Have someone loaded by
AutoModelForCausalLM.from_pretrained(MODEL_PATH, torch_dtype=torch.float16, device_map="auto)
I have tried load starcoder model with A100-40G with approximately 140s to load
but it took me 600+ s to load wizardcoder
Can someone help ?
Is this matter? : starcoder is separated in 7 pieces while wizardcode is a large bin file
Have someone loaded by
AutoModelForCausalLM.from_pretrained(MODEL_PATH, torch_dtype=torch.float16, device_map="auto)
I have tried load starcoder model with A100-40G with approximately 140s to load but it took me 600+ s to load wizardcoder Can someone help ? Is this matter? : starcoder is separated in 7 pieces while wizardcode is a large bin file