biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Misc] Remove qlora_adapter_name_or_path (#17699)

Browse Source 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
This commit is contained in:
Jee Jee Li
2025-05-07 14:10:37 +08:00
committed by GitHub
parent f80ae5bdcf
commit ba7703e659
3 changed files with 50 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
vllm/engine/arg_utils.py
View File

@@ -6,6 +6,7 @@ import dataclasses
import json
import re
import threading
import warnings
from dataclasses import MISSING, dataclass, fields
from itertools import permutations
from typing import (Any, Callable, Dict, List, Literal, Optional, Type,
@@ -394,7 +395,13 @@ class EngineArgs:
if isinstance(self.compilation_config, (int, dict)):
self.compilation_config = CompilationConfig.from_cli(
str(self.compilation_config))
if self.qlora_adapter_name_or_path is not None:
warnings.warn(
"The `qlora_adapter_name_or_path` is deprecated "
"and will be removed in v0.10.0. ",
DeprecationWarning,
stacklevel=2,
)
# Setup plugins
from vllm.plugins import load_general_plugins
load_general_plugins()
@@ -504,10 +511,14 @@ class EngineArgs:
**load_kwargs["ignore_patterns"])
load_group.add_argument("--use-tqdm-on-load",
**load_kwargs["use_tqdm_on_load"])
load_group.add_argument('--qlora-adapter-name-or-path',
type=str,
default=None,
help='Name or path of the QLoRA adapter.')
load_group.add_argument(
"--qlora-adapter-name-or-path",
type=str,
default=None,
help="The `--qlora-adapter-name-or-path` has no effect, do not set"
" it, and it will be removed in v0.10.0.",
deprecated=True,
)
load_group.add_argument('--pt-load-map-location',
**load_kwargs["pt_load_map_location"])
@@ -534,7 +545,7 @@ class EngineArgs:
deprecated=True,
help="[DEPRECATED] The `--enable-reasoning` flag is deprecated as "
"of v0.8.6. Use `--reasoning-parser` to specify the reasoning "
"parser backend insteadThis flag (`--enable-reasoning`) will be "
"parser backend instead. This flag (`--enable-reasoning`) will be "
"removed in v0.10.0. When `--reasoning-parser` is specified, "
"reasoning mode is automatically enabled.")
guided_decoding_group.add_argument(
@@ -896,12 +907,6 @@ class EngineArgs:
def create_load_config(self) -> LoadConfig:
if(self.qlora_adapter_name_or_path is not None) and \
self.quantization != "bitsandbytes":
raise ValueError(
"QLoRA adapter only support "
f"'bitsandbytes' quantization, but got {self.quantization}")
if self.quantization == "bitsandbytes":
self.load_format = "bitsandbytes"
@@ -1098,11 +1103,6 @@ class EngineArgs:
max_cpu_loras=self.max_cpu_loras if self.max_cpu_loras
and self.max_cpu_loras > 0 else None) if self.enable_lora else None
if self.qlora_adapter_name_or_path is not None and \
self.qlora_adapter_name_or_path != "":
self.model_loader_extra_config[
"qlora_adapter_name_or_path"] = self.qlora_adapter_name_or_path
# bitsandbytes pre-quantized model need a specific model loader
if model_config.quantization == "bitsandbytes":
self.quantization = self.load_format = "bitsandbytes"