【feature】model_type支持transformers-MindStudio-ModelSlim-AtomGit

Mconfig/config.ini+2-0

	@@ -4,6 +4,7 @@ question_and_answer_url = https://gitcode.com/Ascend/msmodelslim/blob/master/doc
4		4
5	[ModelAdapter]	5	[ModelAdapter]
6	default = default	6	default = default
		7	+transformers = transformers
7	deepseek_v3 = DeepSeek-V3, DeepSeek-V3-0324, DeepSeek-R1, DeepSeek-R1-0528, DeepSeek-V3.1, DeepSeek-V3.1-Terminus	8	deepseek_v3 = DeepSeek-V3, DeepSeek-V3-0324, DeepSeek-R1, DeepSeek-R1-0528, DeepSeek-V3.1, DeepSeek-V3.1-Terminus
8	deepseek_v3_2 = DeepSeek-V3.2-Exp, DeepSeek-V3.2	9	deepseek_v3_2 = DeepSeek-V3.2-Exp, DeepSeek-V3.2
9	deepseek_v4 = DeepSeek-V4-Flash, DeepSeek-V4-Pro	10	deepseek_v4 = DeepSeek-V4-Flash, DeepSeek-V4-Pro
	@@ -51,6 +52,7 @@ mimo_v2_flash = MiMo-V2-Flash
51		52
52	[ModelAdapterEntryPoints]	53	[ModelAdapterEntryPoints]
53	default = msmodelslim.model.default.loader:DefaultAdapterLoader	54	default = msmodelslim.model.default.loader:DefaultAdapterLoader
		55	+transformers = msmodelslim.model.transformers.loader:TransformersAdapterLoader
54	deepseek_v3 = msmodelslim.model.deepseek_v3.loader:DeepseekV3AdapterLoader	56	deepseek_v3 = msmodelslim.model.deepseek_v3.loader:DeepseekV3AdapterLoader
55	deepseek_v3_2 = msmodelslim.model.deepseek_v3_2.loader:DeepseekV3_2AdapterLoader	57	deepseek_v3_2 = msmodelslim.model.deepseek_v3_2.loader:DeepseekV3_2AdapterLoader
56	deepseek_v4 = msmodelslim.model.deepseek_v4.loader:DeepseekV4AdapterLoader	58	deepseek_v4 = msmodelslim.model.deepseek_v4.loader:DeepseekV4AdapterLoader

Mdocs/zh/api_reference/cli/msmodelslim_quant.md+15-4

	@@ -25,7 +25,7 @@ msmodelslim quant [--model_type <model_type>] --model_path <model_path> --save_p
25		25
26	\| 参数 \| 别名 \| 类型 \| 传入形式 \| 必选/可选 \| 默认值 \| 取值范围或格式 \| 含义 \|	26	\| 参数 \| 别名 \| 类型 \| 传入形式 \| 必选/可选 \| 默认值 \| 取值范围或格式 \| 含义 \|
27	\|------\|------\|------\|----------\|-----------\|--------\|----------------\|------\|	27	\|------\|------\|------\|----------\|-----------\|--------\|----------------\|------\|
28	-\| `--model_type` \| 无 \| `string` \| 单值 \| 条件必选（普通量化路径必选；`--config` 指向 `apiversion: modelslim_convert` 的配置时可省略） \| 无 \| 模型类型名称，如 `Qwen2.5-7B-Instruct` \| 指定待量化模型类型，用于加载对应模型适配器并匹配最佳实践；仅当 `--config` 指向 `apiversion: modelslim_convert` 的配置时可省略。 \|	28	+\| `--model_type` \| 无 \| `string` \| 单值 \| 条件必选（普通量化路径必选；`--config` 指向 `apiversion: modelslim_convert` 的配置时可省略） \| 无 \| 模型类型名称，如 `Qwen2.5-7B-Instruct`、`transformers` \| 指定待量化模型类型，用于加载对应模型适配器并匹配最佳实践；`transformers`支持加载基于 transformers 的通用基础模型适配器，当前仅适用于大语言模型量化；仅当 `--config` 指向 `apiversion: modelslim_convert` 的配置时可省略。 \|
29	\| `--model_path` \| 无 \| `string` \| 单值 \| 必选 \| 无 \| 原始模型权重目录（需存在且可读） \| 待量化模型的权重目录。 \|	29	\| `--model_path` \| 无 \| `string` \| 单值 \| 必选 \| 无 \| 原始模型权重目录（需存在且可读） \| 待量化模型的权重目录。 \|
30	\| `--save_path` \| 无 \| `string` \| 单值 \| 必选 \| 无 \| 输出目录（需可写） \| 量化权重与描述文件的保存目录。 \|	30	\| `--save_path` \| 无 \| `string` \| 单值 \| 必选 \| 无 \| 输出目录（需可写） \| 量化权重与描述文件的保存目录。 \|
31	\| `--device` \| 无 \| `string` \| 单值 \| 可选 \| `npu` \| `npu`、`cpu` \| 运行设备类型；多卡索引请用 `--device_id` 指定。 \|	31	\| `--device` \| 无 \| `string` \| 单值 \| 可选 \| `npu` \| `npu`、`cpu` \| 运行设备类型；多卡索引请用 `--device_id` 指定。 \|
	@@ -46,6 +46,7 @@ msmodelslim quant [--model_type <model_type>] --model_path <model_path> --save_p
46	- 两者都不传时，按默认量化类型 `w8a8` 匹配最佳实践；未匹配到最佳实践时会给出提示并等待确认（输入 `y` 继续，否则退出）。	46	- 两者都不传时，按默认量化类型 `w8a8` 匹配最佳实践；未匹配到最佳实践时会给出提示并等待确认（输入 `y` 继续，否则退出）。
47	- 指定 `--config` 后直接采用该配置，`--quant_type` 与 `--tags` 的最佳实践匹配均被忽略。	47	- 指定 `--config` 后直接采用该配置，`--quant_type` 与 `--tags` 的最佳实践匹配均被忽略。
48	- `--model_type` 在普通量化路径下必须提供；仅当 `--config` 指向 `apiversion: modelslim_convert` 的配置时可省略。	48	- `--model_type` 在普通量化路径下必须提供；仅当 `--config` 指向 `apiversion: modelslim_convert` 的配置时可省略。
		49	+- `--model_type` 指定 `transformers` 时，若使用 `--quant_type`，仅支持 `w8a16`/`w8a8`；若使用 `--config`，当前仅支持大语言模型量化配置（`apiversion: modelslim_v1`）。
49	- `--tags` 指定多个值时须同时出现在同一已验证场景；未提供硬件类型标签时自动匹配当前设备类型。	50	- `--tags` 指定多个值时须同时出现在同一已验证场景；未提供硬件类型标签时自动匹配当前设备类型。
50	- `--debug` 启用后量化上下文写入 `${SAVE_PATH}/debug_info/`。	51	- `--debug` 启用后量化上下文写入 `${SAVE_PATH}/debug_info/`。
51		52
	@@ -81,7 +82,17 @@ msmodelslim quant \
81		82
82	`${MODEL_TYPE}` 为模型类型名称（如 `Qwen2.5-7B-Instruct`），`${MODEL_PATH}` 为浮点权重目录，`${SAVE_PATH}` 为量化输出目录。未指定 `--quant_type` 与 `--config` 时，默认按量化类型 `w8a8` 匹配最佳实践并执行量化；若模型加载需要模型目录内的自定义代码，再补充 `--trust_remote_code true`。	83	`${MODEL_TYPE}` 为模型类型名称（如 `Qwen2.5-7B-Instruct`），`${MODEL_PATH}` 为浮点权重目录，`${SAVE_PATH}` 为量化输出目录。未指定 `--quant_type` 与 `--config` 时，默认按量化类型 `w8a8` 匹配最佳实践并执行量化；若模型加载需要模型目录内的自定义代码，再补充 `--trust_remote_code true`。
83		84
84	-### 7.2 显式指定量化类型与设备	85	+### 7.2 使用 transformers 通用基础模型适配器量化（仅适用于大语言模型）
		C C code_mingming7 天前【review】序号未更新，和下方序号重复。需要更新下方及之后的序号 Z zhangz200102 7 天前评论： 7 天前评论：
		86	+
		87	+```bash
		88	+msmodelslim quant \
		89	+ --model_type transformers \
		90	+ --model_path "${MODEL_PATH}" \
		91	+ --save_path "${SAVE_PATH}" \
		92	+ --quant_type w8a16
		93	+```
		94	+
		95	+### 7.3 显式指定量化类型与设备
85		96
86	```bash	97	```bash
87	msmodelslim quant \	98	msmodelslim quant \
	@@ -95,7 +106,7 @@ msmodelslim quant \
95		106
96	`--quant_type w8a8c8` 表示权重8bit、激活8bit、KVCache 8bit 量化；`--device_id 0 1 2 3` 使用4个 NPU 设备（索引列表形式在 `apiversion: modelslim_v1` 配置下受支持）。	107	`--quant_type w8a8c8` 表示权重8bit、激活8bit、KVCache 8bit 量化；`--device_id 0 1 2 3` 使用4个 NPU 设备（索引列表形式在 `apiversion: modelslim_v1` 配置下受支持）。
97		108
98	-### 7.3 使用自定义配置文件	109	+### 7.4 使用自定义配置文件
99		110
100	```bash	111	```bash
101	msmodelslim quant \	112	msmodelslim quant \
	@@ -107,7 +118,7 @@ msmodelslim quant \
107		118
108	`${CONFIG_PATH}` 指向符合 V1 等协议的量化 YAML；指定后直接采用该配置，不再做最佳实践匹配。字段说明见引用的配置文档。	119	`${CONFIG_PATH}` 指向符合 V1 等协议的量化 YAML；指定后直接采用该配置，不再做最佳实践匹配。字段说明见引用的配置文档。
109		120
110	-### 7.4 使用场景标签匹配最佳实践	121	+### 7.5 使用场景标签匹配最佳实践
111		122
112	```bash	123	```bash
113	msmodelslim quant \	124	msmodelslim quant \

Mdocs/zh/user_guide/usage_weight_quantization.md+9-2

	@@ -103,10 +103,17 @@ msmodelslim quant \
103		103
104	操作：	104	操作：
105		105
106	-- 尚未接入的模型：须先完成适配器开发与注册，再进入步骤 3。	106	+- 尚未接入的模型：通常需要先自主接入模型，完成适配器开发与注册，再进入步骤 3。特例：对于基于 transformers 的大语言模型，使用 `--model_type transformers` 可匹配通用基础模型适配器，无需自行适配。
107	- 支持矩阵中已接入的模型：可跳过本步骤，直接进入步骤 3 编写量化配置；确认所用 `--model_type`（即 `${MODEL_TYPE}`）名称即可。	107	- 支持矩阵中已接入的模型：可跳过本步骤，直接进入步骤 3 编写量化配置；确认所用 `--model_type`（即 `${MODEL_TYPE}`）名称即可。
108		108
109	-尚未接入时按下列操作执行：	109	+`--model_type transformers` 使用限制：
		110	+
		111	+- 该模型适配器仅具备基础量化能力，未实现任何算法。
		112	+- 若使用 `--quant_type`，当前仅支持 `w8a16`、`w8a8(MXFP8)`；若使用 `--config`，仅支持大语言模型量化配置（`apiversion: modelslim_v1`）。
		113	+- 仅推荐用于一键量化，若用于敏感层分析、自动调优等其他功能，用户需自行确保相关接口行为符合预期。
		114	+- 仅支持 `runner: model_wise` ，因而显存要求需足以容纳完整模型。
		115	+
		116	+自主接入时按下列操作执行：
110		117
111	1. 按模型类别阅读对应接入指南，并按其要求完成适配器实现与注册（细节以接入指南为准）：	118	1. 按模型类别阅读对应接入指南，并按其要求完成适配器实现与注册（细节以接入指南为准）：
112		119

Alab_practice/llm_transformers/llm_transformers_w8a16.yaml+28-0

@@ -0,0 +1,28 @@
	1	+apiversion: modelslim_v1
	2	+
	3	+metadata:
	4	+ config_id: llm_transformers_w8a16
	5	+ score: 50
	6	+ verified_model_types:
	7	+ - transformers
	8	+ label:
	9	+ w_bit: 8
	10	+ a_bit: 16
	11	+ is_sparse: False
	12	+ kv_cache: False
	13	+
	14	+spec:
	15	+ runner: "model_wise"
	16	+ process:
	17	+ - type: "linear_quant"
	18	+ qconfig:
	19	+ weight:
	20	+ scope: "per_channel"
	21	+ dtype: "int8"
	22	+ symmetric: True
	23	+ method: "minmax"
	24	+ include: [ "*" ]
	25	+
	26	+ save:
	27	+ - type: "ascendv1_saver"
	28	+ part_file_size: 4

Alab_practice/llm_transformers/llm_transformers_w8a8_mxfp8.yaml+35-0

@@ -0,0 +1,35 @@
	1	+apiversion: modelslim_v1
	2	+
	3	+metadata:
	4	+ config_id: llm_transformers_w8a8_mxfp8
	5	+ score: 50
	6	+ verified_model_types:
	7	+ - transformers
	8	+ label:
	9	+ w_bit: 8
	10	+ a_bit: 8
	11	+ is_sparse: False
	12	+ kv_cache: False
	13	+
	14	+default_w8a8_dynamic: &default_w8a8_mxfp8_dynamic
	15	+ act:
	16	+ scope: "per_block"
	17	+ dtype: "mxfp8"
	18	+ symmetric: True
	19	+ method: "minmax"
	20	+ weight:
	21	+ scope: "per_block"
	22	+ dtype: "mxfp8"
	23	+ symmetric: True
	24	+ method: "minmax"
	25	+
	26	+spec:
	27	+ runner: "model_wise"
	28	+ process:
	29	+ - type: "linear_quant"
	30	+ qconfig: *default_w8a8_mxfp8_dynamic
	31	+ include: [ "*" ]
	32	+
	33	+ save:
	34	+ - type: "ascendv1_saver"
	35	+ part_file_size: 4

Mmsmodelslim/app/naive_quantization/application.py+15-0

	@@ -156,6 +156,20 @@ def validate_device_index(device_index: Optional[List[int]], device_type: Device
156	)	156	)
157		157
158		158
		159	+def check_model_type_transformers(model_adapter: IModel, config: PracticeConfig):
		160	+ model_type = model_adapter.get_model_type()
		161	+ if model_type == "transformers" and config.apiversion == "multimodal_vlm_modelslim_v1":
		162	+ raise UnsupportedError(
		163	+ "VLM quantization is not supported for model_type transformers",
		164	+ action="Please use a dedicated model adapter for multimodal models",
		165	+ )
		166	+ if model_type == "transformers" and config.apiversion == "multimodal_sd_modelslim_v1":
		167	+ raise UnsupportedError(
		168	+ "DiT quantization is not supported for model_type transformers",
		169	+ action="Please use a dedicated model adapter for multimodal models",
		170	+ )
		171	+
		172	+
159	@logger_setter('msmodelslim.app.naive_quantization')	173	@logger_setter('msmodelslim.app.naive_quantization')
160	class NaiveQuantizationApplication:	174	class NaiveQuantizationApplication:
161	def __init__(	175	def __init__(
	@@ -235,6 +249,7 @@ class NaiveQuantizationApplication:
235	if config_path is not None:	249	if config_path is not None:
236	config = PracticeConfig.model_validate(yaml_safe_load(str(config_path)))	250	config = PracticeConfig.model_validate(yaml_safe_load(str(config_path)))
237	get_logger().info("Naive Quant apply config_path: %s", config_path)	251	get_logger().info("Naive Quant apply config_path: %s", config_path)
		252	+ check_model_type_transformers(model_adapter, config)
238	return config	253	return config
239		254
240	if not isinstance(model_adapter, ModelInfoInterface):	255	if not isinstance(model_adapter, ModelInfoInterface):

Amsmodelslim/model/transformers/__init__.py+20-0

@@ -0,0 +1,20 @@
	1	+#!/usr/bin/env python
	2	+# -- coding: UTF-8 --
	3	+
	4	+"""
	5	+-------------------------------------------------------------------------
	6	+This file is part of the MindStudio project.
	7	+Copyright (c) 2026 Huawei Technologies Co.,Ltd.
	8	+
	9	+MindStudio is licensed under Mulan PSL v2.
	10	+You can use this software according to the terms and conditions of the Mulan PSL v2.
	11	+You may obtain a copy of Mulan PSL v2 at:
	12	+
	13	+ http://license.coscl.org.cn/MulanPSL2
	14	+
	15	+THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
	16	+EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
	17	+MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
	18	+See the Mulan PSL v2 for more details.
	19	+-------------------------------------------------------------------------
	20	+"""

Amsmodelslim/model/transformers/detect.py+96-0

@@ -0,0 +1,96 @@
	1	+#!/usr/bin/env python
	2	+# -- coding: UTF-8 --
	3	+
	4	+"""
	5	+-------------------------------------------------------------------------
	6	+This file is part of the MindStudio project.
	7	+Copyright (c) 2026 Huawei Technologies Co.,Ltd.
	8	+
	9	+MindStudio is licensed under Mulan PSL v2.
	10	+You can use this software according to the terms and conditions of the Mulan PSL v2.
	11	+You may obtain a copy of Mulan PSL v2 at:
	12	+
	13	+ http://license.coscl.org.cn/MulanPSL2
	14	+
	15	+THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
	16	+EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
	17	+MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
	18	+See the Mulan PSL v2 for more details.
	19	+-------------------------------------------------------------------------
	20	+"""
	21	+
	22	+from enum import Enum
	23	+from pathlib import Path
	24	+from typing import Any, Dict
	25	+
	26	+from msmodelslim.utils.exception import UnsupportedError, SecurityError
	27	+from msmodelslim.utils.security import json_safe_load
	28	+
	29	+
	30	+class ModelClass(str, Enum):
	31	+ LLM = "llm"
	32	+ VLM = "vlm"
	33	+ DIT = "dit"
	34	+
	35	+
	36	+LLM_ADAPTER_CLASS_PATH = "msmodelslim.model.transformers.model_adapter:LLMTransformersModel"
	37	+
	38	+_VLM_CONFIG_KEYS = ("vision_config",)
	39	+
	40	+_DIT_CONFIG_KEYS = ("_diffusers_version",)
	41	+
	42	+
	43	+def _is_vlm_config(config: Dict[str, Any]) -> bool:
	44	+ if any(key in config for key in _VLM_CONFIG_KEYS):
	45	+ return True
	46	+ return False
	47	+
	48	+
	49	+def _is_dit_config(config: Dict[str, Any]) -> bool:
	50	+ if any(key in config for key in _DIT_CONFIG_KEYS):
	51	+ return True
	52	+ return False
	53	+
	54	+
	55	+def detect_transformers_kind(model_path: Path) -> ModelClass:
	56	+ """Classify a HuggingFace / diffusers directory as llm, vlm, or dit."""
	57	+ config_path = model_path / "config.json"
	58	+ try:
	59	+ config = json_safe_load(str(config_path))
	60	+ except SecurityError:
	61	+ raise UnsupportedError(
	62	+ f"The path {config_path} doesn't exist or isn't a file. "
	63	+ "Only standard transformers-based LLM model is supported currently "
	64	+ "when using model_type transformers.",
	65	+ action=(f"Please check (1) files in {model_path} are complete. (2) the model is a LLM model."),
	66	+ )
	67	+
	68	+ if _is_vlm_config(config):
	69	+ return ModelClass.VLM
	70	+
	71	+ if _is_dit_config(config):
	72	+ return ModelClass.DIT
	73	+
	74	+ return ModelClass.LLM
	75	+
	76	+
	77	+def resolve_adapter_class_path(model_path: Path) -> str:
	78	+ """Return adapter class path for ``--model_type transformers``.
	79	+
	80	+ VLMTransformersModel is not registered yet; VLM/DiT raise until a dedicated
	81	+ generic adapter exists.
	82	+ """
	83	+
	84	+ kind = detect_transformers_kind(model_path)
	85	+
	86	+ if kind == ModelClass.VLM:
	87	+ raise UnsupportedError(
	88	+ "VLM quantization is not supported for model_type transformers",
	89	+ action="Please use a dedicated model adapter for multimodal models (VLM/DiT)",
	90	+ )
	91	+ if kind == ModelClass.DIT:
	92	+ raise UnsupportedError(
	93	+ "DiT quantization is not supported for model_type transformers",
	94	+ action="Please use a dedicated model adapter for multimodal models (VLM/DiT)",
	95	+ )
	96	+ return LLM_ADAPTER_CLASS_PATH

Amsmodelslim/model/transformers/loader.py+20-0

@@ -0,0 +1,20 @@
	1	+# -- coding: UTF-8 --
	2	+
	3	+from pathlib import Path
	4	+
	5	+from msmodelslim.model.plugin_factory.base_loader import BaseModelAdapterLoader
	6	+
	7	+from .detect import LLM_ADAPTER_CLASS_PATH, resolve_adapter_class_path
	8	+
	9	+
	10	+class TransformersAdapterLoader(BaseModelAdapterLoader):
	11	+ ADAPTER_CLASS_PATH = LLM_ADAPTER_CLASS_PATH
	12	+
	13	+ def load(
	14	+ self,
	15	+ model_type: str,
	16	+ model_path: Path,
	17	+ trust_remote_code: bool = False,
	18	+ ):
	19	+ self.ADAPTER_CLASS_PATH = resolve_adapter_class_path(model_path)
	20	+ return super().load(model_type, model_path, trust_remote_code)

Amsmodelslim/model/transformers/model_adapter.py+103-0

@@ -0,0 +1,103 @@
	1	+#!/usr/bin/env python
	2	+# -- coding: UTF-8 --
	3	+
	4	+"""
	5	+-------------------------------------------------------------------------
	6	+This file is part of the MindStudio project.
	7	+Copyright (c) 2026 Huawei Technologies Co.,Ltd.
	8	+
	9	+MindStudio is licensed under Mulan PSL v2.
	10	+You can use this software according to the terms and conditions of the Mulan PSL v2.
	11	+You may obtain a copy of Mulan PSL v2 at:
	12	+
	13	+ http://license.coscl.org.cn/MulanPSL2
	14	+
	15	+THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
	16	+EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
	17	+MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
	18	+See the Mulan PSL v2 for more details.
	19	+-------------------------------------------------------------------------
	20	+"""
	21	+
	22	+from typing import Any, Generator, List
	23	+
	24	+from torch import nn
	25	+
	26	+from msmodelslim.core.base.protocol import ProcessRequest
	27	+from msmodelslim.core.const import DeviceType
	28	+from msmodelslim.utils.logging import logger_setter
	29	+from msmodelslim.utils.exception import InvalidModelError
	30	+from msmodelslim.utils.exception_decorator import exception_handler
	31	+from ..common.layer_wise_forward import (
	32	+ generated_decoder_layer_visit_func,
	33	+ transformers_generated_forward_func,
	34	+)
	35	+from ..common.transformers import TransformersModel
	36	+from ..interface_hub import (
	37	+ ModelInfoInterface,
	38	+ ModelSlimPipelineInterfaceV1,
	39	+)
	40	+
	41	+
	42	+@logger_setter()
	43	+class LLMTransformersModel( # pylint: disable=too-many-ancestors
	44	+ TransformersModel,
	45	+ ModelInfoInterface,
	46	+ ModelSlimPipelineInterfaceV1,
	47	+):
	48	+ """Generic HuggingFace LLM adapter for ``--model_type transformers``.
	49	+
	50	+ Suitable for standard decoder-layer LLMs. MoE / VLM / custom structures
	51	+ still need dedicated adapters.
	52	+ """
	53	+
	54	+ def get_model_type(self) -> str:
	55	+ return self.model_type
	56	+
	57	+ def get_model_pedigree(self) -> str:
	58	+ return "llm_transformers"
	59	+
	60	+ def handle_dataset(self, dataset: Any, device: DeviceType = DeviceType.NPU) -> List[Any]:
	61	+ with exception_handler(
	62	+ 'You are handling dataset with llm transformers model adapter but failed',
	63	+ ms_err_cls=InvalidModelError,
	64	+ action='Please ensure llm transformers model adapter match your model',
	65	+ ):
	66	+ return self._get_tokenized_data(dataset, device)
	67	+
	68	+ def init_model(self, device: DeviceType = DeviceType.NPU) -> nn.Module:
	69	+ with exception_handler(
	70	+ 'You are initializing model with llm transformers model adapter but failed',
	71	+ ms_err_cls=InvalidModelError,
	72	+ action='Please ensure llm transformers model adapter match your model. '
	73	+ 'For multimodal model (VLM/DiT) quantization, please implement a dedicated model adapter',
	74	+ ):
	75	+ return self._load_model(device)
	76	+
	77	+ def generate_model_visit(self, model: nn.Module) -> Generator[ProcessRequest, Any, None]:
	78	+ with exception_handler(
	79	+ 'You are generating model visit with llm transformers model adapter but failed',
	80	+ ms_err_cls=InvalidModelError,
	81	+ action='Please ensure llm transformers model adapter match your model',
	82	+ ):
	83	+ yield from generated_decoder_layer_visit_func(model)
	84	+
	85	+ def generate_model_forward(
	86	+ self,
	87	+ model: nn.Module,
	88	+ inputs: Any,
	89	+ ) -> Generator[ProcessRequest, Any, None]:
	90	+ with exception_handler(
	91	+ 'You are generating model forward with llm transformers model adapter but failed',
	92	+ ms_err_cls=InvalidModelError,
	93	+ action='Please ensure llm transformers model adapter match your model',
	94	+ ):
	95	+ yield from transformers_generated_forward_func(model, inputs)
	96	+
	97	+ def enable_kv_cache(self, model: nn.Module, need_kv_cache: bool) -> None:
	98	+ with exception_handler(
	99	+ 'You are enabling kv cache with llm transformers model adapter but failed',
	100	+ ms_err_cls=InvalidModelError,
	101	+ action='Please ensure llm transformers model adapter match your model',
	102	+ ):
	103	+ return self._enable_kv_cache(model, need_kv_cache)

Mtest/cases/app/naive_quantization/test_naive_quantization_app.py+64-2

	@@ -24,13 +24,13 @@ import shutil
24	import tempfile	24	import tempfile
25	import unittest	25	import unittest
26	from pathlib import Path	26	from pathlib import Path
27	-from unittest.mock import MagicMock	27	+from unittest.mock import MagicMock, patch
28		28
29	from testing_utils.mock import mock_init_config	29	from testing_utils.mock import mock_init_config
30		30
31	from msmodelslim.core.const import QuantType	31	from msmodelslim.core.const import QuantType
32	from msmodelslim.core.practice.interface import Metadata	32	from msmodelslim.core.practice.interface import Metadata
33	-from msmodelslim.utils.exception import SchemaValidateError, ToDoError	33	+from msmodelslim.utils.exception import SchemaValidateError, ToDoError, UnsupportedError
34		34
35	mock_init_config()	35	mock_init_config()
36		36
	@@ -225,6 +225,38 @@ class TestCheckConfig(TestNaiveQuantizationAppBase):
225	class TestGetBestPractice(TestNaiveQuantizationAppBase):	225	class TestGetBestPractice(TestNaiveQuantizationAppBase):
226	"""测试 get_best_practice"""	226	"""测试 get_best_practice"""
227		227
		228	+ def _make_app(self):
		229	+ from msmodelslim.app.naive_quantization.application import NaiveQuantizationApplication
		230	+
		231	+ return NaiveQuantizationApplication(
		232	+ practice_manager=MagicMock(),
		233	+ quant_service=MagicMock(),
		234	+ model_factory=MagicMock(),
		235	+ )
		236	+
		237	+ def _get_best_practice_with_mocked_yaml(self, model_type, apiversion, config_id="test_w8a8"):
		238	+ mock_model_adapter = MagicMock()
		239	+ mock_model_adapter.model_type = model_type
		240	+ mock_model_adapter.get_model_type.return_value = model_type
		241	+ app = self._make_app()
		242	+ mock_config = MagicMock()
		243	+ mock_config.apiversion = apiversion
		244	+ mock_config.metadata.config_id = config_id
		245	+ with (
		246	+ patch(
		247	+ "msmodelslim.app.naive_quantization.application.yaml_safe_load",
		248	+ return_value={},
		249	+ ),
		250	+ patch(
		251	+ "msmodelslim.app.naive_quantization.application.PracticeConfig.model_validate",
		252	+ return_value=mock_config,
		253	+ ),
		254	+ ):
		255	+ return app.get_best_practice(
		256	+ model_adapter=mock_model_adapter,
		257	+ config_path=Path("config.yaml"),
		258	+ )
		259	+
228	def test_get_best_practice_with_config_path(self):	260	def test_get_best_practice_with_config_path(self):
229	"""测试指定 config_path 时直接返回配置"""	261	"""测试指定 config_path 时直接返回配置"""
230	from msmodelslim.app.naive_quantization.application import NaiveQuantizationApplication	262	from msmodelslim.app.naive_quantization.application import NaiveQuantizationApplication
	@@ -290,6 +322,36 @@ spec: {}
290	quant_type=QuantType.W8A8,	322	quant_type=QuantType.W8A8,
291	)	323	)
292		324
		325	+ def test_get_best_practice_raises_unsupported_error_when_transformers_and_vlm_apiversion(self):
		326	+ with self.assertRaises(UnsupportedError) as ctx:
		327	+ self._get_best_practice_with_mocked_yaml("transformers", "multimodal_vlm_modelslim_v1")
		328	+
		329	+ self.assertIn("VLM", str(ctx.exception))
		330	+ self.assertIn("transformers", str(ctx.exception))
		331	+
		332	+ def test_get_best_practice_raises_unsupported_error_when_transformers_and_sd_apiversion(self):
		333	+ with self.assertRaises(UnsupportedError) as ctx:
		334	+ self._get_best_practice_with_mocked_yaml("transformers", "multimodal_sd_modelslim_v1")
		335	+
		336	+ self.assertIn("DiT", str(ctx.exception))
		337	+ self.assertIn("transformers", str(ctx.exception))
		338	+
		339	+ def test_get_best_practice_returns_config_when_transformers_and_modelslim_v1(self):
		340	+ config = self._get_best_practice_with_mocked_yaml("transformers", "modelslim_v1")
		341	+
		342	+ self.assertEqual(config.metadata.config_id, "test_w8a8")
		343	+ self.assertEqual(config.apiversion, "modelslim_v1")
		344	+
		345	+ def test_get_best_practice_returns_config_when_other_model_type_and_vlm_apiversion(self):
		346	+ config = self._get_best_practice_with_mocked_yaml(
		347	+ "Qwen2.5-VL-7B-Instruct",
		348	+ "multimodal_vlm_modelslim_v1",
		349	+ config_id="vlm_ok",
		350	+ )
		351	+
		352	+ self.assertEqual(config.metadata.config_id, "vlm_ok")
		353	+ self.assertEqual(config.apiversion, "multimodal_vlm_modelslim_v1")
		354	+
293		355
294	class TestQuantParameterValidation(TestNaiveQuantizationAppBase):	356	class TestQuantParameterValidation(TestNaiveQuantizationAppBase):
295	"""测试 quant 方法参数校验"""	357	"""测试 quant 方法参数校验"""

Atest/cases/model/transformers/test_detect.py+89-0

@@ -0,0 +1,89 @@
	1	+#!/usr/bin/env python
	2	+# -- coding: UTF-8 --
	3	+
	4	+"""
	5	+-------------------------------------------------------------------------
	6	+This file is part of the MindStudio project.
	7	+Copyright (c) 2026 Huawei Technologies Co.,Ltd.
	8	+
	9	+MindStudio is licensed under Mulan PSL v2.
	10	+You can use this software according to the terms and conditions of the Mulan PSL v2.
	11	+You may obtain a copy of Mulan PSL v2 at:
	12	+
	13	+ http://license.coscl.org.cn/MulanPSL2
	14	+
	15	+THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
	16	+EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
	17	+MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
	18	+See the Mulan PSL v2 for more details.
	19	+-------------------------------------------------------------------------
	20	+"""
	21	+
	22	+import unittest
	23	+from pathlib import Path
	24	+from unittest.mock import patch
	25	+
	26	+from msmodelslim.model.transformers.detect import (
	27	+ LLM_ADAPTER_CLASS_PATH,
	28	+ ModelClass,
	29	+ detect_transformers_kind,
	30	+ resolve_adapter_class_path,
	31	+)
	32	+from msmodelslim.utils.exception import SecurityError, UnsupportedError
	33	+
	34	+
	35	+class TestDetectTransformersKind(unittest.TestCase):
	36	+ def setUp(self):
	37	+ self.model_path = Path("/tmp/transformers-model")
	38	+
	39	+ def test_detect_transformers_kind_returns_llm_when_config_has_no_hints(self):
	40	+ with patch("msmodelslim.model.transformers.detect.json_safe_load", return_value={"model_type": "llama"}):
	41	+ self.assertEqual(detect_transformers_kind(self.model_path), ModelClass.LLM)
	42	+
	43	+ def test_detect_transformers_kind_returns_vlm_when_vision_config_present(self):
	44	+ with patch("msmodelslim.model.transformers.detect.json_safe_load", return_value={"vision_config": {}}):
	45	+ self.assertEqual(detect_transformers_kind(self.model_path), ModelClass.VLM)
	46	+
	47	+ def test_detect_transformers_kind_returns_dit_when_diffusers_version_present(self):
	48	+ with patch(
	49	+ "msmodelslim.model.transformers.detect.json_safe_load",
	50	+ return_value={"_diffusers_version": "0.30.0"},
	51	+ ):
	52	+ self.assertEqual(detect_transformers_kind(self.model_path), ModelClass.DIT)
	53	+
	54	+ def test_detect_transformers_kind_raises_unsupported_error_when_config_missing(self):
	55	+ with patch(
	56	+ "msmodelslim.model.transformers.detect.json_safe_load",
	57	+ side_effect=SecurityError("invalid path"),
	58	+ ):
	59	+ with self.assertRaises(UnsupportedError) as ctx:
	60	+ detect_transformers_kind(self.model_path)
	61	+
	62	+ self.assertIn("config.json", str(ctx.exception))
	63	+ self.assertIn("transformers", str(ctx.exception))
	64	+
	65	+
	66	+class TestResolveAdapterClassPath(unittest.TestCase):
	67	+ def setUp(self):
	68	+ self.model_path = Path("/tmp/transformers-model")
	69	+
	70	+ def test_resolve_adapter_class_path_returns_llm_path_when_llm_config(self):
	71	+ with patch("msmodelslim.model.transformers.detect.json_safe_load", return_value={}):
	72	+ self.assertEqual(resolve_adapter_class_path(self.model_path), LLM_ADAPTER_CLASS_PATH)
	73	+
	74	+ def test_resolve_adapter_class_path_raises_unsupported_error_when_vlm_config(self):
	75	+ with patch("msmodelslim.model.transformers.detect.json_safe_load", return_value={"vision_config": {}}):
	76	+ with self.assertRaises(UnsupportedError) as ctx:
	77	+ resolve_adapter_class_path(self.model_path)
	78	+
	79	+ self.assertIn("VLM", str(ctx.exception))
	80	+
	81	+ def test_resolve_adapter_class_path_raises_unsupported_error_when_dit_config(self):
	82	+ with patch(
	83	+ "msmodelslim.model.transformers.detect.json_safe_load",
	84	+ return_value={"_diffusers_version": "0.30.0"},
	85	+ ):
	86	+ with self.assertRaises(UnsupportedError) as ctx:
	87	+ resolve_adapter_class_path(self.model_path)
	88	+
	89	+ self.assertIn("DiT", str(ctx.exception))

Atest/cases/model/transformers/test_loader_llm_transformers.py+130-0

@@ -0,0 +1,130 @@
	1	+#!/usr/bin/env python
	2	+# -- coding: UTF-8 --
	3	+
	4	+"""
	5	+-------------------------------------------------------------------------
	6	+This file is part of the MindStudio project.
	7	+Copyright (c) 2026 Huawei Technologies Co.,Ltd.
	8	+
	9	+MindStudio is licensed under Mulan PSL v2.
	10	+You can use this software according to the terms and conditions of the Mulan PSL v2.
	11	+You may obtain a copy of Mulan PSL v2 at:
	12	+
	13	+ http://license.coscl.org.cn/MulanPSL2
	14	+
	15	+THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
	16	+EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
	17	+MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
	18	+See the Mulan PSL v2 for more details.
	19	+-------------------------------------------------------------------------
	20	+"""
	21	+
	22	+import unittest
	23	+from pathlib import Path
	24	+from types import SimpleNamespace
	25	+from unittest.mock import patch
	26	+
	27	+from msmodelslim.model.transformers.detect import LLM_ADAPTER_CLASS_PATH
	28	+from msmodelslim.model.transformers.loader import TransformersAdapterLoader
	29	+from msmodelslim.model.transformers.model_adapter import LLMTransformersModel
	30	+from msmodelslim.utils.exception import UnsupportedError, VersionError
	31	+
	32	+
	33	+def _mock_adapter_init(self, model_type, model_path, trust_remote_code=False):
	34	+ self.model_type = model_type
	35	+ self.model_path = model_path
	36	+ self.trust_remote_code = trust_remote_code
	37	+
	38	+
	39	+class TestTransformersAdapterLoader(unittest.TestCase):
	40	+ def setUp(self):
	41	+ self.model_type = "transformers"
	42	+ self.model_path = Path("/tmp/transformers-model")
	43	+ self.loader = TransformersAdapterLoader()
	44	+
	45	+ def test_adapter_class_path_points_to_llm_transformers_model_when_defined(self):
	46	+ self.assertEqual(
	47	+ TransformersAdapterLoader.ADAPTER_CLASS_PATH,
	48	+ LLM_ADAPTER_CLASS_PATH,
	49	+ )
	50	+ self.assertEqual(
	51	+ TransformersAdapterLoader.ADAPTER_CLASS_PATH,
	52	+ "msmodelslim.model.transformers.model_adapter:LLMTransformersModel",
	53	+ )
	54	+
	55	+ def _load_with_patches(self, config=None, **load_kwargs):
	56	+ if config is None:
	57	+ config = {}
	58	+ with patch("msmodelslim.model.transformers.detect.json_safe_load", return_value=config):
	59	+ with patch("msmodelslim.model.plugin_factory.base_loader.DependencyChecker.set_plugin"):
	60	+ with patch("msmodelslim.model.plugin_factory.base_loader.DependencyChecker._check_single"):
	61	+ with patch("msmodelslim.model.plugin_factory.base_loader.get_require_packages", return_value={}):
	62	+ with patch("msmodelslim.model.plugin_factory.base_loader.import_module") as mock_import:
	63	+ mock_import.return_value = SimpleNamespace(LLMTransformersModel=LLMTransformersModel)
	64	+ with patch(
	65	+ "msmodelslim.model.transformers.model_adapter.TransformersModel.__init__",
	66	+ _mock_adapter_init,
	67	+ ):
	68	+ return self.loader.load(
	69	+ model_type=self.model_type,
	70	+ model_path=self.model_path,
	71	+ **load_kwargs,
	72	+ )
	73	+
	74	+ def test_load_returns_llm_transformers_model_when_trust_remote_code_true(self):
	75	+ adapter = self._load_with_patches(trust_remote_code=True)
	76	+
	77	+ self.assertIsInstance(adapter, LLMTransformersModel)
	78	+ self.assertEqual(adapter.model_type, self.model_type)
	79	+ self.assertEqual(adapter.model_path, self.model_path)
	80	+ self.assertTrue(adapter.trust_remote_code)
	81	+
	82	+ def test_load_passes_false_when_trust_remote_code_omitted(self):
	83	+ adapter = self._load_with_patches()
	84	+ self.assertFalse(adapter.trust_remote_code)
	85	+
	86	+ def test_load_raises_unsupported_error_when_vlm_config(self):
	87	+ with self.assertRaises(UnsupportedError) as ctx:
	88	+ self._load_with_patches(config={"vision_config": {}})
	89	+
	90	+ self.assertIn("VLM", str(ctx.exception))
	91	+ self.assertIn("transformers", str(ctx.exception))
	92	+
	93	+ def test_load_raises_unsupported_error_when_dit_config(self):
	94	+ with self.assertRaises(UnsupportedError) as ctx:
	95	+ self._load_with_patches(config={"_diffusers_version": "0.30.0"})
	96	+
	97	+ self.assertIn("DiT", str(ctx.exception))
	98	+ self.assertIn("transformers", str(ctx.exception))
	99	+
	100	+ def test_precheck_sets_plugin_name_when_model_type_valid(self):
	101	+ with patch(
	102	+ "msmodelslim.model.plugin_factory.base_loader.msmodelslim_config",
	103	+ SimpleNamespace(model_adapter_dependencies={}),
	104	+ ):
	105	+ with patch("msmodelslim.model.plugin_factory.base_loader.DependencyChecker.set_plugin") as mock_set:
	106	+ with patch("msmodelslim.model.plugin_factory.base_loader.DependencyChecker._check_single"):
	107	+ self.loader.precheck(
	108	+ model_type=self.model_type,
	109	+ model_path=self.model_path,
	110	+ )
	111	+
	112	+ plugin_name = mock_set.call_args[0][0]
	113	+ self.assertEqual(plugin_name, f"msmodelslim.model_adapter.plugins:{self.model_type}")
	114	+
	115	+ def test_precheck_sets_is_match_false_when_dependency_check_fails(self):
	116	+ self.loader._require_packages = {"numpy": ">=1.26"}
	117	+ with patch(
	118	+ "msmodelslim.model.plugin_factory.base_loader.msmodelslim_config",
	119	+ SimpleNamespace(model_adapter_dependencies={}),
	120	+ ):
	121	+ with patch(
	122	+ "msmodelslim.model.plugin_factory.base_loader.DependencyChecker._check_single",
	123	+ side_effect=VersionError("dependency mismatch"),
	124	+ ):
	125	+ self.loader.precheck(
	126	+ model_type=self.model_type,
	127	+ model_path=self.model_path,
	128	+ )
	129	+
	130	+ self.assertFalse(self.loader._is_match)

Atest/cases/model/transformers/test_model_adapter_llm_transformers.py+155-0

@@ -0,0 +1,155 @@
	1	+#!/usr/bin/env python
	2	+# -- coding: UTF-8 --
	3	+
	4	+"""
	5	+-------------------------------------------------------------------------
	6	+This file is part of the MindStudio project.
	7	+Copyright (c) 2026 Huawei Technologies Co.,Ltd.
	8	+
	9	+MindStudio is licensed under Mulan PSL v2.
	10	+You can use this software according to the terms and conditions of the Mulan PSL v2.
	11	+You may obtain a copy of Mulan PSL v2 at:
	12	+
	13	+ http://license.coscl.org.cn/MulanPSL2
	14	+
	15	+THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
	16	+EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
	17	+MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
	18	+See the Mulan PSL v2 for more details.
	19	+-------------------------------------------------------------------------
	20	+"""
	21	+
	22	+import unittest
	23	+from pathlib import Path
	24	+from unittest.mock import MagicMock, patch
	25	+
	26	+from msmodelslim.core.const import DeviceType
	27	+from msmodelslim.model.transformers.model_adapter import LLMTransformersModel
	28	+from msmodelslim.utils.exception import InvalidModelError
	29	+
	30	+
	31	+class TestLLMTransformersModel(unittest.TestCase):
	32	+ def setUp(self):
	33	+ self.model_type = "transformers"
	34	+ self.model_path = Path(".")
	35	+
	36	+ def _make_adapter(self):
	37	+ with patch(
	38	+ "msmodelslim.model.transformers.model_adapter.TransformersModel.__init__",
	39	+ return_value=None,
	40	+ ):
	41	+ adapter = LLMTransformersModel(model_type=self.model_type, model_path=self.model_path)
	42	+ adapter.model_type = self.model_type
	43	+ adapter.model_path = self.model_path
	44	+ return adapter
	45	+
	46	+ def test_get_model_type_returns_transformers_when_called(self):
	47	+ adapter = self._make_adapter()
	48	+ self.assertEqual(adapter.get_model_type(), "transformers")
	49	+
	50	+ def test_get_model_pedigree_returns_llm_transformers_when_called(self):
	51	+ adapter = self._make_adapter()
	52	+ self.assertEqual(adapter.get_model_pedigree(), "llm_transformers")
	53	+
	54	+ def test_init_model_returns_model_when_load_succeeds(self):
	55	+ adapter = self._make_adapter()
	56	+ mock_model = MagicMock()
	57	+ adapter._load_model = MagicMock(return_value=mock_model)
	58	+
	59	+ result = adapter.init_model(device=DeviceType.NPU)
	60	+
	61	+ self.assertIs(result, mock_model)
	62	+ adapter._load_model.assert_called_once_with(DeviceType.NPU)
	63	+
	64	+ def test_init_model_raises_invalid_model_error_when_load_fails(self):
	65	+ adapter = self._make_adapter()
	66	+ adapter._load_model = MagicMock(side_effect=Exception("Loading failed"))
	67	+
	68	+ with self.assertRaises(InvalidModelError) as ctx:
	69	+ adapter.init_model(device=DeviceType.NPU)
	70	+
	71	+ self.assertIn("dedicated model adapter", ctx.exception.action)
	72	+ self.assertIn("VLM/DiT", ctx.exception.action)
	73	+
	74	+ def test_handle_dataset_returns_tokenized_data_when_helper_succeeds(self):
	75	+ adapter = self._make_adapter()
	76	+ mock_dataset = ["data1", "data2"]
	77	+ adapter._get_tokenized_data = MagicMock(return_value=mock_dataset)
	78	+
	79	+ result = adapter.handle_dataset(dataset="test_data", device=DeviceType.CPU)
	80	+
	81	+ self.assertEqual(result, mock_dataset)
	82	+ adapter._get_tokenized_data.assert_called_once_with("test_data", DeviceType.CPU)
	83	+
	84	+ def test_handle_dataset_raises_invalid_model_error_when_helper_fails(self):
	85	+ adapter = self._make_adapter()
	86	+ adapter._get_tokenized_data = MagicMock(side_effect=Exception("Processing failed"))
	87	+
	88	+ with self.assertRaises(InvalidModelError):
	89	+ adapter.handle_dataset(dataset="test_data", device=DeviceType.CPU)
	90	+
	91	+ @patch("msmodelslim.model.transformers.model_adapter.generated_decoder_layer_visit_func")
	92	+ def test_generate_model_visit_yields_requests_when_helper_succeeds(self, mock_visit):
	93	+ mock_request = MagicMock()
	94	+ mock_visit.return_value = iter([mock_request])
	95	+ adapter = self._make_adapter()
	96	+ mock_model = MagicMock()
	97	+
	98	+ result = list(adapter.generate_model_visit(mock_model))
	99	+
	100	+ self.assertEqual(result, [mock_request])
	101	+ mock_visit.assert_called_once_with(mock_model)
	102	+
	103	+ @patch("msmodelslim.model.transformers.model_adapter.generated_decoder_layer_visit_func")
	104	+ def test_generate_model_visit_raises_invalid_model_error_when_helper_fails(self, mock_visit):
	105	+ mock_visit.side_effect = Exception("visit failed")
	106	+ adapter = self._make_adapter()
	107	+
	108	+ with self.assertRaises(InvalidModelError):
	109	+ list(adapter.generate_model_visit(MagicMock()))
	110	+
	111	+ @patch("msmodelslim.model.transformers.model_adapter.transformers_generated_forward_func")
	112	+ def test_generate_model_forward_yields_requests_when_helper_succeeds(self, mock_forward):
	113	+ mock_request = MagicMock()
	114	+ mock_forward.return_value = iter([mock_request])
	115	+ adapter = self._make_adapter()
	116	+ mock_model = MagicMock()
	117	+ mock_inputs = MagicMock()
	118	+
	119	+ result = list(adapter.generate_model_forward(mock_model, mock_inputs))
	120	+
	121	+ self.assertEqual(result, [mock_request])
	122	+ mock_forward.assert_called_once_with(mock_model, mock_inputs)
	123	+
	124	+ @patch("msmodelslim.model.transformers.model_adapter.transformers_generated_forward_func")
	125	+ def test_generate_model_forward_raises_invalid_model_error_when_helper_fails(self, mock_forward):
	126	+ mock_forward.side_effect = Exception("forward failed")
	127	+ adapter = self._make_adapter()
	128	+
	129	+ with self.assertRaises(InvalidModelError):
	130	+ list(adapter.generate_model_forward(MagicMock(), MagicMock()))
	131	+
	132	+ def test_enable_kv_cache_delegates_to_helper_when_need_kv_cache_true(self):
	133	+ adapter = self._make_adapter()
	134	+ mock_model = MagicMock()
	135	+ adapter._enable_kv_cache = MagicMock(return_value=None)
	136	+
	137	+ adapter.enable_kv_cache(model=mock_model, need_kv_cache=True)
	138	+
	139	+ adapter._enable_kv_cache.assert_called_once_with(mock_model, True)
	140	+
	141	+ def test_enable_kv_cache_delegates_to_helper_when_need_kv_cache_false(self):
	142	+ adapter = self._make_adapter()
	143	+ mock_model = MagicMock()
	144	+ adapter._enable_kv_cache = MagicMock(return_value=None)
	145	+
	146	+ adapter.enable_kv_cache(model=mock_model, need_kv_cache=False)
	147	+
	148	+ adapter._enable_kv_cache.assert_called_once_with(mock_model, False)
	149	+
	150	+ def test_enable_kv_cache_raises_invalid_model_error_when_helper_fails(self):
	151	+ adapter = self._make_adapter()
	152	+ adapter._enable_kv_cache = MagicMock(side_effect=Exception("Enable KV cache failed"))
	153	+
	154	+ with self.assertRaises(InvalidModelError):
	155	+ adapter.enable_kv_cache(model=MagicMock(), need_kv_cache=True)