Mingtian(Benny) Tan

Results 7 comments of Mingtian(Benny) Tan

Your result is impressive.

@mahad852 unfortunately, not yet. I am waiting the official Repo, but they seem to have no plan to release code yet. You can try email them as well.

Sounds good! Thanks so much.

May I kindly ask about the progress of the JSON data file? @marcotet It seems that many people are interested in it!

> Hi [@BennyTMT](https://github.com/BennyTMT) and [@Atharvap14](https://github.com/Atharvap14), I have uploaded a JSON version of the benchmark here: [all_tasks.json](https://github.com/ServiceNow/context-is-key-forecasting/blob/main/results/all_tasks.json). It was created by converting a HuggingFace dataset to JSON, so one way to...

I ran into the same issue. With the same dataset and almost identical configurations, training **Qwen2.5-VL-7B-Instruct** takes only 9 hours, while **Qwen3-VL-8B-Instruct** takes **35 hours!**