Mixtral-8x7B-Instruct-v0_1-AQLM-2Bit-1x16-hf / configuration_mixtral_aqlm.py
BlackSamorez's picture
Upload MixtralForCausalLM
6bfb114 verified
raw
history blame
427 Bytes
from transformers import MixtralConfig as OrigLlamaConfig
class MixtralConfig(OrigLlamaConfig):
model_type = "mixtral_aqlm"
def __init__(
self,
aqlm: dict[str, int] = {
"nbits_per_codebook": 16,
"num_codebooks": 1,
"out_group_size": 8,
"in_group_size": 1,
},
**kwargs,
):
super().__init__(**kwargs)
self.aqlm = aqlm