huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-09-12 04:44:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix: avoid unneeded quantize check

Browse Source
This commit is contained in:
drbh 2024-07-25 15:05:26 +00:00
parent a10f4010d7
commit 6784d5d0a6
1 changed files with 0 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
server/text_generation_server/layers/tensor_parallel.py
View File

@ -2,7 +2,6 @@ import torch
from torch.nn import functional as F from torch.nn import functional as F
from typing import Iterable, List from typing import Iterable, List
from text_generation_server.layers.linear import get_linear, FastLinear from text_generation_server.layers.linear import get_linear, FastLinear
from text_generation_server.layers.exl2 import Exl2Weight
from text_generation_server.utils.import_utils import SYSTEM from text_generation_server.utils.import_utils import SYSTEM
if SYSTEM == "ipex": if SYSTEM == "ipex":
@ -67,16 +66,6 @@ class TensorParallelHead(SuperLayer):
weight = weights.get_tensor(f"{prefix}.weight") weight = weights.get_tensor(f"{prefix}.weight")
should_gather = False should_gather = False
# GPTQ,AWQ,EETQ don't quantize heads (nor embeddings)
if config.quantize in ["gptq", "awq", "eetq", "marlin"]:
# Local variable `quantize` is assigned to but never used
quantize = None # noqa F841
# See above, exl2 LM head can be quantized or not.
elif config.quantize == "exl2" and not isinstance(weight, Exl2Weight):
quantize = None # noqa F841
else:
quantize = config.quantize # noqa F841
return TensorParallelHead( return TensorParallelHead(
get_linear(weight, bias=None), get_linear(weight, bias=None),
process_group=weights.process_group, process_group=weights.process_group,