Skip to page content
SAVRN
Search Plan a project

Open-weight model

visualrl-qwen3vl2b-ppo1-4cx-c50-clean

by Zeyuan Yang Miiche/visualrl-qwen3vl2b-ppo1-4cx-c50-clean

visualrl-qwen3vl2b-ppo1-4cx-c50-clean is an open-weight model from Zeyuan Yang. Its published files total 31.0 GB.

Parameters—
Context—
Weights14.6 GB
License—
AccessOpen weights
Monthly Downloads—

Model Card

The publisher has not written a card for this model.

Identity and Version

Repository
Miiche/visualrl-qwen3vl2b-ppo1-4cx-c50-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
c12bcc775eb2ffbd9f17470f36c74a0cb42c73d3
First published
2026-09-23
Last updated
2026-09-24

Files and Weights

176 files, 31.0 GB in total. The weights are 3 files totalling 14.6 GB in safetensors.

Weights3 files · 14.6 GB
Configuration108 files · 279.2 MB
Tokenizer12 files · 47.6 MB
Other52 files · 16.0 GB
Repository1 file · 2.7 KB
Every file
FileTypeSizeSHA-256
global_step_180_merged/model.safetensorsWeights4.9 GB 2f8888b891a6
global_step_270_merged/model.safetensorsWeights4.9 GB 88a68988dbae
global_step_90_merged/model.safetensorsWeights4.9 GB 19c4f40ad245
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/blink_output.jsonConfiguration6.5 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/blink_scores.jsonConfiguration3.2 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/chartqa_output.jsonConfiguration2.0 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/chartqa_scores.jsonConfiguration193 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/hallusionbench_output.jsonConfiguration1.7 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/hallusionbench_scores.jsonConfiguration491 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/logicvista_output.jsonConfiguration3.3 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/logicvista_scores.jsonConfiguration501 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mathverse_output.jsonConfiguration18.4 MB 15857704320d
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mathverse_scores.jsonConfiguration2.3 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mathvision_output.jsonConfiguration3.5 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mathvision_scores.jsonConfiguration8.9 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mathvista_output.jsonConfiguration4.3 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mathvista_scores.jsonConfiguration7.5 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmbench_output.jsonConfiguration10.4 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmbench_scores.jsonConfiguration6.4 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mme_output.jsonConfiguration2.6 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mme_scores.jsonConfiguration2.0 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmmu_output.jsonConfiguration5.3 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmmu_pro_output.jsonConfiguration196.3 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmmu_pro_scores.jsonConfiguration110 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmmu_scores.jsonConfiguration2.7 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmstar_output.jsonConfiguration5.1 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmstar_scores.jsonConfiguration3.1 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmvet_output.jsonConfiguration275.8 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmvet_scores.jsonConfiguration527 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/perceptionbench_output.jsonConfiguration16.4 MB e2352f1167e3
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/perceptionbench_scores.jsonConfiguration1.0 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/realworldqa_output.jsonConfiguration1.2 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/realworldqa_scores.jsonConfiguration95 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/scienceqa_output.jsonConfiguration4.4 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/scienceqa_scores.jsonConfiguration358 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/visnumbench_output.jsonConfiguration8.9 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/visnumbench_scores.jsonConfiguration955 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/blink_output.jsonConfiguration7.7 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/blink_scores.jsonConfiguration3.2 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/chartqa_output.jsonConfiguration2.3 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/chartqa_scores.jsonConfiguration193 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/logicvista_output.jsonConfiguration3.9 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/logicvista_scores.jsonConfiguration502 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/mmbench_output.jsonConfiguration11.0 MB 56fc7f1ef4ec
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/mmbench_scores.jsonConfiguration6.4 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/mme_output.jsonConfiguration2.7 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/mme_scores.jsonConfiguration2.0 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/mmstar_output.jsonConfiguration5.5 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/mmstar_scores.jsonConfiguration3.1 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/mmvet_output.jsonConfiguration273.0 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/mmvet_scores.jsonConfiguration504 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/perceptionbench_output.jsonConfiguration21.6 MB 96a07712ae54
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/perceptionbench_scores.jsonConfiguration1.1 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/realworldqa_output.jsonConfiguration1.2 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/realworldqa_scores.jsonConfiguration94 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/scienceqa_output.jsonConfiguration4.3 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/scienceqa_scores.jsonConfiguration359 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/visnumbench_output.jsonConfiguration10.0 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/visnumbench_scores.jsonConfiguration948 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/blink_output.jsonConfiguration8.4 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/blink_scores.jsonConfiguration3.2 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/chartqa_output.jsonConfiguration2.1 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/chartqa_scores.jsonConfiguration193 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/hallusionbench_output.jsonConfiguration1.8 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/hallusionbench_scores.jsonConfiguration492 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/logicvista_output.jsonConfiguration3.8 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/logicvista_scores.jsonConfiguration501 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mathverse_output.jsonConfiguration21.8 MB dff4e0239c04
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mathverse_scores.jsonConfiguration2.3 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mathvision_output.jsonConfiguration4.1 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mathvision_scores.jsonConfiguration8.9 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mathvista_output.jsonConfiguration4.9 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mathvista_scores.jsonConfiguration7.5 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmbench_output.jsonConfiguration11.8 MB d82a161669e0
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmbench_scores.jsonConfiguration6.3 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mme_output.jsonConfiguration2.5 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mme_scores.jsonConfiguration2.0 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmmu_output.jsonConfiguration6.1 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmmu_pro_output.jsonConfiguration196.3 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmmu_pro_scores.jsonConfiguration110 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmmu_scores.jsonConfiguration2.7 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmstar_output.jsonConfiguration6.2 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmstar_scores.jsonConfiguration3.1 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmvet_output.jsonConfiguration232.6 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmvet_scores.jsonConfiguration526 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/perceptionbench_output.jsonConfiguration23.0 MB 542ccba90813
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/perceptionbench_scores.jsonConfiguration1.0 KB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/realworldqa_output.jsonConfiguration1.4 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/realworldqa_scores.jsonConfiguration95 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/scienceqa_output.jsonConfiguration5.2 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/scienceqa_scores.jsonConfiguration359 B —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/visnumbench_output.jsonConfiguration10.2 MB —
eval_results/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/visnumbench_scores.jsonConfiguration954 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/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/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/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-q3vl2b_ppo1_4cx_c50.logOther16.0 GB a67c261b86f1
logs/eval-visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm.logOther1.6 MB —
logs/eval-visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm.logOther1.6 MB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/blink.logOther141.5 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/chartqa.logOther91.5 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/hallusionbench.logOther47.7 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/logicvista.logOther35.9 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mathverse.logOther198.7 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mathvision.logOther29.9 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mathvista.logOther59.7 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmbench.logOther157.2 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mme.logOther101.4 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmmu.logOther85.3 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmmu_pro.logOther125.3 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmstar.logOther73.2 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/mmvet.logOther44.9 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/perceptionbench.logOther197.0 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/realworldqa.logOther58.1 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/scienceqa.logOther84.1 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s180-vllm/visnumbench.logOther96.0 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/blink.logOther146.5 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/chartqa.logOther97.8 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/logicvista.logOther37.4 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/mmbench.logOther156.9 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/mme.logOther104.2 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/mmstar.logOther75.2 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/mmvet.logOther44.6 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/perceptionbench.logOther196.6 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/realworldqa.logOther61.0 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/scienceqa.logOther84.7 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s270-vllm/visnumbench.logOther97.7 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/blink.logOther147.5 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/chartqa.logOther90.9 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/hallusionbench.logOther49.5 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/logicvista.logOther36.2 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mathverse.logOther203.8 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mathvision.logOther27.5 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mathvista.logOther61.1 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmbench.logOther161.4 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mme.logOther99.2 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmmu.logOther83.6 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmmu_pro.logOther119.9 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmstar.logOther74.3 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/mmvet.logOther45.3 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/perceptionbench.logOther185.2 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/realworldqa.logOther58.6 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/scienceqa.logOther87.3 KB —
logs/visualrl-qwen3vl2b-ppo1-4cx-c50-clean-s90-vllm/visnumbench.logOther96.3 KB —
train_logs/train_metrics_q3vl2b_ppo1_4cx_c50.jsonlOther542.3 KB —
.gitattributesRepository2.7 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
14.6 GB
Download from Zeyuan Yang

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

Memory Requirements

PrecisionWeights in memory
As published14.6 GB

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