huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-09-10 20:04:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

loading quantization config the proper way

Browse Source
This commit is contained in:
Chris 2023-08-27 16:48:30 +02:00
parent cf178a278a
commit 694a535033
1 changed files with 11 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
server/text_generation_server/models/causal_lm.py
View File

@ -3,7 +3,7 @@ import inspect
from dataclasses import dataclass from dataclasses import dataclass
from opentelemetry import trace from opentelemetry import trace
from transformers import AutoConfig, AutoTokenizer, AutoModelForCausalLM, PreTrainedTokenizerBase from transformers import AutoConfig, AutoTokenizer, AutoModelForCausalLM, PreTrainedTokenizerBase, BitsAndBytesConfig
from typing import Optional, Tuple, List, Type, Dict from typing import Optional, Tuple, List, Type, Dict
from text_generation_server.models import Model from text_generation_server.models import Model
@ -474,14 +474,20 @@ class CausalLM(Model):
truncation_side="left", truncation_side="left",
trust_remote_code=trust_remote_code, trust_remote_code=trust_remote_code,
) )
should_quantize = quantize == "bitsandbytes"
if(should_quantize):
quantization_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_compute_dtype=torch.float16
)
model = AutoModelForCausalLM.from_pretrained( model = AutoModelForCausalLM.from_pretrained(
model_id, model_id,
revision=revision, revision=revision,
torch_dtype=dtype, torch_dtype=dtype,
device_map="auto" device_map="auto" if torch.cuda.is_available() and torch.cuda.device_count() > 1 else None,
if torch.cuda.is_available() and torch.cuda.device_count() > 1 load_in_4bit= False if not should_quantize else None,
else None, quantization_config = quantization_config if should_quantize else None,
load_in_4bit=quantize == "bitsandbytes",
trust_remote_code=trust_remote_code, trust_remote_code=trust_remote_code,
) )
## ValueError: Calling `cuda()` is not supported for `4-bit` or `8-bit` quantized models. Please use the model as it is, since the model has already been set to the correct devices and casted to the correct `dtype`. ## ValueError: Calling `cuda()` is not supported for `4-bit` or `8-bit` quantized models. Please use the model as it is, since the model has already been set to the correct devices and casted to the correct `dtype`.