mahaocong90

Results 3 comments of mahaocong90

[Feat] Support FlashMLA backend with MTP and FP8 KV cache

Hi @quinnrong94, I have a question, Can FlashMLA be used with mtp 3-1-4? I test on H20 141g, and the benchmark reported a memory leak at the end： Scheduler hit...

[Feat] Support FlashMLA backend with MTP and FP8 KV cache

> > Hi @quinnrong94, I have a question, Can FlashMLA be used with mtp 3-1-4? I test on H20 141g, and the benchmark reported a memory leak at the end：...

【Question】What is the minimum number of GPUs required to train deepseek 671B with GRPO? How about using LoRA?

Same question, and then if I set --max_length from 256 to 128, and --batch_size from 24 to 12, does this reduce fine-tuning memory consumption？