Zhiqing Sun

Hi, Thank you for your interest in both projects. The essential differences between this implementation and GraphVite implementation are listed as following: 1. GraphVite uses ASGD rather than mini-batch SGD，the...

Hi @qianyizhang, For FPN in TSP-RCNN, we use the RPN's output (RoI) to select top proposals. For FPN in TSP-FCOS, we have a Feature of Interest (FoI) score. As for...

Hi, Our script should be able to fully reproduce the trained model. [TSP-Detection#rreproducing-results](https://github.com/Edward-Sun/TSP-Detection#rreproducing-results)

Hi, For now, please try the following code to replace the line of `meta_prompt = meta_prompts[int(example["example_id"]) % total_meta_prompt]`. We will add a commit to solve the issue soon. ``` example_id...

Hi Harryis, In our codebase, "example_id" only affects which prompt template to use, so it won't affect too much on the performance. Also, if you inspect the data, you would...

Yes, when we developed this project around March/April, the faster inference techniques (e.g., TGI and vLLM) of llama had not been developed, so we tried our best to use a...

Hi Harryis, Yes. We generated around 120k synthetic topics (after filtering on the topics) from TGRT Self-Instruct, generated the corresponding 120k prompts, and did some filtering on the prompts to...

Hi Harryis, We generate the topics in several rounds (called `generation_epoch` in the [code](https://github.com/IBM/Dromedary/blob/927bf5bdaa2b2fe951e014474692d89813e95b00/training/step1_topic_guided_red_teaming_self_instruct/scripts/tgrt_topic_generate_65b_base.sh#L40)), where in each round, we sample all topics from the previous rounds as the seed to...

Hi, it's unlike the problem of the difference between 7b and 70b. What is the version of the PEFT you used? We use `peft==4.0.0` in our experiments, and perhaps the...