alpaca_lora_4bit/autograd_4bit
Andrey Glushenkov f20570343f
GPTQv2 support
GPTQv2 support.
1. Adds dependency on `triton`
2. Refactors autograd_4bit to include both GPTQv1 and GPTQv2
3. Introduces new environment variable GPTQ_VERSION to select autograd_4bit version
4. Fixes triton kernels
5. Matrix multiplications are in fp16
2023-04-06 02:29:36 +03:00
..
__init__.py GPTQv2 support 2023-04-06 02:29:36 +03:00
autograd_4bit_v1.py GPTQv2 support 2023-04-06 02:29:36 +03:00
autograd_4bit_v2.py GPTQv2 support 2023-04-06 02:29:36 +03:00