[model] add qwen3 next (#9130)

.github/workflows/tests.yml

@@ -40,6 +40,9 @@ jobs:
           - python: "3.9"
             os: "ubuntu-latest"
             transformers: "4.51.0"
+          - python: "3.9"
+            os: "ubuntu-latest"
+            transformers: "4.53.0"
 
     runs-on: ${{ matrix.os }}
 

README.md

@@ -276,7 +276,7 @@ Choose your path:
 | [Gemma 3/Gemma 3n](https://huggingface.co/google)                 | 270M/1B/4B/6B/8B/12B/27B         | gemma3/gemma3n      |
 | [GLM-4/GLM-4-0414/GLM-Z1](https://huggingface.co/zai-org)         | 9B/32B                           | glm4/glmz1          |
 | [GLM-4.1V](https://huggingface.co/zai-org)                        | 9B                               | glm4v               |
-| [GLM-4.5/GLM-4.5V](https://huggingface.co/zai-org)*               | 106B/355B                        | glm4_moe/glm4v_moe  |
+| [GLM-4.5/GLM-4.5V](https://huggingface.co/zai-org)                | 106B/355B                        | glm4_moe/glm4v_moe  |
 | [GPT-2](https://huggingface.co/openai-community)                  | 0.1B/0.4B/0.8B/1.5B              | -                   |
 | [GPT-OSS](https://huggingface.co/openai)                          | 20B/120B                         | gpt                 |
 | [Granite 3.0-3.3](https://huggingface.co/ibm-granite)             | 1B/2B/3B/8B                      | granite3            |
@@ -296,7 +296,7 @@ Choose your path:
 | [LLaVA-NeXT](https://huggingface.co/llava-hf)                     | 7B/8B/13B/34B/72B/110B           | llava_next          |
 | [LLaVA-NeXT-Video](https://huggingface.co/llava-hf)               | 7B/34B                           | llava_next_video    |
 | [MiMo](https://huggingface.co/XiaomiMiMo)                         | 7B                               | mimo                |
-| [MiniCPM](https://huggingface.co/openbmb)                         | 0.5B/1B/2B/4B/8B                 | cpm/cpm3/cpm4       |
+| [MiniCPM 1-4.1](https://huggingface.co/openbmb)                   | 0.5B/1B/2B/4B/8B                 | cpm/cpm3/cpm4       |
 | [MiniCPM-o-2.6/MiniCPM-V-2.6](https://huggingface.co/openbmb)     | 8B                               | minicpm_o/minicpm_v |
 | [Ministral/Mistral-Nemo](https://huggingface.co/mistralai)        | 8B/12B                           | ministral           |
 | [Mistral/Mixtral](https://huggingface.co/mistralai)               | 7B/8x7B/8x22B                    | mistral             |
@@ -309,11 +309,11 @@ Choose your path:
 | [Phi-4](https://huggingface.co/microsoft)                         | 14B                              | phi4                |
 | [Pixtral](https://huggingface.co/mistralai)                       | 12B                              | pixtral             |
 | [Qwen (1-2.5) (Code/Math/MoE/QwQ)](https://huggingface.co/Qwen)   | 0.5B/1.5B/3B/7B/14B/32B/72B/110B | qwen                |
-| [Qwen3 (MoE/Instruct/Thinking)](https://huggingface.co/Qwen)      | 0.6B/1.7B/4B/8B/14B/32B/235B     | qwen3/qwen3_nothink |
+| [Qwen3 (MoE/Instruct/Thinking/Next)](https://huggingface.co/Qwen) | 0.6B/1.7B/4B/8B/14B/32B/80B/235B | qwen3/qwen3_nothink |
 | [Qwen2-Audio](https://huggingface.co/Qwen)                        | 7B                               | qwen2_audio         |
 | [Qwen2.5-Omni](https://huggingface.co/Qwen)                       | 3B/7B                            | qwen2_omni          |
 | [Qwen2-VL/Qwen2.5-VL/QVQ](https://huggingface.co/Qwen)            | 2B/3B/7B/32B/72B                 | qwen2_vl            |
-| [Seed Coder](https://huggingface.co/ByteDance-Seed)               | 8B                               | seed_coder          |
+| [Seed (OSS/Coder)](https://huggingface.co/ByteDance-Seed)         | 8B/36B                           | seed_oss/seed_coder |
 | [Skywork o1](https://huggingface.co/Skywork)                      | 8B                               | skywork_o1          |
 | [StarCoder 2](https://huggingface.co/bigcode)                     | 3B/7B/15B                        | -                   |
 | [TeleChat2](https://huggingface.co/Tele-AI)                       | 3B/7B/35B/115B                   | telechat2           |

README_zh.md

@@ -278,7 +278,7 @@ https://github.com/user-attachments/assets/43b700c6-a178-41db-b1f8-8190a5d3fcfc
 | [Gemma 3/Gemma 3n](https://huggingface.co/google)                 | 270M/1B/4B/6B/8B/12B/27B         | gemma3/gemma3n      |
 | [GLM-4/GLM-4-0414/GLM-Z1](https://huggingface.co/zai-org)         | 9B/32B                           | glm4/glmz1          |
 | [GLM-4.1V](https://huggingface.co/zai-org)                        | 9B                               | glm4v               |
-| [GLM-4.5/GLM-4.5V](https://huggingface.co/zai-org)*               | 106B/355B                        | glm4_moe/glm4v_moe  |
+| [GLM-4.5/GLM-4.5V](https://huggingface.co/zai-org)                | 106B/355B                        | glm4_moe/glm4v_moe  |
 | [GPT-2](https://huggingface.co/openai-community)                  | 0.1B/0.4B/0.8B/1.5B              | -                   |
 | [GPT-OSS](https://huggingface.co/openai)                          | 20B/120B                         | gpt                 |
 | [Granite 3.0-3.3](https://huggingface.co/ibm-granite)             | 1B/2B/3B/8B                      | granite3            |
@@ -298,7 +298,7 @@ https://github.com/user-attachments/assets/43b700c6-a178-41db-b1f8-8190a5d3fcfc
 | [LLaVA-NeXT](https://huggingface.co/llava-hf)                     | 7B/8B/13B/34B/72B/110B           | llava_next          |
 | [LLaVA-NeXT-Video](https://huggingface.co/llava-hf)               | 7B/34B                           | llava_next_video    |
 | [MiMo](https://huggingface.co/XiaomiMiMo)                         | 7B                               | mimo                |
-| [MiniCPM](https://huggingface.co/openbmb)                         | 0.5B/1B/2B/4B/8B                 | cpm/cpm3/cpm4       |
+| [MiniCPM 1-4.1](https://huggingface.co/openbmb)                   | 0.5B/1B/2B/4B/8B                 | cpm/cpm3/cpm4       |
 | [MiniCPM-o-2.6/MiniCPM-V-2.6](https://huggingface.co/openbmb)     | 8B                               | minicpm_o/minicpm_v |
 | [Ministral/Mistral-Nemo](https://huggingface.co/mistralai)        | 8B/12B                           | ministral           |
 | [Mistral/Mixtral](https://huggingface.co/mistralai)               | 7B/8x7B/8x22B                    | mistral             |
@@ -311,11 +311,11 @@ https://github.com/user-attachments/assets/43b700c6-a178-41db-b1f8-8190a5d3fcfc
 | [Phi-4](https://huggingface.co/microsoft)                         | 14B                              | phi4                |
 | [Pixtral](https://huggingface.co/mistralai)                       | 12B                              | pixtral             |
 | [Qwen (1-2.5) (Code/Math/MoE/QwQ)](https://huggingface.co/Qwen)   | 0.5B/1.5B/3B/7B/14B/32B/72B/110B | qwen                |
-| [Qwen3 (MoE/Instruct/Thinking)](https://huggingface.co/Qwen)      | 0.6B/1.7B/4B/8B/14B/32B/235B     | qwen3/qwen3_nothink |
+| [Qwen3 (MoE/Instruct/Thinking/Next)](https://huggingface.co/Qwen) | 0.6B/1.7B/4B/8B/14B/32B/80B/235B | qwen3/qwen3_nothink |
 | [Qwen2-Audio](https://huggingface.co/Qwen)                        | 7B                               | qwen2_audio         |
 | [Qwen2.5-Omni](https://huggingface.co/Qwen)                       | 3B/7B                            | qwen2_omni          |
 | [Qwen2-VL/Qwen2.5-VL/QVQ](https://huggingface.co/Qwen)            | 2B/3B/7B/32B/72B                 | qwen2_vl            |
-| [Seed Coder](https://huggingface.co/ByteDance-Seed)               | 8B                               | seed_coder          |
+| [Seed (OSS/Coder)](https://huggingface.co/ByteDance-Seed)         | 8B/36B                           | seed_oss/seed_coder |
 | [Skywork o1](https://huggingface.co/Skywork)                      | 8B                               | skywork_o1          |
 | [StarCoder 2](https://huggingface.co/bigcode)                     | 3B/7B/15B                        | -                   |
 | [TeleChat2](https://huggingface.co/Tele-AI)                       | 3B/7B/35B/115B                   | telechat2           |

requirements.txt

@@ -1,12 +1,11 @@
 # core deps
-transformers>=4.49.0,<=4.55.0,!=4.52.0
+transformers>=4.49.0,<=4.56.1,!=4.52.0
-datasets>=2.16.0,<=3.6.0
+datasets>=2.16.0,<=4.0.0
-accelerate>=1.3.0,<=1.7.0
+accelerate>=1.3.0,<=1.10.1
-peft>=0.14.0,<=0.15.2
+peft>=0.14.0,<=0.17.1
 trl>=0.8.6,<=0.9.6
-tokenizers>=0.19.0,<=0.21.1
 # gui
-gradio>=4.38.0,<=5.42.0
+gradio>=4.38.0,<=5.45.0
 matplotlib>=3.7.0
 tyro<0.9.0
 # ops

src/llamafactory/data/converter.py

@@ -11,7 +11,7 @@
 # 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
 from abc import abstractmethod
 from dataclasses import dataclass
@@ -227,9 +227,150 @@ class SharegptDatasetConverter(DatasetConverter):
         return output
 
 
+@dataclass
+class OpenAIDatasetConverter(DatasetConverter):
+    def __call__(self, example: dict[str, Any]) -> dict[str, Any]:
+        tag_mapping = {
+            self.dataset_attr.user_tag: Role.USER.value,
+            self.dataset_attr.assistant_tag: Role.ASSISTANT.value,
+            self.dataset_attr.observation_tag: Role.OBSERVATION.value,
+            self.dataset_attr.function_tag: Role.FUNCTION.value,
+            self.dataset_attr.system_tag: Role.SYSTEM.value,
+        }
+
+        messages = example[self.dataset_attr.messages]
+        if (
+            self.dataset_attr.system_tag
+            and len(messages) != 0
+            and messages[0][self.dataset_attr.role_tag] == self.dataset_attr.system_tag
+        ):
+            system = messages[0][self.dataset_attr.content_tag]
+            messages = messages[1:]
+        else:
+            system = example.get(self.dataset_attr.system, "") if self.dataset_attr.system else ""
+
+        aligned_messages = []
+        tool_responses = []
+        broken_data = False
+        for turn_idx, message in enumerate(messages):
+            role = message[self.dataset_attr.role_tag]
+            content = message[self.dataset_attr.content_tag]
+
+            if role in [self.dataset_attr.assistant_tag, self.dataset_attr.function_tag]:
+                if "tool_calls" in message and len(message["tool_calls"]) > 0:
+                    tool_calls_list = [tool["function"] for tool in message["tool_calls"]]
+                    content = json.dumps(tool_calls_list, ensure_ascii=False)
+                    role = self.dataset_attr.function_tag
+
+            if role == self.dataset_attr.observation_tag:
+                tool_responses.append(content)
+                continue
+            elif len(tool_responses) > 0:
+                _content = "\n</tool_response>\n<tool_response>\n".join(tool_responses)
+                aligned_messages.append(
+                    {
+                        "role": Role.OBSERVATION.value,
+                        "content": _content,
+                    }
+                )
+                tool_responses = []
+
+            aligned_messages.append(
+                {
+                    "role": tag_mapping[role],
+                    "content": content,
+                }
+            )
+
+        odd_tags = (Role.USER.value, Role.OBSERVATION.value)
+        even_tags = (Role.ASSISTANT.value, Role.FUNCTION.value)
+        accept_tags = (odd_tags, even_tags)
+        for turn_idx, message in enumerate(aligned_messages):
+            if message["role"] not in accept_tags[turn_idx % 2]:
+                logger.warning_rank0(f"Invalid role tag in {messages}.")
+                broken_data = True
+                break
+
+        if (not self.dataset_attr.ranking and len(aligned_messages) % 2 != 0) or (
+            self.dataset_attr.ranking and len(aligned_messages) % 2 == 0
+        ):
+            logger.warning_rank0(f"Invalid message count in {messages}.")
+            broken_data = True
+
+        if broken_data:
+            logger.warning_rank0("Skipping this abnormal example.")
+            prompt, response = [], []
+        elif self.dataset_attr.kto_tag and isinstance(example[self.dataset_attr.kto_tag], bool):  # kto example
+            prompt = aligned_messages[:-1]
+            response = aligned_messages[-1:]
+            if example[self.dataset_attr.kto_tag]:
+                response = response + [{"role": Role.ASSISTANT.value, "content": ""}]
+            else:
+                response = [{"role": Role.ASSISTANT.value, "content": ""}] + response
+        elif (
+            self.dataset_attr.ranking
+            and isinstance(example[self.dataset_attr.chosen], dict)
+            and isinstance(example[self.dataset_attr.rejected], dict)
+        ):  # pairwise example
+            chosen = example[self.dataset_attr.chosen]
+            rejected = example[self.dataset_attr.rejected]
+            if (
+                chosen[self.dataset_attr.role_tag] not in accept_tags[-1]
+                or rejected[self.dataset_attr.role_tag] not in accept_tags[-1]
+            ):
+                logger.warning_rank0(f"Invalid role tag in {[chosen, rejected]}.")
+                broken_data = True
+
+            prompt = aligned_messages
+            response = [
+                {
+                    "role": tag_mapping[chosen[self.dataset_attr.role_tag]],
+                    "content": chosen[self.dataset_attr.content_tag],
+                },
+                {
+                    "role": tag_mapping[rejected[self.dataset_attr.role_tag]],
+                    "content": rejected[self.dataset_attr.content_tag],
+                },
+            ]
+        else:  # normal example
+            prompt = aligned_messages[:-1]
+            response = aligned_messages[-1:]
+
+        tools = example.get(self.dataset_attr.tools, "") if self.dataset_attr.tools else ""
+        if isinstance(tools, dict) or isinstance(tools, list):
+            tools = json.dumps(tools, ensure_ascii=False)
+
+        short_system_prompt = "detailed thinking off"
+        if not system:
+            if not tools:
+                system = short_system_prompt
+            else:
+                pass
+        else:
+            if not tools:
+                if "detailed thinking on" in system or "detailed thinking off" in system:
+                    pass
+                else:
+                    system += "\n" + short_system_prompt
+            else:
+                system += "\n"
+
+        output = {
+            "_prompt": prompt,
+            "_response": response,
+            "_system": system,
+            "_tools": tools,
+            "_images": self._find_medias(example[self.dataset_attr.images]) if self.dataset_attr.images else None,
+            "_videos": self._find_medias(example[self.dataset_attr.videos]) if self.dataset_attr.videos else None,
+            "_audios": self._find_medias(example[self.dataset_attr.audios]) if self.dataset_attr.audios else None,
+        }
+        return output
+
+
 DATASET_CONVERTERS = {
     "alpaca": AlpacaDatasetConverter,
     "sharegpt": SharegptDatasetConverter,
+    "openai": OpenAIDatasetConverter,
 }
 
 

src/llamafactory/data/template.py

@@ -679,6 +679,23 @@ register_template(
 )
 
 
+register_template(
+    name="bailing_v2",
+    format_user=StringFormatter(slots=["<role>HUMAN</role>{{content}}<|role_end|><role>ASSISTANT</role>"]),
+    format_system=StringFormatter(slots=["<role>SYSTEM</role>{{content}}<|role_end|>"]),
+    format_assistant=StringFormatter(slots=["{{content}}<|role_end|>"]),
+    format_observation=StringFormatter(
+        slots=[
+            "<role>OBSERVATION</role>\n<tool_response>\n{{content}}\n</tool_response><|role_end|><role>ASSISTANT</role>"
+        ]
+    ),
+    format_function=FunctionFormatter(slots=["{{content}}<|role_end|>"], tool_format="ling"),
+    format_tools=ToolFormatter(tool_format="ling"),
+    stop_words=["<|endoftext|>"],
+    efficient_eos=True,
+)
+
+
 register_template(
     name="belle",
     format_user=StringFormatter(slots=["Human: {{content}}\n\nBelle: "]),
@@ -900,6 +917,18 @@ register_template(
 )
 
 
+# copied from chatml template
+register_template(
+    name="ernie",
+    format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n\n<|im_start|>assistant\n"]),
+    format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\n\n"]),
+    format_system=StringFormatter(slots=["<|im_start|>system\n{{content}}<|im_end|>\n\n"]),
+    format_observation=StringFormatter(slots=["<|im_start|>tool\n{{content}}<|im_end|>\n\n<|im_start|>assistant\n"]),
+    default_system="<global_setting>\nthink_mode=True\n</global_setting>",
+    stop_words=["<|im_end|>"],
+)
+
+
 register_template(
     name="exaone",
     format_user=StringFormatter(slots=["[|user|]{{content}}\n[|assistant|]"]),

src/llamafactory/data/tool_utils.py

@@ -78,6 +78,14 @@ SEED_TOOL_PROMPT = (
     "lines</parameter>\n</function>\n</seed:tool_call>\n"
 )
 
+LING_TOOL_PROMPT = (
+    "# Tools\n\nYou may call one or more functions to assist with the user query.\n\n"
+    "You are provided with function signatures within <tools></tools> XML tags:\n<tools>{tool_text}"
+    "\n</tools>\n\nFor each function call, return a json object with function name and arguments within "
+    """<tool_call></tool_call> XML tags:\n<tool_call>\n{{"name": <function-name>, """
+    """"arguments": <args-json-object>}}\n</tool_call>"""
+)
+
 
 @dataclass
 class ToolUtils(ABC):
@@ -406,6 +414,20 @@ class SeedToolUtils(ToolUtils):
         return results
 
 
+class LingToolUtils(QwenToolUtils):
+    r"""Ling v2 tool using template."""
+
+    @override
+    @staticmethod
+    def tool_formatter(tools: list[dict[str, Any]]) -> str:
+        tool_text = ""
+        for tool in tools:
+            wrapped_tool = tool if tool.get("type") == "function" else {"type": "function", "function": tool}
+            tool_text += "\n" + json.dumps(wrapped_tool, ensure_ascii=False)
+
+        return LING_TOOL_PROMPT.format(tool_text=tool_text) + "\n" + "detailed thinking off"
+
+
 TOOLS = {
     "default": DefaultToolUtils(),
     "glm4": GLM4ToolUtils(),
@@ -414,6 +436,7 @@ TOOLS = {
     "qwen": QwenToolUtils(),
     "glm4_moe": GLM4MOEToolUtils(),
     "seed_oss": SeedToolUtils(),
+    "ling": LingToolUtils(),
 }
 
 

src/llamafactory/extras/constants.py

@@ -601,6 +601,17 @@ register_model_group(
 )
 
 
+register_model_group(
+    models={
+        "ERNIE-4.5-21B-A3B-Thinking": {
+            DownloadSource.DEFAULT: "baidu/ERNIE-4.5-21B-A3B-Thinking",
+            DownloadSource.MODELSCOPE: "PaddlePaddle/ERNIE-4.5-21B-A3B-Thinking",
+        },
+    },
+    template="ernie",
+)
+
+
 register_model_group(
     models={
         "EXAONE-3.0-7.8B-Instruct": {
@@ -1783,6 +1794,10 @@ register_model_group(
             DownloadSource.DEFAULT: "openbmb/MiniCPM4-8B",
             DownloadSource.MODELSCOPE: "OpenBMB/MiniCPM4-8B",
         },
+        "MiniCPM4.1-8B-Chat": {
+            DownloadSource.DEFAULT: "openbmb/MiniCPM4.1-8B",
+            DownloadSource.MODELSCOPE: "OpenBMB/MiniCPM4.1-8B",
+        },
     },
     template="cpm4",
 )
@@ -1790,7 +1805,7 @@ register_model_group(
 
 register_model_group(
     models={
-        "MiniCPM-o-2_6": {
+        "MiniCPM-o-2.6": {
             DownloadSource.DEFAULT: "openbmb/MiniCPM-o-2_6",
             DownloadSource.MODELSCOPE: "OpenBMB/MiniCPM-o-2_6",
         },
@@ -1802,7 +1817,7 @@ register_model_group(
 
 register_model_group(
     models={
-        "MiniCPM-V-2_6": {
+        "MiniCPM-V-2.6": {
             DownloadSource.DEFAULT: "openbmb/MiniCPM-V-2_6",
             DownloadSource.MODELSCOPE: "OpenBMB/MiniCPM-V-2_6",
         },
@@ -1826,7 +1841,7 @@ register_model_group(
 
 register_model_group(
     models={
-        "MiniCPM-V-4_5": {
+        "MiniCPM-V-4.5": {
             DownloadSource.DEFAULT: "openbmb/MiniCPM-V-4_5",
             DownloadSource.MODELSCOPE: "OpenBMB/MiniCPM-V-4_5",
         },
@@ -1944,6 +1959,37 @@ register_model_group(
 )
 
 
+register_model_group(
+    models={
+        "MobileLLM-R1-140M-Base": {
+            DownloadSource.DEFAULT: "facebook/MobileLLM-R1-140M-base",
+            DownloadSource.MODELSCOPE: "facebook/MobileLLM-R1-140M-base",
+        },
+        "MobileLLM-R1-360M-Base": {
+            DownloadSource.DEFAULT: "facebook/MobileLLM-R1-360M-base",
+            DownloadSource.MODELSCOPE: "facebook/MobileLLM-R1-360M-base",
+        },
+        "MobileLLM-R1-950M-Base": {
+            DownloadSource.DEFAULT: "facebook/MobileLLM-R1-950M-base",
+            DownloadSource.MODELSCOPE: "facebook/MobileLLM-R1-950M-base",
+        },
+        "MobileLLM-R1-140M-Instruct": {
+            DownloadSource.DEFAULT: "facebook/MobileLLM-R1-140M",
+            DownloadSource.MODELSCOPE: "facebook/MobileLLM-R1-140M",
+        },
+        "MobileLLM-R1-360M-Instruct": {
+            DownloadSource.DEFAULT: "facebook/MobileLLM-R1-360M",
+            DownloadSource.MODELSCOPE: "facebook/MobileLLM-R1-360M",
+        },
+        "MobileLLM-R1-950M-Instruct": {
+            DownloadSource.DEFAULT: "facebook/MobileLLM-R1-950M",
+            DownloadSource.MODELSCOPE: "facebook/MobileLLM-R1-950M",
+        },
+    },
+    template="llama3",
+)
+
+
 register_model_group(
     models={
         "Moonlight-16B-A3B": {
@@ -2912,6 +2958,10 @@ register_model_group(
             DownloadSource.DEFAULT: "Qwen/Qwen3-235B-A22B-GPTQ-Int4",
             DownloadSource.MODELSCOPE: "Qwen/Qwen3-235B-A22B-GPTQ-Int4",
         },
+        "Qwen/Qwen3-Next-80B-A3B-Thinking": {
+            DownloadSource.DEFAULT: "Qwen/Qwen3-Next-80B-A3B-Thinking",
+            DownloadSource.MODELSCOPE: "Qwen/Qwen3-Next-80B-A3B-Thinking",
+        },
     },
     template="qwen3",
 )
@@ -2931,6 +2981,10 @@ register_model_group(
             DownloadSource.DEFAULT: "Qwen/Qwen3-235B-A22B-Instruct-2507",
             DownloadSource.MODELSCOPE: "Qwen/Qwen3-235B-A22B-Instruct-2507",
         },
+        "Qwen3-Next-80B-A3B-Instruct": {
+            DownloadSource.DEFAULT: "Qwen/Qwen3-Next-80B-A3B-Instruct",
+            DownloadSource.MODELSCOPE: "Qwen/Qwen3-Next-80B-A3B-Instruct",
+        },
     },
     template="qwen3_nothink",
 )

src/llamafactory/extras/misc.py

@@ -94,10 +94,10 @@ def check_version(requirement: str, mandatory: bool = False) -> None:
 
 def check_dependencies() -> None:
     r"""Check the version of the required packages."""
-    check_version("transformers>=4.49.0,<=4.55.0")
+    check_version("transformers>=4.49.0,<=4.56.1")
-    check_version("datasets>=2.16.0,<=3.6.0")
+    check_version("datasets>=2.16.0,<=4.0.0")
-    check_version("accelerate>=1.3.0,<=1.7.0")
+    check_version("accelerate>=1.3.0,<=1.10.1")
-    check_version("peft>=0.14.0,<=0.15.2")
+    check_version("peft>=0.14.0,<=0.17.1")
     check_version("trl>=0.8.6,<=0.9.6")
 
 

src/llamafactory/extras/packages.py

@@ -58,6 +58,10 @@ def is_apollo_available():
     return _is_package_available("apollo_torch")
 
 
+def is_jieba_available():
+    return _is_package_available("jieba")
+
+
 def is_gradio_available():
     return _is_package_available("gradio")
 

src/llamafactory/model/patcher.py

@@ -211,9 +211,23 @@ def patch_valuehead_model(model: "AutoModelForCausalLMWithValueHead") -> None:
         if isinstance(self.pretrained_model, PeftModel):
             self.pretrained_model.create_or_update_model_card(output_dir)
 
+    def get_rope_index_func(self: "AutoModelForCausalLMWithValueHead"):
+        if isinstance(self.pretrained_model, PeftModel):
+            base_model = self.pretrained_model.base_model.model
+        else:
+            base_model = self.pretrained_model
+
+        if base_model and hasattr(base_model, "get_rope_index"):
+            return base_model.get_rope_index
+        elif base_model and hasattr(base_model, "model") and hasattr(base_model.model, "get_rope_index"):
+            return base_model.model.get_rope_index
+        else:
+            return None
+
     ignore_modules = [name for name, _ in model.named_parameters() if "pretrained_model" in name]
     setattr(model, "_keys_to_ignore_on_save", ignore_modules)
     setattr(model, "tie_weights", MethodType(tie_weights, model))
     setattr(model, "get_input_embeddings", MethodType(get_input_embeddings, model))
     setattr(model, "get_output_embeddings", MethodType(get_output_embeddings, model))
+    setattr(model, "get_rope_index", get_rope_index_func(model))
     setattr(model, "create_or_update_model_card", MethodType(create_or_update_model_card, model))

src/llamafactory/train/sft/metric.py

@@ -21,11 +21,11 @@ from typing import TYPE_CHECKING, Optional
 
 import numpy as np
 import torch
-from transformers.utils import is_jieba_available, is_nltk_available
+from transformers.utils import is_nltk_available
 
 from ...extras.constants import IGNORE_INDEX
 from ...extras.misc import numpify
-from ...extras.packages import is_rouge_available
+from ...extras.packages import is_jieba_available, is_rouge_available
 
 
 if TYPE_CHECKING:

src/llamafactory/webui/locales.py

@@ -34,31 +34,36 @@ LOCALES = {
         "en": {
             "value": (
                 "<h3><center>Visit <a href='https://github.com/hiyouga/LLaMA-Factory' target='_blank'>"
-                "GitHub Page</a></center></h3>"
+                "GitHub Page</a> <a href='https://llamafactory.readthedocs.io/en/latest/' target='_blank'>"
+                "Documentation</a></center></h3>"
             ),
         },
         "ru": {
             "value": (
                 "<h3><center>Посетить <a href='https://github.com/hiyouga/LLaMA-Factory' target='_blank'>"
-                "страницу GitHub</a></center></h3>"
+                "страницу GitHub</a> <a href='https://llamafactory.readthedocs.io/en/latest/' target='_blank'>"
+                "Документацию</a></center></h3>"
             ),
         },
         "zh": {
             "value": (
                 "<h3><center>访问 <a href='https://github.com/hiyouga/LLaMA-Factory' target='_blank'>"
-                "GitHub 主页</a></center></h3>"
+                "GitHub 主页</a> <a href='https://llamafactory.readthedocs.io/zh-cn/latest/' target='_blank'>"
+                "官方文档</a></center></h3>"
             ),
         },
         "ko": {
             "value": (
                 "<h3><center><a href='https://github.com/hiyouga/LLaMA-Factory' target='_blank'>"
-                "GitHub 페이지</a>를 방문하세요.</center></h3>"
+                "GitHub 페이지</a> <a href='https://llamafactory.readthedocs.io/en/latest/' target='_blank'>"
+                "공식 문서</a>를 방문하세요.</center></h3>"
             ),
         },
         "ja": {
             "value": (
                 "<h3><center><a href='https://github.com/hiyouga/LLaMA-Factory' target='_blank'>"
-                "GitHub ページ</a>にアクセスする</center></h3>"
+                "GitHub ページ</a> <a href='https://llamafactory.readthedocs.io/en/latest/' target='_blank'>"
+                "ドキュメント</a>にアクセスする</center></h3>"
             ),
         },
     },