trace good first issues (Python)

Independently reproduce the released 3B TRACE validation result 3 months ago

help wanted reproduction

2

Trace: A Taxonomy-Guided Environment for Multidomain Visual Reasoning

Python

#benchmark#dataset-generation#grpo#multimodal#multimodal-reasoning#post-training#procedural-generation#qwen2-5-vl#reinforcement-learning#rlvr#synthetic-data#verifiable-rewards#vision-language-models#visual-question-answering#visual-reasoning#vlm-training

Validate a memory-efficient 4-bit inference path for the 3B checkpoint 3 months ago

enhancement help wanted

maveryn/trace

2

Trace: A Taxonomy-Guided Environment for Multidomain Visual Reasoning

Python

#benchmark#dataset-generation#grpo#multimodal#multimodal-reasoning#post-training#procedural-generation#qwen2-5-vl#reinforcement-learning#rlvr#synthetic-data#verifiable-rewards#vision-language-models#visual-question-answering#visual-reasoning#vlm-training

Add a minimal veRL GRPO integration example 3 months ago

enhancement help wanted integration

maveryn/trace

2

Trace: A Taxonomy-Guided Environment for Multidomain Visual Reasoning

Python

#benchmark#dataset-generation#grpo#multimodal#multimodal-reasoning#post-training#procedural-generation#qwen2-5-vl#reinforcement-learning#rlvr#synthetic-data#verifiable-rewards#vision-language-models#visual-question-answering#visual-reasoning#vlm-training

Add a minimal TRL GRPO integration example 3 months ago

enhancement help wanted integration

maveryn/trace

2

Trace: A Taxonomy-Guided Environment for Multidomain Visual Reasoning

Python

#benchmark#dataset-generation#grpo#multimodal#multimodal-reasoning#post-training#procedural-generation#qwen2-5-vl#reinforcement-learning#rlvr#synthetic-data#verifiable-rewards#vision-language-models#visual-question-answering#visual-reasoning#vlm-training

maveryn/trace

Open Issues Need Help