mmarch/text-generation-webui
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Add --num_experts_per_token parameter (ExLlamav2) (#4955)

Browse source
This commit is contained in:
oobabooga 2023-12-17 12:08:33 -03:00 • committed by GitHub
parent 12690d3ffc
commit f1f2c4c3f4
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
7 changed files with 28 additions and 20 deletions
Download patch file Download diff file Expand all files Collapse all files

1
modules/shared.py
View file

@ -125,6 +125,7 @@ parser.add_argument('--max_seq_len', type=int, default=2048, help='Maximum seque
parser.add_argument('--cfg-cache', action='store_true', help='ExLlama_HF: Create an additional cache for CFG negative prompts. Necessary to use CFG with that loader, but not necessary for CFG with base ExLlama.')
parser.add_argument('--no_flash_attn', action='store_true', help='Force flash-attention to not be used.')
parser.add_argument('--cache_8bit', action='store_true', help='Use 8-bit cache to save VRAM.')
parser.add_argument('--num_experts_per_token', type=int, default=2, help='Number of experts to use for generation. Applies to MoE models like Mixtral.')
# AutoGPTQ
parser.add_argument('--triton', action='store_true', help='Use triton.')