rixty/KoboldAI-Client
1
1
Fork 0
mirror of https://github.com/KoboldAI/KoboldAI-Client.git synced 2025-06-05 21:59:24 +02:00
Code Issues Projects Releases Wiki Activity

Fix 4-bit mpt

Browse Source
This commit is contained in:
0cc4m
2023-05-06 21:58:23 +02:00
parent a9fa199c49
commit 9ec50c9972
1 changed files with 3 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
modeling/inference_models/hf_torch_4bit.py
View File

@@ -34,6 +34,7 @@ from gptq.gptj import load_quant as gptj_load_quant
from gptq.gptneox import load_quant as gptneox_load_quant from gptq.gptneox import load_quant as gptneox_load_quant
from gptq.llama import load_quant as llama_load_quant from gptq.llama import load_quant as llama_load_quant
from gptq.opt import load_quant as opt_load_quant from gptq.opt import load_quant as opt_load_quant
from gptq.mpt import load_quant as mpt_load_quant
from gptq.offload import load_quant_offload from gptq.offload import load_quant_offload
@@ -369,6 +370,8 @@ class HFTorch4BitInferenceModel(HFTorchInferenceModel):
model = load_quant_offload(llama_load_quant, utils.koboldai_vars.custmodpth, path_4bit, 4, groupsize, self.gpu_layers_list) model = load_quant_offload(llama_load_quant, utils.koboldai_vars.custmodpth, path_4bit, 4, groupsize, self.gpu_layers_list)
elif utils.koboldai_vars.model_type == "opt": elif utils.koboldai_vars.model_type == "opt":
model = load_quant_offload(opt_load_quant, utils.koboldai_vars.custmodpth, path_4bit, 4, groupsize, self.gpu_layers_list) model = load_quant_offload(opt_load_quant, utils.koboldai_vars.custmodpth, path_4bit, 4, groupsize, self.gpu_layers_list)
elif utils.koboldai_vars.model_type == "mpt":
model = load_quant_offload(mpt_load_quant, utils.koboldai_vars.custmodpth, path_4bit, 4, groupsize, self.gpu_layers_list)
else: else:
raise RuntimeError(f"4-bit load failed. Model type {utils.koboldai_vars.model_type} not supported in 4-bit") raise RuntimeError(f"4-bit load failed. Model type {utils.koboldai_vars.model_type} not supported in 4-bit")