add grpo

data/mllm_video_audio_demo.json

+[
+  {
+    "messages": [
+      {
+        "content": "<video><audio>What is the video describing?",
+        "role": "user"
+      },
+      {
+        "content": "A girl who is drawing a picture of a guitar and feel nervous.",
+        "role": "assistant"
+      }
+    ],
+    "videos": [
+      "mllm_demo_data/4.mp4"
+    ],
+    "audios": [
+      "mllm_demo_data/4.mp3"
+    ]
+  },
+  {
+    "messages": [
+      {
+        "content": "<video><audio>What does this girl say?",
+        "role": "user"
+      },
+      {
+        "content": "She says: 'Hello! Take a look at what am I drawing!'",
+        "role": "assistant"
+      }
+    ],
+    "videos": [
+      "mllm_demo_data/4.mp4"
+    ],
+    "audios": [
+      "mllm_demo_data/4.mp3"
+    ]
+  },
+  {
+    "messages": [
+      {
+        "content": "<video><audio>What is this girl drawing with?",
+        "role": "user"
+      },
+      {
+        "content": "She is drawing with an iPad.",
+        "role": "assistant"
+      }
+    ],
+    "videos": [
+      "mllm_demo_data/4.mp4"
+    ],
+    "audios": [
+      "mllm_demo_data/4.mp3"
+    ]
+  }
+]

data/mllm_video_demo.json

+[
+  {
+    "messages": [
+      {
+        "content": "<video>Why is this video funny?",
+        "role": "user"
+      },
+      {
+        "content": "Because a baby is reading, and he is so cute!",
+        "role": "assistant"
+      }
+    ],
+    "videos": [
+      "mllm_demo_data/1.mp4"
+    ]
+  },
+  {
+    "messages": [
+      {
+        "content": "<video>What is she doing?",
+        "role": "user"
+      },
+      {
+        "content": "She is cooking.",
+        "role": "assistant"
+      }
+    ],
+    "videos": [
+      "mllm_demo_data/2.avi"
+    ]
+  },
+  {
+    "messages": [
+      {
+        "content": "<video>What's in the video?",
+        "role": "user"
+      },
+      {
+        "content": "A baby is playing in the living room.",
+        "role": "assistant"
+      }
+    ],
+    "videos": [
+      "mllm_demo_data/3.mp4"
+    ]
+  }
+]

data/ultra_chat/ultra_chat.py

+# Copyright 2025 the LlamaFactory team.
+# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+import os
+
+import datasets
+
+
+_HF_ENDPOINT = os.getenv("HF_ENDPOINT", "https://huggingface.co")
+
+_DESCRIPTION = "UltraChat: Large-scale, Informative, and Diverse Multi-round Dialogue Data."
+
+_CITATION = """\
+@misc{UltraChat,
+  author = {Ding, Ning and Chen, Yulin and Xu, Bokai and Hu, Shengding and others},
+  title = {UltraChat: A Large-scale Auto-generated Multi-round Dialogue Data},
+  year = {2023},
+  publisher = {GitHub},
+  journal = {GitHub repository},
+  howpublished = {\\url{https://github.com/thunlp/ultrachat}},
+}
+"""
+
+_HOMEPAGE = f"{_HF_ENDPOINT}/datasets/stingning/ultrachat"
+_LICENSE = "cc-by-nc-4.0"
+_BASE_DATA_URL = f"{_HF_ENDPOINT}/datasets/stingning/ultrachat/resolve/main/train_{{idx}}.jsonl"
+
+
+class UltraChat(datasets.GeneratorBasedBuilder):
+    VERSION = datasets.Version("0.0.0")
+
+    def _info(self):
+        features = datasets.Features(
+            {"conversations": [{"from": datasets.Value("string"), "value": datasets.Value("string")}]}
+        )
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION, features=features, homepage=_HOMEPAGE, license=_LICENSE, citation=_CITATION
+        )
+
+    def _split_generators(self, dl_manager: datasets.DownloadManager):
+        file_paths = [dl_manager.download(_BASE_DATA_URL.format(idx=idx)) for idx in range(10)]  # multiple shards
+        return [datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepaths": file_paths})]
+
+    def _generate_examples(self, filepaths: list[str]):
+        for filepath in filepaths:
+            with open(filepath, encoding="utf-8") as f:
+                for row in f:
+                    try:
+                        data = json.loads(row)
+                    except Exception:
+                        continue
+                    key: int = data["id"]
+                    content: list[str] = data["data"]
+                    if len(content) % 2 == 1:
+                        content.pop(-1)
+                    if len(content) < 2:
+                        continue
+                    conversations = [
+                        {"from": "human" if i % 2 == 0 else "gpt", "value": content[i]} for i in range(len(content))
+                    ]
+                    yield key, {"conversations": conversations}

docker/Dockerfile

+FROM docker pull image.sourcefind.cn:5000/dcu/admin/base/vllm:0.8.5-ubuntu22.04-dtk25.04.1-rc5-das1.6-py3.10-20250711
\ No newline at end of file

evaluation/ceval/ceval.py

+# Copyright 2025 the LlamaFactory team.
+# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+
+import datasets
+import pandas as pd
+
+
+_CITATION = """\
+@article{huang2023ceval,
+  title={C-Eval: A Multi-Level Multi-Discipline Chinese Evaluation Suite for Foundation Models},
+  author={Huang, Yuzhen and Bai, Yuzhuo and Zhu, Zhihao and others},
+  journal={arXiv preprint arXiv:2305.08322},
+  year={2023}
+}
+"""
+
+_DESCRIPTION = """\
+C-Eval is a comprehensive Chinese evaluation suite for foundation models.
+It consists of 13948 multi-choice questions spanning 52 diverse disciplines and four difficulty levels.
+"""
+
+_HOMEPAGE = "https://cevalbenchmark.com"
+
+_LICENSE = "Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License"
+
+_URL = "ceval.zip"
+
+task_list = [
+    "computer_network",
+    "operating_system",
+    "computer_architecture",
+    "college_programming",
+    "college_physics",
+    "college_chemistry",
+    "advanced_mathematics",
+    "probability_and_statistics",
+    "discrete_mathematics",
+    "electrical_engineer",
+    "metrology_engineer",
+    "high_school_mathematics",
+    "high_school_physics",
+    "high_school_chemistry",
+    "high_school_biology",
+    "middle_school_mathematics",
+    "middle_school_biology",
+    "middle_school_physics",
+    "middle_school_chemistry",
+    "veterinary_medicine",
+    "college_economics",
+    "business_administration",
+    "marxism",
+    "mao_zedong_thought",
+    "education_science",
+    "teacher_qualification",
+    "high_school_politics",
+    "high_school_geography",
+    "middle_school_politics",
+    "middle_school_geography",
+    "modern_chinese_history",
+    "ideological_and_moral_cultivation",
+    "logic",
+    "law",
+    "chinese_language_and_literature",
+    "art_studies",
+    "professional_tour_guide",
+    "legal_professional",
+    "high_school_chinese",
+    "high_school_history",
+    "middle_school_history",
+    "civil_servant",
+    "sports_science",
+    "plant_protection",
+    "basic_medicine",
+    "clinical_medicine",
+    "urban_and_rural_planner",
+    "accountant",
+    "fire_engineer",
+    "environmental_impact_assessment_engineer",
+    "tax_accountant",
+    "physician",
+]
+
+
+class CevalConfig(datasets.BuilderConfig):
+    def __init__(self, **kwargs):
+        super().__init__(version=datasets.Version("1.0.0"), **kwargs)
+
+
+class Ceval(datasets.GeneratorBasedBuilder):
+    BUILDER_CONFIGS = [
+        CevalConfig(
+            name=task_name,
+        )
+        for task_name in task_list
+    ]
+
+    def _info(self):
+        features = datasets.Features(
+            {
+                "id": datasets.Value("int32"),
+                "question": datasets.Value("string"),
+                "A": datasets.Value("string"),
+                "B": datasets.Value("string"),
+                "C": datasets.Value("string"),
+                "D": datasets.Value("string"),
+                "answer": datasets.Value("string"),
+                "explanation": datasets.Value("string"),
+            }
+        )
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        data_dir = dl_manager.download_and_extract(_URL)
+        task_name = self.config.name
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TEST,
+                gen_kwargs={
+                    "filepath": os.path.join(data_dir, "test", f"{task_name}_test.csv"),
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.VALIDATION,
+                gen_kwargs={
+                    "filepath": os.path.join(data_dir, "val", f"{task_name}_val.csv"),
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={
+                    "filepath": os.path.join(data_dir, "dev", f"{task_name}_dev.csv"),
+                },
+            ),
+        ]
+
+    def _generate_examples(self, filepath):
+        df = pd.read_csv(filepath, encoding="utf-8")
+        for i, instance in enumerate(df.to_dict(orient="records")):
+            if "answer" not in instance.keys():
+                instance["answer"] = ""
+            if "explanation" not in instance.keys():
+                instance["explanation"] = ""
+            yield i, instance

evaluation/ceval/mapping.json

+{
+  "accountant": {
+    "name": "注册会计师",
+    "category": "Other"
+  },
+  "advanced_mathematics": {
+    "name": "高等数学",
+    "category": "STEM"
+  },
+  "art_studies": {
+    "name": "艺术学",
+    "category": "Humanities"
+  },
+  "basic_medicine": {
+    "name": "基础医学",
+    "category": "Other"
+  },
+  "business_administration": {
+    "name": "工商管理",
+    "category": "Social Sciences"
+  },
+  "chinese_language_and_literature": {
+    "name": "中国语言文学",
+    "category": "Humanities"
+  },
+  "civil_servant": {
+    "name": "公务员",
+    "category": "Other"
+  },
+  "clinical_medicine": {
+    "name": "临床医学",
+    "category": "Other"
+  },
+  "college_chemistry": {
+    "name": "大学化学",
+    "category": "STEM"
+  },
+  "college_economics": {
+    "name": "大学经济学",
+    "category": "Social Sciences"
+  },
+  "college_physics": {
+    "name": "大学物理",
+    "category": "STEM"
+  },
+  "college_programming": {
+    "name": "大学编程",
+    "category": "STEM"
+  },
+  "computer_architecture": {
+    "name": "计算机组成",
+    "category": "STEM"
+  },
+  "computer_network": {
+    "name": "计算机网络",
+    "category": "STEM"
+  },
+  "discrete_mathematics": {
+    "name": "离散数学",
+    "category": "STEM"
+  },
+  "education_science": {
+    "name": "教育学",
+    "category": "Social Sciences"
+  },
+  "electrical_engineer": {
+    "name": "注册电气工程师",
+    "category": "STEM"
+  },
+  "environmental_impact_assessment_engineer": {
+    "name": "环境影响评价工程师",
+    "category": "Other"
+  },
+  "fire_engineer": {
+    "name": "注册消防工程师",
+    "category": "Other"
+  },
+  "high_school_biology": {
+    "name": "高中生物",
+    "category": "STEM"
+  },
+  "high_school_chemistry": {
+    "name": "高中化学",
+    "category": "STEM"
+  },
+  "high_school_chinese": {
+    "name": "高中语文",
+    "category": "Humanities"
+  },
+  "high_school_geography": {
+    "name": "高中地理",
+    "category": "Social Sciences"
+  },
+  "high_school_history": {
+    "name": "高中历史",
+    "category": "Humanities"
+  },
+  "high_school_mathematics": {
+    "name": "高中数学",
+    "category": "STEM"
+  },
+  "high_school_physics": {
+    "name": "高中物理",
+    "category": "STEM"
+  },
+  "high_school_politics": {
+    "name": "高中政治",
+    "category": "Social Sciences"
+  },
+  "ideological_and_moral_cultivation": {
+    "name": "思想道德修养与法律基础",
+    "category": "Humanities"
+  },
+  "law": {
+    "name": "法学",
+    "category": "Humanities"
+  },
+  "legal_professional": {
+    "name": "法律职业资格",
+    "category": "Humanities"
+  },
+  "logic": {
+    "name": "逻辑学",
+    "category": "Humanities"
+  },
+  "mao_zedong_thought": {
+    "name": "毛泽东思想和中国特色社会主义理论体系概论",
+    "category": "Social Sciences"
+  },
+  "marxism": {
+    "name": "马克思主义基本原理",
+    "category": "Social Sciences"
+  },
+  "metrology_engineer": {
+    "name": "注册计量师",
+    "category": "STEM"
+  },
+  "middle_school_biology": {
+    "name": "初中生物",
+    "category": "STEM"
+  },
+  "middle_school_chemistry": {
+    "name": "初中化学",
+    "category": "STEM"
+  },
+  "middle_school_geography": {
+    "name": "初中地理",
+    "category": "Social Sciences"
+  },
+  "middle_school_history": {
+    "name": "初中历史",
+    "category": "Humanities"
+  },
+  "middle_school_mathematics": {
+    "name": "初中数学",
+    "category": "STEM"
+  },
+  "middle_school_physics": {
+    "name": "初中物理",
+    "category": "STEM"
+  },
+  "middle_school_politics": {
+    "name": "初中政治",
+    "category": "Social Sciences"
+  },
+  "modern_chinese_history": {
+    "name": "近代史纲要",
+    "category": "Humanities"
+  },
+  "operating_system": {
+    "name": "操作系统",
+    "category": "STEM"
+  },
+  "physician": {
+    "name": "医师资格",
+    "category": "Other"
+  },
+  "plant_protection": {
+    "name": "植物保护",
+    "category": "Other"
+  },
+  "probability_and_statistics": {
+    "name": "概率统计",
+    "category": "STEM"
+  },
+  "professional_tour_guide": {
+    "name": "导游资格",
+    "category": "Humanities"
+  },
+  "sports_science": {
+    "name": "体育学",
+    "category": "Other"
+  },
+  "tax_accountant": {
+    "name": "税务师",
+    "category": "Other"
+  },
+  "teacher_qualification": {
+    "name": "教师资格",
+    "category": "Social Sciences"
+  },
+  "urban_and_rural_planner": {
+    "name": "注册城乡规划师",
+    "category": "Other"
+  },
+  "veterinary_medicine": {
+    "name": "兽医学",
+    "category": "STEM"
+  }
+}

evaluation/cmmlu/cmmlu.py

+# Copyright 2025 the LlamaFactory team.
+# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+
+import datasets
+import pandas as pd
+
+
+_CITATION = """\
+@article{li2023cmmlu,
+  title={CMMLU: Measuring massive multitask language understanding in Chinese},
+  author={Haonan Li and Yixuan Zhang and Fajri Koto and Yifei Yang and others,
+  journal={arXiv preprint arXiv:2306.09212},
+  year={2023}
+}
+"""
+
+_DESCRIPTION = """\
+CMMLU is a comprehensive Chinese assessment suite specifically designed to evaluate the advanced knowledge
+and reasoning abilities of LLMs within the Chinese language and cultural context.
+"""
+
+_HOMEPAGE = "https://github.com/haonan-li/CMMLU"
+
+_LICENSE = "Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License"
+
+_URL = "cmmlu.zip"
+
+task_list = [
+    "agronomy",
+    "anatomy",
+    "ancient_chinese",
+    "arts",
+    "astronomy",
+    "business_ethics",
+    "chinese_civil_service_exam",
+    "chinese_driving_rule",
+    "chinese_food_culture",
+    "chinese_foreign_policy",
+    "chinese_history",
+    "chinese_literature",
+    "chinese_teacher_qualification",
+    "clinical_knowledge",
+    "college_actuarial_science",
+    "college_education",
+    "college_engineering_hydrology",
+    "college_law",
+    "college_mathematics",
+    "college_medical_statistics",
+    "college_medicine",
+    "computer_science",
+    "computer_security",
+    "conceptual_physics",
+    "construction_project_management",
+    "economics",
+    "education",
+    "electrical_engineering",
+    "elementary_chinese",
+    "elementary_commonsense",
+    "elementary_information_and_technology",
+    "elementary_mathematics",
+    "ethnology",
+    "food_science",
+    "genetics",
+    "global_facts",
+    "high_school_biology",
+    "high_school_chemistry",
+    "high_school_geography",
+    "high_school_mathematics",
+    "high_school_physics",
+    "high_school_politics",
+    "human_sexuality",
+    "international_law",
+    "journalism",
+    "jurisprudence",
+    "legal_and_moral_basis",
+    "logical",
+    "machine_learning",
+    "management",
+    "marketing",
+    "marxist_theory",
+    "modern_chinese",
+    "nutrition",
+    "philosophy",
+    "professional_accounting",
+    "professional_law",
+    "professional_medicine",
+    "professional_psychology",
+    "public_relations",
+    "security_study",
+    "sociology",
+    "sports_science",
+    "traditional_chinese_medicine",
+    "virology",
+    "world_history",
+    "world_religions",
+]
+
+
+class CMMLUConfig(datasets.BuilderConfig):
+    def __init__(self, **kwargs):
+        super().__init__(version=datasets.Version("1.0.1"), **kwargs)
+
+
+class CMMLU(datasets.GeneratorBasedBuilder):
+    BUILDER_CONFIGS = [
+        CMMLUConfig(
+            name=task_name,
+        )
+        for task_name in task_list
+    ]
+
+    def _info(self):
+        features = datasets.Features(
+            {
+                "question": datasets.Value("string"),
+                "A": datasets.Value("string"),
+                "B": datasets.Value("string"),
+                "C": datasets.Value("string"),
+                "D": datasets.Value("string"),
+                "answer": datasets.Value("string"),
+            }
+        )
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        data_dir = dl_manager.download_and_extract(_URL)
+        task_name = self.config.name
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TEST,
+                gen_kwargs={
+                    "filepath": os.path.join(data_dir, f"test/{task_name}.csv"),
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={
+                    "filepath": os.path.join(data_dir, f"dev/{task_name}.csv"),
+                },
+            ),
+        ]
+
+    def _generate_examples(self, filepath):
+        df = pd.read_csv(filepath, header=0, index_col=0, encoding="utf-8")
+        for i, instance in enumerate(df.to_dict(orient="records")):
+            question = instance.pop("Question", "")
+            answer = instance.pop("Answer", "")
+            instance["question"] = question
+            instance["answer"] = answer
+            yield i, instance

evaluation/cmmlu/mapping.json

+{
+  "agronomy": {
+    "name": "农学",
+    "category": "Other"
+  },
+  "anatomy": {
+    "name": "解剖学",
+    "category": "STEM"
+  },
+  "ancient_chinese": {
+    "name": "古汉语",
+    "category": "Social Sciences"
+  },
+  "arts": {
+    "name": "艺术学",
+    "category": "Humanities"
+  },
+  "astronomy": {
+    "name": "天文学",
+    "category": "STEM"
+  },
+  "business_ethics": {
+    "name": "商业伦理",
+    "category": "Social Sciences"
+  },
+  "chinese_civil_service_exam": {
+    "name": "中国公务员考试",
+    "category": "Social Sciences"
+  },
+  "chinese_driving_rule": {
+    "name": "中国驾驶规则",
+    "category": "Other"
+  },
+  "chinese_food_culture": {
+    "name": "中国饮食文化",
+    "category": "Social Sciences"
+  },
+  "chinese_foreign_policy": {
+    "name": "中国外交政策",
+    "category": "Social Sciences"
+  },
+  "chinese_history": {
+    "name": "中国历史",
+    "category": "Humanities"
+  },
+  "chinese_literature": {
+    "name": "中国文学",
+    "category": "Humanities"
+  },
+  "chinese_teacher_qualification": {
+    "name": "中国教师资格",
+    "category": "Social Sciences"
+  },
+  "college_actuarial_science": {
+    "name": "大学精算学",
+    "category": "STEM"
+  },
+  "college_education": {
+    "name": "大学教育学",
+    "category": "Social Sciences"
+  },
+  "college_engineering_hydrology": {
+    "name": "大学工程水文学",
+    "category": "STEM"
+  },
+  "college_law": {
+    "name": "大学法律",
+    "category": "Humanities"
+  },
+  "college_mathematics": {
+    "name": "大学数学",
+    "category": "STEM"
+  },
+  "college_medical_statistics": {
+    "name": "大学医学统计",
+    "category": "STEM"
+  },
+  "clinical_knowledge": {
+    "name": "临床知识",
+    "category": "Other"
+  },
+  "college_medicine": {
+    "name": "大学医学",
+    "category": "Other"
+  },
+  "computer_science": {
+    "name": "计算机科学",
+    "category": "STEM"
+  },
+  "computer_security": {
+    "name": "计算机安全",
+    "category": "Other"
+  },
+  "conceptual_physics": {
+    "name": "概念物理学",
+    "category": "STEM"
+  },
+  "construction_project_management": {
+    "name": "建设工程管理",
+    "category": "Other"
+  },
+  "economics": {
+    "name": "经济学",
+    "category": "Social Sciences"
+  },
+  "education": {
+    "name": "教育学",
+    "category": "Social Sciences"
+  },
+  "elementary_chinese": {
+    "name": "小学语文",
+    "category": "Social Sciences"
+  },
+  "elementary_commonsense": {
+    "name": "小学常识",
+    "category": "Other"
+  },
+  "elementary_information_and_technology": {
+    "name": "小学信息技术",
+    "category": "Other"
+  },
+  "electrical_engineering": {
+    "name": "电气工程",
+    "category": "STEM"
+  },
+  "elementary_mathematics": {
+    "name": "初等数学",
+    "category": "STEM"
+  },
+  "ethnology": {
+    "name": "民族学",
+    "category": "Social Sciences"
+  },
+  "food_science": {
+    "name": "食品科学",
+    "category": "Other"
+  },
+  "genetics": {
+    "name": "遗传学",
+    "category": "STEM"
+  },
+  "global_facts": {
+    "name": "全球事实",
+    "category": "Humanities"
+  },
+  "high_school_biology": {
+    "name": "高中生物",
+    "category": "STEM"
+  },
+  "high_school_chemistry": {
+    "name": "高中化学",
+    "category": "STEM"
+  },
+  "high_school_geography": {
+    "name": "高中地理",
+    "category": "Social Sciences"
+  },
+  "high_school_mathematics": {
+    "name": "高中数学",
+    "category": "STEM"
+  },
+  "high_school_physics": {
+    "name": "高中物理学",
+    "category": "STEM"
+  },
+  "high_school_politics": {
+    "name": "高中政治",
+    "category": "Social Sciences"
+  },
+  "human_sexuality": {
+    "name": "人类性行为",
+    "category": "Other"
+  },
+  "international_law": {
+    "name": "国际法学",
+    "category": "Humanities"
+  },
+  "journalism": {
+    "name": "新闻学",
+    "category": "Social Sciences"
+  },
+  "jurisprudence": {
+    "name": "法理学",
+    "category": "Humanities"
+  },
+  "legal_and_moral_basis": {
+    "name": "法律与道德基础",
+    "category": "Other"
+  },
+  "logical": {
+    "name": "逻辑学",
+    "category": "Humanities"
+  },
+  "machine_learning": {
+    "name": "机器学习",
+    "category": "STEM"
+  },
+  "management": {
+    "name": "管理学",
+    "category": "Social Sciences"
+  },
+  "marketing": {
+    "name": "市场营销",
+    "category": "Social Sciences"
+  },
+  "marxist_theory": {
+    "name": "马克思主义理论",
+    "category": "Humanities"
+  },
+  "modern_chinese": {
+    "name": "现代汉语",
+    "category": "Social Sciences"
+  },
+  "nutrition": {
+    "name": "营养学",
+    "category": "Other"
+  },
+  "philosophy": {
+    "name": "哲学",
+    "category": "Humanities"
+  },
+  "professional_accounting": {
+    "name": "专业会计",
+    "category": "Social Sciences"
+  },
+  "professional_law": {
+    "name": "专业法学",
+    "category": "Humanities"
+  },
+  "professional_medicine": {
+    "name": "专业医学",
+    "category": "Other"
+  },
+  "professional_psychology": {
+    "name": "专业心理学",
+    "category": "Social Sciences"
+  },
+  "public_relations": {
+    "name": "公共关系",
+    "category": "Social Sciences"
+  },
+  "security_study": {
+    "name": "安全研究",
+    "category": "Social Sciences"
+  },
+  "sociology": {
+    "name": "社会学",
+    "category": "Social Sciences"
+  },
+  "sports_science": {
+    "name": "体育学",
+    "category": "Other"
+  },
+  "traditional_chinese_medicine": {
+    "name": "中医中药",
+    "category": "Other"
+  },
+  "virology": {
+    "name": "病毒学",
+    "category": "STEM"
+  },
+  "world_history": {
+    "name": "世界历史",
+    "category": "Humanities"
+  },
+  "world_religions": {
+    "name": "世界宗教",
+    "category": "Humanities"
+  }
+}

evaluation/mmlu/mapping.json

+{
+  "abstract_algebra": {
+    "name": "abstract algebra",
+    "category": "STEM"
+  },
+  "anatomy": {
+    "name": "anatomy",
+    "category": "Other"
+  },
+  "astronomy": {
+    "name": "astronomy",
+    "category": "STEM"
+  },
+  "business_ethics": {
+    "name": "business ethics",
+    "category": "Other"
+  },
+  "clinical_knowledge": {
+    "name": "clinical knowledge",
+    "category": "Other"
+  },
+  "college_biology": {
+    "name": "college biology",
+    "category": "STEM"
+  },
+  "college_chemistry": {
+    "name": "college chemistry",
+    "category": "STEM"
+  },
+  "college_computer_science": {
+    "name": "college computer science",
+    "category": "STEM"
+  },
+  "college_mathematics": {
+    "name": "college mathematics",
+    "category": "STEM"
+  },
+  "college_medicine": {
+    "name": "college medicine",
+    "category": "Other"
+  },
+  "college_physics": {
+    "name": "college physics",
+    "category": "STEM"
+  },
+  "computer_security": {
+    "name": "computer security",
+    "category": "STEM"
+  },
+  "conceptual_physics": {
+    "name": "conceptual physics",
+    "category": "STEM"
+  },
+  "econometrics": {
+    "name": "econometrics",
+    "category": "Social Sciences"
+  },
+  "electrical_engineering": {
+    "name": "electrical engineering",
+    "category": "STEM"
+  },
+  "elementary_mathematics": {
+    "name": "elementary mathematics",
+    "category": "STEM"
+  },
+  "formal_logic": {
+    "name": "formal logic",
+    "category": "Humanities"
+  },
+  "global_facts": {
+    "name": "global facts",
+    "category": "Other"
+  },
+  "high_school_biology": {
+    "name": "high school biology",
+    "category": "STEM"
+  },
+  "high_school_chemistry": {
+    "name": "high school chemistry",
+    "category": "STEM"
+  },
+  "high_school_computer_science": {
+    "name": "high school computer science",
+    "category": "STEM"
+  },
+  "high_school_european_history": {
+    "name": "high school european history",
+    "category": "Humanities"
+  },
+  "high_school_geography": {
+    "name": "high school geography",
+    "category": "Social Sciences"
+  },
+  "high_school_government_and_politics": {
+    "name": "high school government and politics",
+    "category": "Social Sciences"
+  },
+  "high_school_macroeconomics": {
+    "name": "high school macroeconomics",
+    "category": "Social Sciences"
+  },
+  "high_school_mathematics": {
+    "name": "high school mathematics",
+    "category": "STEM"
+  },
+  "high_school_microeconomics": {
+    "name": "high school microeconomics",
+    "category": "Social Sciences"
+  },
+  "high_school_physics": {
+    "name": "high school physics",
+    "category": "STEM"
+  },
+  "high_school_psychology": {
+    "name": "high school psychology",
+    "category": "Social Sciences"
+  },
+  "high_school_statistics": {
+    "name": "high school statistics",
+    "category": "STEM"
+  },
+  "high_school_us_history": {
+    "name": "high school us history",
+    "category": "Humanities"
+  },
+  "high_school_world_history": {
+    "name": "high school world history",
+    "category": "Humanities"
+  },
+  "human_aging": {
+    "name": "human aging",
+    "category": "Other"
+  },
+  "human_sexuality": {
+    "name": "human sexuality",
+    "category": "Social Sciences"
+  },
+  "international_law": {
+    "name": "international law",
+    "category": "Humanities"
+  },
+  "jurisprudence": {
+    "name": "jurisprudence",
+    "category": "Humanities"
+  },
+  "logical_fallacies": {
+    "name": "logical fallacies",
+    "category": "Humanities"
+  },
+  "machine_learning": {
+    "name": "machine learning",
+    "category": "STEM"
+  },
+  "management": {
+    "name": "management",
+    "category": "Other"
+  },
+  "marketing": {
+    "name": "marketing",
+    "category": "Other"
+  },
+  "medical_genetics": {
+    "name": "medical genetics",
+    "category": "Other"
+  },
+  "miscellaneous": {
+    "name": "miscellaneous",
+    "category": "Other"
+  },
+  "moral_disputes": {
+    "name": "moral disputes",
+    "category": "Humanities"
+  },
+  "moral_scenarios": {
+    "name": "moral scenarios",
+    "category": "Humanities"
+  },
+  "nutrition": {
+    "name": "nutrition",
+    "category": "Other"
+  },
+  "philosophy": {
+    "name": "philosophy",
+    "category": "Humanities"
+  },
+  "prehistory": {
+    "name": "prehistory",
+    "category": "Humanities"
+  },
+  "professional_accounting": {
+    "name": "professional accounting",
+    "category": "Other"
+  },
+  "professional_law": {
+    "name": "professional law",
+    "category": "Humanities"
+  },
+  "professional_medicine": {
+    "name": "professional medicine",
+    "category": "Other"
+  },
+  "professional_psychology": {
+    "name": "professional psychology",
+    "category": "Social Sciences"
+  },
+  "public_relations": {
+    "name": "public relations",
+    "category": "Social Sciences"
+  },
+  "security_studies": {
+    "name": "security studies",
+    "category": "Social Sciences"
+  },
+  "sociology": {
+    "name": "sociology",
+    "category": "Social Sciences"
+  },
+  "us_foreign_policy": {
+    "name": "us foreign policy",
+    "category": "Social Sciences"
+  },
+  "virology": {
+    "name": "virology",
+    "category": "Other"
+  },
+  "world_religions": {
+    "name": "world religions",
+    "category": "Humanities"
+  }
+}

evaluation/mmlu/mmlu.py

+# Copyright 2025 the LlamaFactory team.
+# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+
+import datasets
+import pandas as pd
+
+
+_CITATION = """\
+@article{hendryckstest2021,
+  title={Measuring Massive Multitask Language Understanding},
+  author={Dan Hendrycks and Collin Burns and others},
+  journal={Proceedings of the International Conference on Learning Representations (ICLR)},
+  year={2021}
+}
+"""
+
+_DESCRIPTION = """\
+Measuring Massive Multitask Language Understanding by Dan Hendrycks, Collin Burns, Steven Basart,
+Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt (ICLR 2021).
+"""
+
+_HOMEPAGE = "https://github.com/hendrycks/test"
+
+_LICENSE = "MIT"
+
+_URL = "mmlu.zip"
+
+task_list = [
+    "high_school_european_history",
+    "business_ethics",
+    "clinical_knowledge",
+    "medical_genetics",
+    "high_school_us_history",
+    "high_school_physics",
+    "high_school_world_history",
+    "virology",
+    "high_school_microeconomics",
+    "econometrics",
+    "college_computer_science",
+    "high_school_biology",
+    "abstract_algebra",
+    "professional_accounting",
+    "philosophy",
+    "professional_medicine",
+    "nutrition",
+    "global_facts",
+    "machine_learning",
+    "security_studies",
+    "public_relations",
+    "professional_psychology",
+    "prehistory",
+    "anatomy",
+    "human_sexuality",
+    "college_medicine",
+    "high_school_government_and_politics",
+    "college_chemistry",
+    "logical_fallacies",
+    "high_school_geography",
+    "elementary_mathematics",
+    "human_aging",
+    "college_mathematics",
+    "high_school_psychology",
+    "formal_logic",
+    "high_school_statistics",
+    "international_law",
+    "high_school_mathematics",
+    "high_school_computer_science",
+    "conceptual_physics",
+    "miscellaneous",
+    "high_school_chemistry",
+    "marketing",
+    "professional_law",
+    "management",
+    "college_physics",
+    "jurisprudence",
+    "world_religions",
+    "sociology",
+    "us_foreign_policy",
+    "high_school_macroeconomics",
+    "computer_security",
+    "moral_scenarios",
+    "moral_disputes",
+    "electrical_engineering",
+    "astronomy",
+    "college_biology",
+]
+
+
+class MMLUConfig(datasets.BuilderConfig):
+    def __init__(self, **kwargs):
+        super().__init__(version=datasets.Version("1.0.0"), **kwargs)
+
+
+class MMLU(datasets.GeneratorBasedBuilder):
+    BUILDER_CONFIGS = [
+        MMLUConfig(
+            name=task_name,
+        )
+        for task_name in task_list
+    ]
+
+    def _info(self):
+        features = datasets.Features(
+            {
+                "question": datasets.Value("string"),
+                "A": datasets.Value("string"),
+                "B": datasets.Value("string"),
+                "C": datasets.Value("string"),
+                "D": datasets.Value("string"),
+                "answer": datasets.Value("string"),
+            }
+        )
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        data_dir = dl_manager.download_and_extract(_URL)
+        task_name = self.config.name
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TEST,
+                gen_kwargs={
+                    "filepath": os.path.join(data_dir, "data", "test", f"{task_name}_test.csv"),
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.VALIDATION,
+                gen_kwargs={
+                    "filepath": os.path.join(data_dir, "data", "val", f"{task_name}_val.csv"),
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={
+                    "filepath": os.path.join(data_dir, "data", "dev", f"{task_name}_dev.csv"),
+                },
+            ),
+        ]
+
+    def _generate_examples(self, filepath):
+        df = pd.read_csv(filepath, header=None)
+        df.columns = ["question", "A", "B", "C", "D", "answer"]
+
+        yield from enumerate(df.to_dict(orient="records"))

examples/README.md

+We provide diverse examples about fine-tuning LLMs.
+
+Make sure to execute these commands in the `LLaMA-Factory` directory.
+
+## Table of Contents
+
+- [LoRA Fine-Tuning](#lora-fine-tuning)
+- [QLoRA Fine-Tuning](#qlora-fine-tuning)
+- [Full-Parameter Fine-Tuning](#full-parameter-fine-tuning)
+- [Merging LoRA Adapters and Quantization](#merging-lora-adapters-and-quantization)
+- [Inferring LoRA Fine-Tuned Models](#inferring-lora-fine-tuned-models)
+- [Extras](#extras)
+
+Use `CUDA_VISIBLE_DEVICES` (GPU) or `ASCEND_RT_VISIBLE_DEVICES` (NPU) to choose computing devices.
+
+By default, LLaMA-Factory uses all visible computing devices.
+
+Basic usage:
+
+```bash
+llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
+```
+
+Advanced usage:
+
+```bash
+CUDA_VISIBLE_DEVICES=0,1 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml \
+    learning_rate=1e-5 \
+    logging_steps=1
+```
+
+```bash
+bash examples/train_lora/llama3_lora_sft.sh
+```
+
+## Examples
+
+### LoRA Fine-Tuning
+
+#### (Continuous) Pre-Training
+
+```bash
+llamafactory-cli train examples/train_lora/llama3_lora_pretrain.yaml
+```
+
+#### Supervised Fine-Tuning
+
+```bash
+llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
+```
+
+#### Multimodal Supervised Fine-Tuning
+
+```bash
+llamafactory-cli train examples/train_lora/qwen2_5vl_lora_sft.yaml
+```
+
+#### DPO/ORPO/SimPO Training
+
+```bash
+llamafactory-cli train examples/train_lora/llama3_lora_dpo.yaml
+```
+
+#### Multimodal DPO/ORPO/SimPO Training
+
+```bash
+llamafactory-cli train examples/train_lora/qwen2_5vl_lora_dpo.yaml
+```
+
+#### Reward Modeling
+
+```bash
+llamafactory-cli train examples/train_lora/llama3_lora_reward.yaml
+```
+
+#### PPO Training
+
+```bash
+llamafactory-cli train examples/train_lora/llama3_lora_ppo.yaml
+```
+
+#### KTO Training
+
+```bash
+llamafactory-cli train examples/train_lora/llama3_lora_kto.yaml
+```
+
+#### Preprocess Dataset
+
+It is useful for large dataset, use `tokenized_path` in config to load the preprocessed dataset.
+
+```bash
+llamafactory-cli train examples/train_lora/llama3_preprocess.yaml
+```
+
+#### Evaluating on MMLU/CMMLU/C-Eval Benchmarks
+
+```bash
+llamafactory-cli eval examples/train_lora/llama3_lora_eval.yaml
+```
+
+#### Supervised Fine-Tuning on Multiple Nodes
+
+```bash
+FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=0 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
+FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=1 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
+```
+
+#### Supervised Fine-Tuning with DeepSpeed ZeRO-3 (Weight Sharding)
+
+```bash
+FORCE_TORCHRUN=1 llamafactory-cli train examples/train_lora/llama3_lora_sft_ds3.yaml
+```
+
+#### Supervised Fine-Tuning with Ray on 4 GPUs
+
+```bash
+USE_RAY=1 llamafactory-cli train examples/train_lora/llama3_lora_sft_ray.yaml
+```
+
+### QLoRA Fine-Tuning
+
+#### Supervised Fine-Tuning with 4/8-bit Bitsandbytes/HQQ/EETQ Quantization (Recommended)
+
+```bash
+llamafactory-cli train examples/train_qlora/llama3_lora_sft_otfq.yaml
+```
+
+#### Supervised Fine-Tuning with 4-bit Bitsandbytes Quantization on Ascend NPU
+
+```bash
+llamafactory-cli train examples/train_qlora/llama3_lora_sft_bnb_npu.yaml
+```
+
+#### Supervised Fine-Tuning with 4/8-bit GPTQ Quantization
+
+```bash
+llamafactory-cli train examples/train_qlora/llama3_lora_sft_gptq.yaml
+```
+
+#### Supervised Fine-Tuning with 4-bit AWQ Quantization
+
+```bash
+llamafactory-cli train examples/train_qlora/llama3_lora_sft_awq.yaml
+```
+
+#### Supervised Fine-Tuning with 2-bit AQLM Quantization
+
+```bash
+llamafactory-cli train examples/train_qlora/llama3_lora_sft_aqlm.yaml
+```
+
+### Full-Parameter Fine-Tuning
+
+#### Supervised Fine-Tuning on Single Node
+
+```bash
+FORCE_TORCHRUN=1 llamafactory-cli train examples/train_full/llama3_full_sft.yaml
+```
+
+#### Supervised Fine-Tuning on Multiple Nodes
+
+```bash
+FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=0 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft.yaml
+FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=1 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft.yaml
+```
+
+### Elastic and Fault-Tolerant Supervised Fine-Tuning on Multiple Nodes
+
+To launch an elastic job with `MAX_RESTARTS` failures retries, run the following on at least `MIN_NNODES` nodes and at most `MAX_NNODES` nodes. `RDZV_ID` should be set as a unique job id (shared by all nodes participating in the job). See also [torchrun](https://docs.pytorch.org/docs/stable/elastic/run.html).
+
+```bash
+FORCE_TORCHRUN=1 MIN_NNODES=1 MAX_NNODES=3 MAX_RESTARTS=3 RDZV_ID=llamafactory MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft.yaml
+```
+
+#### Multimodal Supervised Fine-Tuning
+
+```bash
+FORCE_TORCHRUN=1 llamafactory-cli train examples/train_full/qwen2_5vl_full_sft.yaml
+```
+
+### Merging LoRA Adapters and Quantization
+
+#### Merge LoRA Adapters
+
+Note: DO NOT use quantized model or `quantization_bit` when merging LoRA adapters.
+
+```bash
+llamafactory-cli export examples/merge_lora/llama3_lora_sft.yaml
+```
+
+#### Quantizing Model using AutoGPTQ
+
+```bash
+llamafactory-cli export examples/merge_lora/llama3_gptq.yaml
+```
+
+### Save Ollama modelfile
+
+```bash
+llamafactory-cli export examples/merge_lora/llama3_full_sft.yaml
+```
+
+### Inferring LoRA Fine-Tuned Models
+
+#### Evaluation using vLLM's Multi-GPU Inference
+
+```
+python scripts/vllm_infer.py --model_name_or_path meta-llama/Meta-Llama-3-8B-Instruct --template llama3 --dataset alpaca_en_demo
+python scripts/eval_bleu_rouge.py generated_predictions.jsonl
+```
+
+#### Use CLI ChatBox
+
+```bash
+llamafactory-cli chat examples/inference/llama3_lora_sft.yaml
+```
+
+#### Use Web UI ChatBox
+
+```bash
+llamafactory-cli webchat examples/inference/llama3_lora_sft.yaml
+```
+
+#### Launch OpenAI-style API
+
+```bash
+llamafactory-cli api examples/inference/llama3_lora_sft.yaml
+```
+
+### Extras
+
+#### Full-Parameter Fine-Tuning using GaLore
+
+```bash
+llamafactory-cli train examples/extras/galore/llama3_full_sft.yaml
+```
+
+#### Full-Parameter Fine-Tuning using APOLLO
+
+```bash
+llamafactory-cli train examples/extras/apollo/llama3_full_sft.yaml
+```
+
+#### Full-Parameter Fine-Tuning using BAdam
+
+```bash
+llamafactory-cli train examples/extras/badam/llama3_full_sft.yaml
+```
+
+#### Full-Parameter Fine-Tuning using Adam-mini
+
+```bash
+llamafactory-cli train examples/extras/adam_mini/qwen2_full_sft.yaml
+```
+
+#### Full-Parameter Fine-Tuning using Muon
+
+```bash
+llamafactory-cli train examples/extras/muon/qwen2_full_sft.yaml
+```
+
+#### LoRA+ Fine-Tuning
+
+```bash
+llamafactory-cli train examples/extras/loraplus/llama3_lora_sft.yaml
+```
+
+#### PiSSA Fine-Tuning
+
+```bash
+llamafactory-cli train examples/extras/pissa/llama3_lora_sft.yaml
+```
+
+#### Mixture-of-Depths Fine-Tuning
+
+```bash
+llamafactory-cli train examples/extras/mod/llama3_full_sft.yaml
+```
+
+#### LLaMA-Pro Fine-Tuning
+
+```bash
+bash examples/extras/llama_pro/expand.sh
+llamafactory-cli train examples/extras/llama_pro/llama3_freeze_sft.yaml
+```
+
+#### FSDP+QLoRA Fine-Tuning
+
+```bash
+bash examples/extras/fsdp_qlora/train.sh
+```

examples/accelerate/fsdp_config.yaml

+compute_environment: LOCAL_MACHINE
+debug: false
+distributed_type: FSDP
+downcast_bf16: 'no'
+fsdp_config:
+  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
+  fsdp_backward_prefetch: BACKWARD_PRE
+  fsdp_forward_prefetch: false
+  fsdp_cpu_ram_efficient_loading: true
+  fsdp_offload_params: false
+  fsdp_sharding_strategy: FULL_SHARD
+  fsdp_state_dict_type: FULL_STATE_DICT
+  fsdp_sync_module_states: true
+  fsdp_use_orig_params: true
+machine_rank: 0
+main_training_function: main
+mixed_precision: bf16  # or fp16
+num_machines: 1  # the number of nodes
+num_processes: 2  # the number of GPUs in all nodes
+rdzv_backend: static
+same_network: true
+tpu_env: []
+tpu_use_cluster: false
+tpu_use_sudo: false
+use_cpu: false