This is the final version, every subset appears here is verified by Gemini/GPT.
Given that I tried my best to improve the quality of the dataset, some models are just not in the same league of others in this context -- that's said I suggest you to "weight" the models when training.
Powerful models are: Gemini ~ Qwen-family > gemma-family ~ medical-based models (lingshu, huato, hulumed)
Normal models are: the rest (they are mostly general-purpose open-source VLMs)
Eventhough the data are verified by LLMs, there will be no 100%-guarantee that there is no hallucinations in the data (from Normal models as mentioned above). Please be picky.
- Downloads last month
- 21
Collection including VietMedTeam/V-RLHF-o-v5
Collection
[Dev versions] RLHF for Medical Visual CoT Monitoring
•
12 items
•
Updated