fix windows (#2733)

Author: Jintao-Huang

docs/source/Instruction/命令行参数.md

@@ -365,7 +365,7 @@ RLHF参数继承于[训练参数](#训练参数)
 - 🔥output_dir: 导出结果存储路径，默认为None
 
 - 🔥quant_method: 可选为'gptq', 'awq'，默认为None
-- quant_n_samples: gptq/awq的校验集抽样数，默认为256
+- quant_n_samples: gptq/awq的校验集抽样数，默认为128
 - max_length: 校准集的max_length, 默认值2048
 - quant_batch_size: 量化batch_size，默认为1
 - group_size: 量化group大小，默认为128

docs/source_en/Instruction/Command-line-parameters.md

@@ -367,7 +367,7 @@ Export Arguments include the [basic arguments](#base-arguments) and [merge argum
 - 🔥output_dir: Path for storing export results, default is None.
 
 - 🔥quant_method: Options are 'gptq' and 'awq', default is None.
-- quant_n_samples: Sampling size for the validation set in gptq/awq, default is 256.
+- quant_n_samples: Sampling size for the validation set in gptq/awq, default is 128.
 - max_length: Max length for the calibration set, default value is 2048.
 - quant_batch_size: Quantization batch size, default is 1.
 - group_size: Group size for quantization, default is 128.

examples/export/quantize/awq.sh

@@ -3,7 +3,7 @@ swift export \
     --model Qwen/Qwen2.5-1.5B-Instruct \
     --dataset AI-ModelScope/alpaca-gpt4-data-zh#500 \
               AI-ModelScope/alpaca-gpt4-data-en#500 \
-    --quant_n_samples 256 \
+    --quant_n_samples 128 \
     --quant_batch_size 1 \
     --max_length 2048 \
     --quant_method awq \

examples/export/quantize/gptq.sh

@@ -5,7 +5,7 @@ swift export \
     --model Qwen/Qwen2.5-1.5B-Instruct \
     --dataset AI-ModelScope/alpaca-gpt4-data-zh#500 \
               AI-ModelScope/alpaca-gpt4-data-en#500 \
-    --quant_n_samples 256 \
+    --quant_n_samples 128 \
     --quant_batch_size 1 \
     --max_length 2048 \
     --quant_method gptq \

swift/llm/argument/export_args.py

@@ -33,7 +33,7 @@ class ExportArguments(MergeArguments, BaseArguments):
 
     # awq/gptq
     quant_method: Literal['awq', 'gptq', 'bnb'] = None
-    quant_n_samples: int = 256
+    quant_n_samples: int = 128
     max_length: int = 2048
     quant_batch_size: int = 1
     group_size: int = 128

swift/llm/model/register.py

@@ -1,5 +1,6 @@
 # Copyright (c) Alibaba, Inc. and its affiliates.
 import os
+import platform
 import re
 from copy import deepcopy
 from dataclasses import asdict, dataclass, field
@@ -333,9 +334,11 @@ def get_model_name(model_id_or_path: str) -> Optional[str]:
     model_id_or_path = model_id_or_path.rstrip('/')
     match_ = re.search('/models--.+?--(.+?)/snapshots/', model_id_or_path)
     if match_ is not None:
-        model_name = match_.group(1)
-    else:
-        model_name = model_id_or_path.rsplit('/', 1)[-1]
+        return match_.group(1)
+
+    model_name = model_id_or_path.rsplit('/', 1)[-1]
+    if platform.system().lower() == 'windows':
+        model_name = model_name.rsplit('\\', 1)[-1]
     # compat modelscope snapshot_download
     model_name = model_name.replace('___', '.')
     return model_name

tests/export/quant.py

@@ -24,6 +24,16 @@ def test_vlm_quant(quant_method: Literal['gptq', 'awq'] = 'awq'):
             quant_method=quant_method))
 
 
+def test_audio_quant(quant_method: Literal['gptq', 'awq'] = 'awq'):
+    from swift.llm import export_main, ExportArguments
+    export_main(
+        ExportArguments(
+            model='Qwen/Qwen2-Audio-7B-Instruct',
+            quant_bits=4,
+            dataset=['speech_asr/speech_asr_aishell1_trainsets:validation#1000'],
+            quant_method=quant_method))
+
+
 def test_vlm_bnb_quant():
     from swift.llm import export_main, ExportArguments, infer_main, InferArguments
     export_main(ExportArguments(model='Qwen/Qwen2-VL-7B-Instruct', quant_bits=4, quant_method='bnb'))
@@ -34,4 +44,5 @@ def test_vlm_bnb_quant():
 if __name__ == '__main__':
     # test_llm_quant('gptq')
     # test_vlm_quant('gptq')
-    test_vlm_bnb_quant()
+    test_audio_quant('gptq')
+    # test_vlm_bnb_quant()