Daniël de Kok 77ac0f364b Add support for Marlin-quantized models ... This change adds support for Marlin-quantized models. Marlin is an FP16xINT4 matmul kernel, which provides good speedups decoding batches of 16-32 tokens. It supports quantized models with symmetric quantization, groupsize -1 or 128, and 4-bit. Tested with: - Llama 2 - Llama 3 - Phi 3		2024-09-24 03:38:05 +00:00
..
__init__.py	Aligin the source code with main branch 2.0.4	2024-09-24 03:06:55 +00:00
convert.py	Force weights_only (before fully breaking pickle files anyway). (#1710)	2024-04-25 15:10:53 +03:00
dist.py	Aligin the source code with main branch 2.0.4	2024-09-24 03:06:55 +00:00
hub.py	Fixing the download strategy for ibm-fms (#1917)	2024-07-17 05:36:58 +00:00
import_utils.py	Purely refactors paged/attention into layers/attention and make hardware differences more obvious with 1 file per hardware. (#1986)	2024-09-24 03:19:39 +00:00
log.py	v1.3.4	2024-04-22 09:08:34 +03:00
logits_process.py	Aligin the source code with main branch 2.0.4	2024-09-24 03:06:55 +00:00
peft.py	fix: fix local loading for .bin models (#1419)	2024-04-22 09:17:52 +03:00
speculate.py	chore: formatting	2024-04-18 16:26:00 +03:00
tokens.py	Aligin the source code with main branch 2.0.4	2024-09-24 03:06:55 +00:00
watermark.py	Aligin the source code with main branch 2.0.4	2024-09-24 03:06:55 +00:00
weights.py	Add support for Marlin-quantized models	2024-09-24 03:38:05 +00:00