ffgcc

ITM threshold

3

HI, I am very fortunate to read your paper BLIP. It's very exciting. I wonder how to set the ITM threshold when filtering? Thanks! Looking forward to your reply.

segmentation layer

1

Hi, Why use the sixth layer as the segmentation, has the author tried to use other layers as the segmentation? such as 8,4? Thanks!

Questions about the paper.

8

Hello, I was very fortunate to read your paper, and the experimental results are exciting. The paper mentions two observations: Observation 1: Original BERT layers fail to improve the performance....

Hi， The experimental results on the LSDMC dataset in the paper reach a very high level, even higher than the models initialized with CLIP, such as x-pool, CenterCLIP, etc. However,...

关于DeepSeek-Coder-V2-Lite-Base的128k捞针测试结果

1

您好， 我测试了deepseek-ai/DeepSeek-Coder-V2-Lite-Base 在128k捞针任务上的表现，结果的正确率不足50%。并且受限于硬件，我无法在deepseek-ai/DeepSeek-Coder-V2-Base 上进行1k到128k的捞针测试。 不知是DeepSeek-Coder-V2-Lite-Base的捞针任务表现一般还是我的测试代码有问题，您可否提供捞针测试代码以便于我重新测试？ 感谢！

关于Llama-3.1-8B-Instruct在Longbench v2 测试结果和排行榜有出入的问题

5

您好， 我测试的Llama-3.1-8B-Instruct 结果如下： Model Overall Easy Hard Short Medium Long Llama-3.1-8B-Instruct 29.0 30.7 28.0 33.9 25.6 27.8 和排行榜中的Overall 有一个点的差距（29.0 vs 30.0），我的环境如下： vllm==0.5.3.post1 transformers==4.45.0 请问测试Llama-3.1-8B-Instruct 还需要什么特殊处理吗