git
ask
hub
Privacy
Terms
Sign in with GitHub
chloeqxq/macd
↗
Stars ·
3
Language ·
Python
Simplify and Visualize This Repo
Scan the safety of this repo
Self Host this repo
Check out more work of Developer: chloeqxq
Ask anything about this repo to start.
I want to test MACD on a Qwen2.5-VL checkpoint with a yes/no video QA benchmark. Walk me through the required JSON question format and the shell script command to run the full pipeline.
Explain how MACD's counterfactual masking works: how does YOLO decide which objects to mask, and how is the masking strength tuned automatically per video clip?
I'm seeing hallucinations in my video question-answering model on multiple-choice questions. How do I apply MACD's contrastive decoding to compare original vs masked outputs and suppress the pattern-matching answers?
What are the exact Python version and transformers library version requirements to run MACD, and what breaks if I use a newer version of transformers?
How does MACD compute its final score by comparing model output on the original video vs the object-masked counterfactual? Walk me through the math behind the contrastive step.
Full explanation on explaingit →
📎
Send
By chatting or signing in you agree to the
Terms
and chat-message logging (revocable in
History
).