You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
when using LoRA to fine-tune, the original model parameters are locked, which can be converted to INT8 mode to inference, and other trainable parameters with fp16/bf16/fp32/tf32 mode. It is just like what peft library do.
Uing INT8 to run original model will save GPU memory and accelerate training speed.
The text was updated successfully, but these errors were encountered:
when using LoRA to fine-tune, the original model parameters are locked, which can be converted to INT8 mode to inference, and other trainable parameters with fp16/bf16/fp32/tf32 mode. It is just like what peft library do.
Uing INT8 to run original model will save GPU memory and accelerate training speed.
The text was updated successfully, but these errors were encountered: