[REMINDER] Use Ray for vLLM, and clean up your repo

Do you clone whole repo or particular branch/tag? Can we keep different models in different branches?

View post 0

How to download for huggingface LLM pretrained model

One more question please. Sorry, maybe I missed this. Is the 25GB limit still valid? Or after moving to http://gitlab.aicrowd.com we can use more?

View post 0

How to download for huggingface LLM pretrained model

Hi @yilun_jin,

Just to clarify, we can not download models using Dockerfile either, right?

View post 0

Failed tag with successful evaluation

My submission was scored correctly, but in the end tagged as “failed”.

image2028×670 88.9 KB

@aicrowd-bot

View post 0

Wrong samples in Track 2

It’s here.

View post 0

Wrong samples in Track 2

I successfully submitted to Track 2, but I noticed that the progress bar not finished.

image1506×270 29.1 KB

I also noticed Retrieval Score in the issue logs, but I can’t see this score in the leaderboard.

View post 0

Submission with 500 Server Error

I tried to resubmit it, and I get timeout issue. @yilun_jin could you take a look please?

View post 0

Submission with 500 Server Error

I’m not sure if we can submit between phases, but I catch this exception right after I pushed the tag:

Submission failed : 500 Server Error: INTERNAL SERVER ERROR for url: https://aicrowd-api.aws-internal.k8s.aicrowd.com/challenges/?slug=amazon-kdd-cup-24-shopping-knowledge-reasoning

@yilun_jin Could you please rerun it somehow?

View post 0

Not qualified submission

I still can’t submit to Track 3.

But my teammate can.

I think it’s happened because I never submitted to this track during Phase 1.

Could you please grant me some access?

View post 0

Not qualified submission

It’s here.

View post 0

Not qualified submission

Hi, I’m from the team GPU_IS_ALL_YOU_NEED and AFAIK we are qualified for Track 3 in Phase 2. But when I submitted, I’ve got this:

Submission failed : You have not qualified for this round. Please review the challenge rules at www.aicrowd.com

View post 4

Optimization opportunities in the Flatland environment

I highly recommend to replace np.isclose with simple equivalent function:

def my_isclose(x, y, rtol=1.e-5, atol=1.e-8):
    return abs(x - y) <= atol + rtol * abs(y)

Here is performance for RailEnv.step before:

912200 function calls (870670 primitive calls) in 1.818 seconds

After

492968 function calls (477640 primitive calls) in 1.161 seconds

View post 1