration5.3 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mme_scores.jsonConfiguration2.0 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmmu_output.jsonConfiguration5.4 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmmu_pro_output.jsonConfiguration196.1 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmmu_pro_scores.jsonConfiguration108 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmmu_scores.jsonConfiguration2.7 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmstar_output.jsonConfiguration4.9 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmstar_scores.jsonConfiguration3.1 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmvet_output.jsonConfiguration198.2 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmvet_scores.jsonConfiguration502 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/perceptionbench_output.jsonConfiguration19.8 MB bcfe9b6969f9
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/perceptionbench_scores.jsonConfiguration1.1 KB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/realworldqa_output.jsonConfiguration1.2 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/realworldqa_scores.jsonConfiguration94 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/scienceqa_output.jsonConfiguration3.4 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/scienceqa_scores.jsonConfiguration359 B —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/visnumbench_output.jsonConfiguration7.3 MB —
eval_results/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/visnumbench_scores.jsonConfiguration936 B —
global_step_180_merged/added_tokens.jsonConfiguration707 B —
global_step_180_merged/config.jsonConfiguration1.6 KB —
global_step_180_merged/generation_config.jsonConfiguration147 B —
global_step_180_merged/model.safetensors.index.jsonConfiguration64.8 KB —
global_step_180_merged/preprocessor_config.jsonConfiguration782 B —
global_step_180_merged/special_tokens_map.jsonConfiguration613 B —
global_step_180_merged/video_preprocessor_config.jsonConfiguration817 B —
global_step_270_merged/added_tokens.jsonConfiguration707 B —
global_step_270_merged/config.jsonConfiguration1.6 KB —
global_step_270_merged/generation_config.jsonConfiguration147 B —
global_step_270_merged/model.safetensors.index.jsonConfiguration64.8 KB —
global_step_270_merged/preprocessor_config.jsonConfiguration782 B —
global_step_270_merged/special_tokens_map.jsonConfiguration613 B —
global_step_270_merged/video_preprocessor_config.jsonConfiguration817 B —
global_step_90_merged/added_tokens.jsonConfiguration707 B —
global_step_90_merged/config.jsonConfiguration1.6 KB —
global_step_90_merged/generation_config.jsonConfiguration147 B —
global_step_90_merged/model.safetensors.index.jsonConfiguration64.8 KB —
global_step_90_merged/preprocessor_config.jsonConfiguration782 B —
global_step_90_merged/special_tokens_map.jsonConfiguration613 B —
global_step_90_merged/video_preprocessor_config.jsonConfiguration817 B —
global_step_180_merged/chat_template.jinjaOther5.3 KB —
global_step_270_merged/chat_template.jinjaOther5.3 KB —
global_step_90_merged/chat_template.jinjaOther5.3 KB —
logs/eval-train-q3vl4b_ppo1_4bx_c50_hires.logOther9.5 GB b37d2f976ba7
logs/eval-visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm.logOther2.3 MB —
logs/eval-visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm.logOther1.0 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/blink.logOther166.6 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/chartqa.logOther106.1 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/hallusionbench.logOther56.0 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/logicvista.logOther54.2 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mathverse.logOther306.9 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mathvision.logOther47.5 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mathvista.logOther77.5 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmbench.logOther190.4 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mme.logOther114.4 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmmu.logOther111.0 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmmu_pro.logOther174.7 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmstar.logOther94.5 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/mmvet.logOther165.5 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/perceptionbench.logOther275.9 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/realworldqa.logOther58.0 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/scienceqa.logOther96.3 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s180-vllm/visnumbench.logOther165.5 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s270-vllm/chartqa.logOther104.7 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s270-vllm/mmvet.logOther167.8 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s270-vllm/visnumbench.logOther100.3 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/blink.logOther162.7 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/chartqa.logOther108.4 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/hallusionbench.logOther60.7 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/logicvista.logOther51.5 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mathverse.logOther317.0 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mathvision.logOther42.0 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mathvista.logOther78.8 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmbench.logOther197.1 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mme.logOther126.0 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmmu.logOther105.4 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmmu_pro.logOther160.6 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmstar.logOther97.7 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/mmvet.logOther165.2 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/perceptionbench.logOther266.9 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/realworldqa.logOther58.5 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/scienceqa.logOther97.2 KB —
logs/visualrl-qwen3vl4b-ppo1-4bx-c50-hires-clean-s90-vllm/visnumbench.logOther175.7 KB —
train_logs/train_metrics_q3vl4b_ppo1_4bx_c50_hires.jsonlOther362.5 KB —
.gitattributesRepository2.3 KB —
global_step_180_merged/merges.txtTokenizer1.7 MB —
global_step_180_merged/tokenizer.jsonTokenizer11.4 MB aeb13307a71a
global_step_180_merged/tokenizer_config.jsonTokenizer5.5 KB —
global_step_180_merged/vocab.jsonTokenizer2.8 MB —
global_step_270_merged/merges.txtTokenizer1.7 MB —
global_step_270_merged/tokenizer.jsonTokenizer11.4 MB aeb13307a71a
global_step_270_merged/tokenizer_config.jsonTokenizer5.5 KB —
global_step_270_merged/vocab.jsonTokenizer2.8 MB —
global_step_90_merged/merges.txtTokenizer1.7 MB —
global_step_90_merged/tokenizer.jsonTokenizer11.4 MB aeb13307a71a
global_step_90_merged/tokenizer_config.jsonTokenizer5.5 KB —
global_step_90_merged/vocab.jsonTokenizer2.8 MB —

License and Download

License
Not stated by the source
Access
Open weights, no gate
Download size
29.0 GB
Download from Zeyuan Yang

Released by Zeyuan Yang through its official repository on Hugging Face.

Memory Requirements

PrecisionWeights in memory
As published29.0 GB

Weights only, from the published parameter count; the key-value cache and runtime add to this.