Question: Whether to cancel this evaluation ... (float32 kevinpro/Hydra-LLaMA3-8B-0513-preview)
#13
by
CombinHorizon
- opened
https://huggingface.co/datasets/OALL/requests/blob/main/kevinpro/Hydra-LLaMA3-8B-0513-preview_eval_request_False_float32_Original.json
has been running since 2024-05may-31 (more than a month)
- this is a float32 run, which takes, longer, and float16 and bfloat32 is also running, is this a test or something? (was it intentional?)
- https://huggingface.co/kevinpro/Hydra-LLaMA3-8B-0513-preview was deleted, only the gguf version is still available online
around the same time frame this was submitted, (1 day before), these other two models are now also stuck running ...
- https://huggingface.co/datasets/OALL/requests/blob/main/cloudyu/Yi-34Bx2-MoE-60B_eval_request_False_bfloat16_Original.json (2024-05may-30)
- https://huggingface.co/datasets/OALL/requests/blob/main/chujiezheng/Smaug-34B-v0.1-ExPO_eval_request_False_bfloat16_Original.json (2024-05may-30)
Is this float32 model holding and delaying these models? It usually takes much faster than this. (are they using the same cluster, given the timing?)
similar models (cloudyu/Yi-34Bx2-MoE-60B-DPO, abacusai/Smaug-34B-v0.1) finished successfully, so perhaps they have been cut/held short, before they could have finished?
attempting to re-run cloudyu/Yi-34Bx2-MoE-60B to see if doing that un-stucks it from the float32 run ...
addressed.
CombinHorizon
changed discussion status to
closed