Search results

请问MMMUPro_full的分数是将mmmu_pro_std4、mmmu_pro_std10和mmmu_pro_vision全测了然后取平均吗？

Status: Open.

#1984 In QwenLM/Qwen3-VL;

· JjjjjZzz opened on Jan 8, 2026

为什么推理时加载了lora权重后显存占用反而变低

Status: Open.

#1983 In QwenLM/Qwen3-VL;

· ggbondrighthere24 opened on Jan 7, 2026

What is the recommended SamplingParams configuration for vLLM offline inference to match Hugging Face generation quality?

Status: Open.

#1982 In QwenLM/Qwen3-VL;

· siddhi-lipare opened on Jan 7, 2026

[Fine Tune] Self-made training dataset results in Loss = 0

Status: Open.

#1981 In QwenLM/Qwen3-VL;

· piqiuni opened on Jan 7, 2026

Qwen3-vl-4b-thinking + GRPO + geo3k + verl会训崩么

Status: Open.

#1980 In QwenLM/Qwen3-VL;

· xuhang-2 opened on Jan 7, 2026

Qwen3-VL如何在vllm offline推理时设置视频帧数？

Status: Open.

#1979 In QwenLM/Qwen3-VL;

· tiesanguaixia opened on Jan 6, 2026

qwen3-vl processor gives inconsistent vision features

Status: Open.

#1978 In QwenLM/Qwen3-VL;

· kindernerd opened on Jan 6, 2026

Suggestion: Rebuild the official docker image with vllm >=0.11.2

Status: Open.

#1976 In QwenLM/Qwen3-VL;

· JIANSUJIAN opened on Jan 5, 2026

box作为Qwen3-VL的prompt输入时，应该使用归一化1000的相对坐标，还是resize后的绝对坐标？

Status: Open.

#1975 In QwenLM/Qwen3-VL;

· user00644 opened on Jan 5, 2026

think with image调用image_zoom_in_tool不会自动往下走到answer，只走到工具调用

Status: Open.

#1974 In QwenLM/Qwen3-VL;

· MarStarck opened on Jan 5, 2026

Question about inconsistent "vocab_size" settings in Qwen2.5-VL 3B and 7B models

Status: Open.

#1973 In QwenLM/Qwen3-VL;

· Hou9612 opened on Jan 4, 2026

LazySupervisedDataset似乎不是真的lazy？

Status: Open.

#1970 In QwenLM/Qwen3-VL;

· g2bond opened on Jan 4, 2026