Update requirements.txt

Update requirements.txt

Update pipeline.py

Update pipeline.py

Test

Load in 8bit

Add back load_in_4bit

Remove triton

Test revision pr/47

Test max_memory for GPU

Remove dtype=torch.bfloat16

Test

Removed device_map="auto"

Init load model on GPU

Change log in pipeline

Update custom pipeline do not load in 4-bit precision

Update requirements.txt

Update requirements.txt

Update requirements.txt

upgrade accelerate in custom handler

use static path oleksandrfluxon/mpt-7b-chat-4bit in custom handler

use PreTrainedPipeline in custom handler

add custom handler

Duplicate from mosaicml/mpt-7b-chat