423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-09-12 16:12:48 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'main' into cpei/refactor

Browse Source 
Former-commit-id: 2c6262c3cd57906806e53aecd4e5a4423399c1e0
This commit is contained in:
hoshi-hiyouga 2024-10-08 17:31:17 +08:00 committed by GitHub
commit c639e52c6b
19 changed files with 775 additions and 240 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

61
README.md
View File

@ -30,6 +30,9 @@ Choose your path:
- **Local machine**: Please refer to [usage](#getting-started) - **Local machine**: Please refer to [usage](#getting-started)
- **Documentation (WIP)**: https://llamafactory.readthedocs.io/zh-cn/latest/ - **Documentation (WIP)**: https://llamafactory.readthedocs.io/zh-cn/latest/
> [!NOTE]
> Except for the above links, all other websites are unauthorized third-party websites. Please carefully use them.
## Table of Contents ## Table of Contents
- [Features](#features) - [Features](#features)
@ -162,34 +165,36 @@ Compared to ChatGLM's [P-Tuning](https://github.com/THUDM/ChatGLM2-6B/tree/main/
## Supported Models ## Supported Models
| Model | Model size | Template | | Model | Model size | Template |
| ----------------------------------------------------------------- | -------------------------------- | --------- | | ----------------------------------------------------------------- | -------------------------------- | ---------------- |
| [Baichuan 2](https://huggingface.co/baichuan-inc) | 7B/13B | baichuan2 | | [Baichuan 2](https://huggingface.co/baichuan-inc) | 7B/13B | baichuan2 |
| [BLOOM/BLOOMZ](https://huggingface.co/bigscience) | 560M/1.1B/1.7B/3B/7.1B/176B | - | | [BLOOM/BLOOMZ](https://huggingface.co/bigscience) | 560M/1.1B/1.7B/3B/7.1B/176B | - |
| [ChatGLM3](https://huggingface.co/THUDM) | 6B | chatglm3 | | [ChatGLM3](https://huggingface.co/THUDM) | 6B | chatglm3 |
| [Command R](https://huggingface.co/CohereForAI) | 35B/104B | cohere | | [Command R](https://huggingface.co/CohereForAI) | 35B/104B | cohere |
| [DeepSeek (Code/MoE)](https://huggingface.co/deepseek-ai) | 7B/16B/67B/236B | deepseek | | [DeepSeek (Code/MoE)](https://huggingface.co/deepseek-ai) | 7B/16B/67B/236B | deepseek |
| [Falcon](https://huggingface.co/tiiuae) | 7B/11B/40B/180B | falcon | | [Falcon](https://huggingface.co/tiiuae) | 7B/11B/40B/180B | falcon |
| [Gemma/Gemma 2/CodeGemma](https://huggingface.co/google) | 2B/7B/9B/27B | gemma | | [Gemma/Gemma 2/CodeGemma](https://huggingface.co/google) | 2B/7B/9B/27B | gemma |
| [GLM-4](https://huggingface.co/THUDM) | 9B | glm4 | | [GLM-4](https://huggingface.co/THUDM) | 9B | glm4 |
| [InternLM2/InternLM2.5](https://huggingface.co/internlm) | 7B/20B | intern2 | | [InternLM2/InternLM2.5](https://huggingface.co/internlm) | 7B/20B | intern2 |
| [Llama](https://github.com/facebookresearch/llama) | 7B/13B/33B/65B | - | | [Llama](https://github.com/facebookresearch/llama) | 7B/13B/33B/65B | - |
| [Llama 2](https://huggingface.co/meta-llama) | 7B/13B/70B | llama2 | | [Llama 2](https://huggingface.co/meta-llama) | 7B/13B/70B | llama2 |
| [Llama 3/Llama 3.1](https://huggingface.co/meta-llama) | 8B/70B | llama3 | | [Llama 3-3.2](https://huggingface.co/meta-llama) | 1B/3B/8B/70B | llama3 |
| [LLaVA-1.5](https://huggingface.co/llava-hf) | 7B/13B | llava | | [LLaVA-1.5](https://huggingface.co/llava-hf) | 7B/13B | llava |
| [MiniCPM](https://huggingface.co/openbmb) | 1B/2B/4B | cpm/cpm3 | | [LLaVA-NeXT](https://huggingface.co/llava-hf) | 7B/8B/13B/34B/72B/110B | llava_next |
| [Mistral/Mixtral](https://huggingface.co/mistralai) | 7B/8x7B/8x22B | mistral | | [LLaVA-NeXT-Video](https://huggingface.co/llava-hf) | 7B/34B | llava_next_video |
| [OLMo](https://huggingface.co/allenai) | 1B/7B | - | | [MiniCPM](https://huggingface.co/openbmb) | 1B/2B/4B | cpm/cpm3 |
| [PaliGemma](https://huggingface.co/google) | 3B | paligemma | | [Mistral/Mixtral](https://huggingface.co/mistralai) | 7B/8x7B/8x22B | mistral |
| [Phi-1.5/Phi-2](https://huggingface.co/microsoft) | 1.3B/2.7B | - | | [OLMo](https://huggingface.co/allenai) | 1B/7B | - |
| [Phi-3](https://huggingface.co/microsoft) | 4B/7B/14B | phi | | [PaliGemma](https://huggingface.co/google) | 3B | paligemma |
| [Qwen (1-2.5) (Code/Math/MoE)](https://huggingface.co/Qwen) | 0.5B/1.5B/3B/7B/14B/32B/72B/110B | qwen | | [Phi-1.5/Phi-2](https://huggingface.co/microsoft) | 1.3B/2.7B | - |
| [Qwen2-VL](https://huggingface.co/Qwen) | 2B/7B/72B | qwen2_vl | | [Phi-3](https://huggingface.co/microsoft) | 4B/7B/14B | phi |
| [StarCoder 2](https://huggingface.co/bigcode) | 3B/7B/15B | - | | [Qwen (1-2.5) (Code/Math/MoE)](https://huggingface.co/Qwen) | 0.5B/1.5B/3B/7B/14B/32B/72B/110B | qwen |
| [XVERSE](https://huggingface.co/xverse) | 7B/13B/65B | xverse | | [Qwen2-VL](https://huggingface.co/Qwen) | 2B/7B/72B | qwen2_vl |
| [Yi/Yi-1.5 (Code)](https://huggingface.co/01-ai) | 1.5B/6B/9B/34B | yi | | [StarCoder 2](https://huggingface.co/bigcode) | 3B/7B/15B | - |
| [Yi-VL](https://huggingface.co/01-ai) | 6B/34B | yi_vl | | [XVERSE](https://huggingface.co/xverse) | 7B/13B/65B | xverse |
| [Yuan 2](https://huggingface.co/IEITYuan) | 2B/51B/102B | yuan | | [Yi/Yi-1.5 (Code)](https://huggingface.co/01-ai) | 1.5B/6B/9B/34B | yi |
| [Yi-VL](https://huggingface.co/01-ai) | 6B/34B | yi_vl |
| [Yuan 2](https://huggingface.co/IEITYuan) | 2B/51B/102B | yuan |
> [!NOTE] > [!NOTE]
> For the "base" models, the `template` argument can be chosen from `default`, `alpaca`, `vicuna` etc. But make sure to use the **corresponding template** for the "instruct/chat" models. > For the "base" models, the `template` argument can be chosen from `default`, `alpaca`, `vicuna` etc. But make sure to use the **corresponding template** for the "instruct/chat" models.

61
README_zh.md
View File

@ -31,6 +31,9 @@ https://github.com/user-attachments/assets/e6ce34b0-52d5-4f3e-a830-592106c4c272
- **入门教程**：https://zhuanlan.zhihu.com/p/695287607 - **入门教程**：https://zhuanlan.zhihu.com/p/695287607
- **框架文档**：https://llamafactory.readthedocs.io/zh-cn/latest/ - **框架文档**：https://llamafactory.readthedocs.io/zh-cn/latest/
> [!NOTE]
> 除上述链接以外的其他网站均为未经许可的第三方网站，请小心甄别。
## 目录 ## 目录
- [项目特色](#项目特色) - [项目特色](#项目特色)
@ -163,34 +166,36 @@ https://github.com/user-attachments/assets/e6ce34b0-52d5-4f3e-a830-592106c4c272
## 模型 ## 模型
| 模型名 | 模型大小 | Template | | 模型名 | 模型大小 | Template |
| ----------------------------------------------------------------- | -------------------------------- | --------- | | ----------------------------------------------------------------- | -------------------------------- | ---------------- |
| [Baichuan 2](https://huggingface.co/baichuan-inc) | 7B/13B | baichuan2 | | [Baichuan 2](https://huggingface.co/baichuan-inc) | 7B/13B | baichuan2 |
| [BLOOM/BLOOMZ](https://huggingface.co/bigscience) | 560M/1.1B/1.7B/3B/7.1B/176B | - | | [BLOOM/BLOOMZ](https://huggingface.co/bigscience) | 560M/1.1B/1.7B/3B/7.1B/176B | - |
| [ChatGLM3](https://huggingface.co/THUDM) | 6B | chatglm3 | | [ChatGLM3](https://huggingface.co/THUDM) | 6B | chatglm3 |
| [Command R](https://huggingface.co/CohereForAI) | 35B/104B | cohere | | [Command R](https://huggingface.co/CohereForAI) | 35B/104B | cohere |
| [DeepSeek (Code/MoE)](https://huggingface.co/deepseek-ai) | 7B/16B/67B/236B | deepseek | | [DeepSeek (Code/MoE)](https://huggingface.co/deepseek-ai) | 7B/16B/67B/236B | deepseek |
| [Falcon](https://huggingface.co/tiiuae) | 7B/11B/40B/180B | falcon | | [Falcon](https://huggingface.co/tiiuae) | 7B/11B/40B/180B | falcon |
| [Gemma/Gemma 2/CodeGemma](https://huggingface.co/google) | 2B/7B/9B/27B | gemma | | [Gemma/Gemma 2/CodeGemma](https://huggingface.co/google) | 2B/7B/9B/27B | gemma |
| [GLM-4](https://huggingface.co/THUDM) | 9B | glm4 | | [GLM-4](https://huggingface.co/THUDM) | 9B | glm4 |
| [InternLM2/InternLM2.5](https://huggingface.co/internlm) | 7B/20B | intern2 | | [InternLM2/InternLM2.5](https://huggingface.co/internlm) | 7B/20B | intern2 |
| [Llama](https://github.com/facebookresearch/llama) | 7B/13B/33B/65B | - | | [Llama](https://github.com/facebookresearch/llama) | 7B/13B/33B/65B | - |
| [Llama 2](https://huggingface.co/meta-llama) | 7B/13B/70B | llama2 | | [Llama 2](https://huggingface.co/meta-llama) | 7B/13B/70B | llama2 |
| [Llama 3/Llama 3.1](https://huggingface.co/meta-llama) | 8B/70B | llama3 | | [Llama 3-3.2](https://huggingface.co/meta-llama) | 1B/3B/8B/70B | llama3 |
| [LLaVA-1.5](https://huggingface.co/llava-hf) | 7B/13B | llava | | [LLaVA-1.5](https://huggingface.co/llava-hf) | 7B/13B | llava |
| [MiniCPM](https://huggingface.co/openbmb) | 1B/2B/4B | cpm/cpm3 | | [LLaVA-NeXT](https://huggingface.co/llava-hf) | 7B/8B/13B/34B/72B/110B | llava_next |
| [Mistral/Mixtral](https://huggingface.co/mistralai) | 7B/8x7B/8x22B | mistral | | [LLaVA-NeXT-Video](https://huggingface.co/llava-hf) | 7B/34B | llava_next_video |
| [OLMo](https://huggingface.co/allenai) | 1B/7B | - | | [MiniCPM](https://huggingface.co/openbmb) | 1B/2B/4B | cpm/cpm3 |
| [PaliGemma](https://huggingface.co/google) | 3B | paligemma | | [Mistral/Mixtral](https://huggingface.co/mistralai) | 7B/8x7B/8x22B | mistral |
| [Phi-1.5/Phi-2](https://huggingface.co/microsoft) | 1.3B/2.7B | - | | [OLMo](https://huggingface.co/allenai) | 1B/7B | - |
| [Phi-3](https://huggingface.co/microsoft) | 4B/7B/14B | phi | | [PaliGemma](https://huggingface.co/google) | 3B | paligemma |
| [Qwen (1-2.5) (Code/Math/MoE)](https://huggingface.co/Qwen) | 0.5B/1.5B/3B/7B/14B/32B/72B/110B | qwen | | [Phi-1.5/Phi-2](https://huggingface.co/microsoft) | 1.3B/2.7B | - |
| [Qwen2-VL](https://huggingface.co/Qwen) | 2B/7B/72B | qwen2_vl | | [Phi-3](https://huggingface.co/microsoft) | 4B/7B/14B | phi |
| [StarCoder 2](https://huggingface.co/bigcode) | 3B/7B/15B | - | | [Qwen (1-2.5) (Code/Math/MoE)](https://huggingface.co/Qwen) | 0.5B/1.5B/3B/7B/14B/32B/72B/110B | qwen |
| [XVERSE](https://huggingface.co/xverse) | 7B/13B/65B | xverse | | [Qwen2-VL](https://huggingface.co/Qwen) | 2B/7B/72B | qwen2_vl |
| [Yi/Yi-1.5 (Code)](https://huggingface.co/01-ai) | 1.5B/6B/9B/34B | yi | | [StarCoder 2](https://huggingface.co/bigcode) | 3B/7B/15B | - |
| [Yi-VL](https://huggingface.co/01-ai) | 6B/34B | yi_vl | | [XVERSE](https://huggingface.co/xverse) | 7B/13B/65B | xverse |
| [Yuan 2](https://huggingface.co/IEITYuan) | 2B/51B/102B | yuan | | [Yi/Yi-1.5 (Code)](https://huggingface.co/01-ai) | 1.5B/6B/9B/34B | yi |
| [Yi-VL](https://huggingface.co/01-ai) | 6B/34B | yi_vl |
| [Yuan 2](https://huggingface.co/IEITYuan) | 2B/51B/102B | yuan |
> [!NOTE] > [!NOTE]
> 对于所有“基座”（Base）模型，`template` 参数可以是 `default`, `alpaca`, `vicuna` 等任意值。但“对话”（Instruct/Chat）模型请务必使用**对应的模板**。 > 对于所有“基座”（Base）模型，`template` 参数可以是 `default`, `alpaca`, `vicuna` 等任意值。但“对话”（Instruct/Chat）模型请务必使用**对应的模板**。

BIN
assets/wechat.jpg
View File

Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 164 KiB

After

Width:  |  Height:  |  Size: 199 KiB

BIN
assets/wechat_npu.jpg
View File

Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 167 KiB

After

Width:  |  Height:  |  Size: 168 KiB

1
requirements.txt
View File

@ -19,3 +19,4 @@ fire
packaging packaging
pyyaml pyyaml
numpy<2.0.0 numpy<2.0.0
av

2
setup.py
View File

@ -54,7 +54,7 @@ extra_require = {
"gptq": ["optimum>=1.17.0", "auto-gptq>=0.5.0"], "gptq": ["optimum>=1.17.0", "auto-gptq>=0.5.0"],
"awq": ["autoawq"], "awq": ["autoawq"],
"aqlm": ["aqlm[gpu]>=1.1.0"], "aqlm": ["aqlm[gpu]>=1.1.0"],
"vllm": ["vllm>=0.4.3,<=0.6.0"], "vllm": ["vllm>=0.4.3,<=0.6.2"],
"galore": ["galore-torch"], "galore": ["galore-torch"],
"badam": ["badam>=1.2.1"], "badam": ["badam>=1.2.1"],
"adam-mini": ["adam-mini"], "adam-mini": ["adam-mini"],

198
src/llamafactory/data/mm_plugin.py
View File

@ -4,6 +4,7 @@ from io import BytesIO
from typing import TYPE_CHECKING, Dict, List, Optional, Sequence, Tuple, TypedDict, Union from typing import TYPE_CHECKING, Dict, List, Optional, Sequence, Tuple, TypedDict, Union
import numpy as np import numpy as np
from transformers.image_utils import get_image_size, to_numpy_array
from typing_extensions import override from typing_extensions import override
from ..extras.constants import IGNORE_INDEX, IMAGE_PLACEHOLDER, VIDEO_PLACEHOLDER from ..extras.constants import IGNORE_INDEX, IMAGE_PLACEHOLDER, VIDEO_PLACEHOLDER
@ -157,6 +158,7 @@ class BasePlugin:
It holds num_patches == torch.prod(image_grid_thw) It holds num_patches == torch.prod(image_grid_thw)
""" """
image_processor: "BaseImageProcessor" = getattr(processor, "image_processor") image_processor: "BaseImageProcessor" = getattr(processor, "image_processor")
video_processor: "BaseImageProcessor" = getattr(processor, "video_processor", image_processor)
input_dict = {"images": None} # default key input_dict = {"images": None} # default key
if len(images) != 0: if len(images) != 0:
images = self._regularize_images( images = self._regularize_images(
@ -174,10 +176,16 @@ class BasePlugin:
) )
input_dict["videos"] = videos input_dict["videos"] = videos
if input_dict.get("images", None) is not None or input_dict.get("videos", None) is not None: mm_inputs = {}
return image_processor(**input_dict, return_tensors="pt") if image_processor != video_processor:
else: if input_dict.get("images") is not None:
return {} mm_inputs.update(image_processor(input_dict["images"], return_tensors="pt"))
if input_dict.get("videos") is not None:
mm_inputs.update(video_processor(input_dict["videos"], return_tensors="pt"))
elif input_dict.get("images") is not None or input_dict.get("videos") is not None: # same processor (qwen2-vl)
mm_inputs.update(image_processor(**input_dict, return_tensors="pt"))
return mm_inputs
def process_messages( def process_messages(
self, self,
@ -263,6 +271,122 @@ class LlavaPlugin(BasePlugin):
return self._get_mm_inputs(images, videos, processor) return self._get_mm_inputs(images, videos, processor)
class LlavaNextPlugin(BasePlugin):
@override
def process_messages(
self,
messages: Sequence[Dict[str, str]],
images: Sequence["ImageInput"],
videos: Sequence["VideoInput"],
processor: Optional["ProcessorMixin"],
) -> List[Dict[str, str]]:
self._validate_input(images, videos)
num_image_tokens = 0
messages = deepcopy(messages)
mm_inputs = self._get_mm_inputs(images, videos, processor)
if "image_sizes" in mm_inputs:
image_sizes = iter(mm_inputs["image_sizes"])
if "pixel_values" in mm_inputs:
height, width = get_image_size(to_numpy_array(mm_inputs["pixel_values"][0][0]))
for message in messages:
content = message["content"]
while self.image_token in content:
image_size = next(image_sizes)
orig_height, orig_width = image_size
image_seqlen = processor._get_number_of_features(orig_height, orig_width, height, width)
if processor.vision_feature_select_strategy == "default":
image_seqlen -= 1
num_image_tokens += 1
content = content.replace(self.image_token, "{{image}}" * image_seqlen, 1)
message["content"] = content.replace("{{image}}", self.image_token)
if len(images) != num_image_tokens:
raise ValueError("The number of images does not match the number of {} tokens".format(IMAGE_PLACEHOLDER))
return messages
@override
def get_mm_inputs(
self,
images: Sequence["ImageInput"],
videos: Sequence["VideoInput"],
imglens: Sequence[int],
vidlens: Sequence[int],
seqlens: Sequence[int],
processor: Optional["ProcessorMixin"],
) -> Dict[str, Union[List[int], "torch.Tensor"]]:
self._validate_input(images, videos)
res = self._get_mm_inputs(images, videos, processor)
return res
class LlavaNextVideoPlugin(BasePlugin):
@override
def process_messages(
self,
messages: Sequence[Dict[str, str]],
images: Sequence["ImageInput"],
videos: Sequence["VideoInput"],
processor: Optional["ProcessorMixin"],
) -> List[Dict[str, str]]:
self._validate_input(images, videos)
num_image_tokens = 0
num_video_tokens = 0
messages = deepcopy(messages)
mm_inputs = self._get_mm_inputs(images, videos, processor)
if "pixel_values" in mm_inputs:
image_sizes = iter(mm_inputs["image_sizes"])
height, width = get_image_size(to_numpy_array(mm_inputs["pixel_values"][0][0]))
for message in messages:
content = message["content"]
while self.image_token in content:
image_size = next(image_sizes)
orig_height, orig_width = image_size
image_seqlen = processor._get_number_of_features(orig_height, orig_width, height, width)
if processor.vision_feature_select_strategy == "default":
image_seqlen -= 1
num_image_tokens += 1
content = content.replace(self.image_token, "{{image}}" * image_seqlen, 1)
message["content"] = content.replace("{{image}}", self.image_token)
if "pixel_values_videos" in mm_inputs:
pixel_values_video = to_numpy_array(mm_inputs.get("pixel_values_videos")[0])
height, width = get_image_size(pixel_values_video[0])
num_frames = pixel_values_video.shape[0] # frame dim is always after batch dim
image_seqlen = (height // processor.patch_size) * (width // processor.patch_size)
video_seqlen = image_seqlen // 4 * num_frames # divide by 4 needed for avg pooling layer
for message in messages:
content = message["content"]
while self.video_token in content:
num_video_tokens += 1
content = content.replace(self.video_token, "{{video}}", 1)
message["content"] = content.replace("{{video}}", self.video_token * video_seqlen)
if len(images) != num_image_tokens:
raise ValueError("The number of images does not match the number of {} tokens".format(IMAGE_PLACEHOLDER))
if len(videos) != num_video_tokens:
raise ValueError("The number of videos does not match the number of {} tokens".format(IMAGE_PLACEHOLDER))
return messages
@override
def get_mm_inputs(
self,
images: Sequence["ImageInput"],
videos: Sequence["VideoInput"],
imglens: Sequence[int],
vidlens: Sequence[int],
seqlens: Sequence[int],
processor: Optional["ProcessorMixin"],
) -> Dict[str, Union[List[int], "torch.Tensor"]]:
self._validate_input(images, videos)
return self._get_mm_inputs(images, videos, processor)
class PaliGemmaPlugin(BasePlugin): class PaliGemmaPlugin(BasePlugin):
@override @override
def process_messages( def process_messages(
@ -417,11 +541,77 @@ class Qwen2vlPlugin(BasePlugin):
return self._get_mm_inputs(images, videos, processor) return self._get_mm_inputs(images, videos, processor)
class VideoLlavaPlugin(BasePlugin):
@override
def process_messages(
self,
messages: Sequence[Dict[str, str]],
images: Sequence["ImageInput"],
videos: Sequence["VideoInput"],
processor: Optional["ProcessorMixin"],
) -> List[Dict[str, str]]:
self._validate_input(images, videos)
num_image_tokens = 0
num_video_tokens = 0
messages = deepcopy(messages)
mm_inputs = self._get_mm_inputs(images, videos, processor)
num_frames = 0
exist_images = "pixel_values_images" in mm_inputs
exist_videos = "pixel_values_videos" in mm_inputs
if exist_videos or exist_images:
if exist_images:
height, width = get_image_size(to_numpy_array(mm_inputs.get("pixel_values_images")[0]))
num_frames = 1
if exist_videos:
pixel_values_video = to_numpy_array(mm_inputs.get("pixel_values_videos")[0])
height, width = get_image_size(pixel_values_video[0])
num_frames = pixel_values_video.shape[0] # frame dim is always after batch dim
image_seqlen = (height // processor.patch_size) * (width // processor.patch_size) + 1
video_seqlen = image_seqlen * num_frames
if processor.vision_feature_select_strategy == "default":
image_seqlen -= 1
for message in messages:
content = message["content"]
while self.image_token in content:
num_image_tokens += 1
content = content.replace(self.image_token, "{{image}}", 1)
while self.video_token in content:
num_video_tokens += 1
content = content.replace(self.video_token, "{{video}}", 1)
content = content.replace("{{image}}", self.image_token * image_seqlen)
message["content"] = content.replace("{{video}}", self.video_token * video_seqlen)
if len(images) != num_image_tokens:
raise ValueError("The number of images does not match the number of {} tokens".format(self.image_token))
if len(videos) != num_video_tokens:
raise ValueError("The number of videos does not match the number of {} tokens".format(self.video_token))
return messages
@override
def get_mm_inputs(
self,
images: Sequence["ImageInput"],
videos: Sequence["VideoInput"],
imglens: Sequence[int],
vidlens: Sequence[int],
seqlens: Sequence[int],
processor: Optional["ProcessorMixin"],
) -> Dict[str, Union[List[int], "torch.Tensor"]]:
self._validate_input(images, videos)
return self._get_mm_inputs(images, videos, processor)
PLUGINS = { PLUGINS = {
"base": BasePlugin, "base": BasePlugin,
"llava": LlavaPlugin, "llava": LlavaPlugin,
"llava_next": LlavaNextPlugin,
"llava_next_video": LlavaNextVideoPlugin,
"paligemma": PaliGemmaPlugin, "paligemma": PaliGemmaPlugin,
"qwen2_vl": Qwen2vlPlugin, "qwen2_vl": Qwen2vlPlugin,
"video_llava": VideoLlavaPlugin,
} }

120
src/llamafactory/data/template.py
View File

@ -644,6 +644,14 @@ _register_template(
) )
_register_template(
name="exaone",
format_user=StringFormatter(slots=["[|user|]{{content}}\n[|assistant|]"]),
format_system=StringFormatter(slots=["[|system|]{{content}}[|endofturn|]\n"]),
format_separator=EmptyFormatter(slots=["\n"]),
)
_register_template( _register_template(
name="falcon", name="falcon",
format_user=StringFormatter(slots=["User: {{content}}\nFalcon:"]), format_user=StringFormatter(slots=["User: {{content}}\nFalcon:"]),
@ -760,6 +768,107 @@ _register_template(
) )
_register_template(
name="llava_next",
format_user=StringFormatter(slots=["USER: {{content}} ASSISTANT:"]),
default_system=(
"A chat between a curious user and an artificial intelligence assistant. "
"The assistant gives helpful, detailed, and polite answers to the user's questions."
),
mm_plugin=get_mm_plugin(name="llava_next", image_token="<image>"),
)
_register_template(
name="llava_next_llama3",
format_user=StringFormatter(
slots=[
(
"<|start_header_id|>user<|end_header_id|>\n\n{{content}}<|eot_id|>"
"<|start_header_id|>assistant<|end_header_id|>\n\n"
)
]
),
format_system=StringFormatter(slots=["<|start_header_id|>system<|end_header_id|>\n\n{{content}}<|eot_id|>"]),
format_observation=StringFormatter(
slots=[
(
"<|start_header_id|>tool<|end_header_id|>\n\n{{content}}<|eot_id|>"
"<|start_header_id|>assistant<|end_header_id|>\n\n"
)
]
),
format_prefix=EmptyFormatter(slots=[{"bos_token"}]),
stop_words=["<|eot_id|>"],
replace_eos=True,
replace_jinja_template=False,
mm_plugin=get_mm_plugin(name="llava_next", image_token="<image>"),
)
_register_template(
name="llava_next_mistral",
format_user=StringFormatter(slots=["[INST] {{content}} [/INST]"]),
format_prefix=EmptyFormatter(slots=[{"bos_token"}]),
mm_plugin=get_mm_plugin(name="llava_next", image_token="<image>"),
)
_register_template(
name="llava_next_qwen",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_system=StringFormatter(slots=["<|im_start|>system\n{{content}}<|im_end|>\n"]),
format_observation=StringFormatter(slots=["<|im_start|>tool\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_separator=EmptyFormatter(slots=["\n"]),
default_system="You are a helpful assistant.",
stop_words=["<|im_end|>"],
replace_eos=True,
replace_jinja_template=False,
mm_plugin=get_mm_plugin(name="llava_next", image_token="<image>"),
)
_register_template(
name="llava_next_yi",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_system=StringFormatter(slots=["<|im_start|>system\n{{content}}<|im_end|>\n"]),
format_separator=EmptyFormatter(slots=["\n"]),
stop_words=["<|im_end|>"],
replace_eos=True,
mm_plugin=get_mm_plugin(name="llava_next", image_token="<image>"),
)
_register_template(
name="llava_next_video",
format_user=StringFormatter(slots=["USER: {{content}} ASSISTANT:"]),
default_system=(
"A chat between a curious user and an artificial intelligence assistant. "
"The assistant gives helpful, detailed, and polite answers to the user's questions."
),
mm_plugin=get_mm_plugin(name="llava_next_video", image_token="<image>", video_token="<video>"),
)
_register_template(
name="llava_next_video_mistral",
format_user=StringFormatter(slots=["[INST] {{content}} [/INST]"]),
format_prefix=EmptyFormatter(slots=[{"bos_token"}]),
mm_plugin=get_mm_plugin(name="llava_next_video", image_token="<image>", video_token="<video>"),
)
_register_template(
name="llava_next_video_yi",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_system=StringFormatter(slots=["<|im_start|>system\n{{content}}<|im_end|>\n"]),
format_separator=EmptyFormatter(slots=["\n"]),
stop_words=["<|im_end|>"],
replace_eos=True,
mm_plugin=get_mm_plugin(name="llava_next_video", image_token="<image>", video_token="<video>"),
)
_register_template( _register_template(
name="mistral", name="mistral",
format_user=StringFormatter(slots=["[INST] {{content}} [/INST]"]), format_user=StringFormatter(slots=["[INST] {{content}} [/INST]"]),
@ -906,6 +1015,17 @@ _register_template(
) )
_register_template(
name="video_llava",
format_user=StringFormatter(slots=["USER: {{content}} ASSISTANT:"]),
default_system=(
"A chat between a curious user and an artificial intelligence assistant. "
"The assistant gives helpful, detailed, and polite answers to the user's questions."
),
mm_plugin=get_mm_plugin(name="video_llava", image_token="<image>", video_token="<video>"),
)
_register_template( _register_template(
name="xuanyuan", name="xuanyuan",
format_user=StringFormatter(slots=["Human: {{content}} Assistant:"]), format_user=StringFormatter(slots=["Human: {{content}} Assistant:"]),

377
src/llamafactory/extras/constants.py
View File

@ -114,17 +114,12 @@ def register_model_group(
template: Optional[str] = None, template: Optional[str] = None,
vision: bool = False, vision: bool = False,
) -> None: ) -> None:
prefix = None
for name, path in models.items(): for name, path in models.items():
if prefix is None:
prefix = name.split("-")[0]
else:
assert prefix == name.split("-")[0], "prefix should be identical."
SUPPORTED_MODELS[name] = path SUPPORTED_MODELS[name] = path
if template is not None: if template is not None and any(suffix in name for suffix in ("-Chat", "-Instruct")):
DEFAULT_TEMPLATE[prefix] = template DEFAULT_TEMPLATE[name] = template
if vision: if vision:
VISION_MODELS.add(prefix) VISION_MODELS.add(name)
register_model_group( register_model_group(
@ -274,27 +269,27 @@ register_model_group(
register_model_group( register_model_group(
models={ models={
"ChineseLLaMA2-1.3B": { "Chinese-Llama-2-1.3B": {
DownloadSource.DEFAULT: "hfl/chinese-llama-2-1.3b", DownloadSource.DEFAULT: "hfl/chinese-llama-2-1.3b",
DownloadSource.MODELSCOPE: "AI-ModelScope/chinese-llama-2-1.3b", DownloadSource.MODELSCOPE: "AI-ModelScope/chinese-llama-2-1.3b",
}, },
"ChineseLLaMA2-7B": { "Chinese-Llama-2-7B": {
DownloadSource.DEFAULT: "hfl/chinese-llama-2-7b", DownloadSource.DEFAULT: "hfl/chinese-llama-2-7b",
DownloadSource.MODELSCOPE: "AI-ModelScope/chinese-llama-2-7b", DownloadSource.MODELSCOPE: "AI-ModelScope/chinese-llama-2-7b",
}, },
"ChineseLLaMA2-13B": { "Chinese-Llama-2-13B": {
DownloadSource.DEFAULT: "hfl/chinese-llama-2-13b", DownloadSource.DEFAULT: "hfl/chinese-llama-2-13b",
DownloadSource.MODELSCOPE: "AI-ModelScope/chinese-llama-2-13b", DownloadSource.MODELSCOPE: "AI-ModelScope/chinese-llama-2-13b",
}, },
"ChineseLLaMA2-1.3B-Chat": { "Chinese-Alpaca-2-1.3B-Chat": {
DownloadSource.DEFAULT: "hfl/chinese-alpaca-2-1.3b", DownloadSource.DEFAULT: "hfl/chinese-alpaca-2-1.3b",
DownloadSource.MODELSCOPE: "AI-ModelScope/chinese-alpaca-2-1.3b", DownloadSource.MODELSCOPE: "AI-ModelScope/chinese-alpaca-2-1.3b",
}, },
"ChineseLLaMA2-7B-Chat": { "Chinese-Alpaca-2-7B-Chat": {
DownloadSource.DEFAULT: "hfl/chinese-alpaca-2-7b", DownloadSource.DEFAULT: "hfl/chinese-alpaca-2-7b",
DownloadSource.MODELSCOPE: "AI-ModelScope/chinese-alpaca-2-7b", DownloadSource.MODELSCOPE: "AI-ModelScope/chinese-alpaca-2-7b",
}, },
"ChineseLLaMA2-13B-Chat": { "Chinese-Alpaca-2-13B-Chat": {
DownloadSource.DEFAULT: "hfl/chinese-alpaca-2-13b", DownloadSource.DEFAULT: "hfl/chinese-alpaca-2-13b",
DownloadSource.MODELSCOPE: "AI-ModelScope/chinese-alpaca-2-13b", DownloadSource.MODELSCOPE: "AI-ModelScope/chinese-alpaca-2-13b",
}, },
@ -450,25 +445,25 @@ register_model_group(
register_model_group( register_model_group(
models={ models={
"DeepSeekCoder-6.7B-Base": { "DeepSeek-Coder-6.7B-Base": {
DownloadSource.DEFAULT: "deepseek-ai/deepseek-coder-6.7b-base", DownloadSource.DEFAULT: "deepseek-ai/deepseek-coder-6.7b-base",
DownloadSource.MODELSCOPE: "deepseek-ai/deepseek-coder-6.7b-base", DownloadSource.MODELSCOPE: "deepseek-ai/deepseek-coder-6.7b-base",
}, },
"DeepSeekCoder-7B-Base": { "DeepSeek-Coder-7B-Base": {
DownloadSource.DEFAULT: "deepseek-ai/deepseek-coder-7b-base-v1.5", DownloadSource.DEFAULT: "deepseek-ai/deepseek-coder-7b-base-v1.5",
}, },
"DeepSeekCoder-33B-Base": { "DeepSeek-Coder-33B-Base": {
DownloadSource.DEFAULT: "deepseek-ai/deepseek-coder-33b-base", DownloadSource.DEFAULT: "deepseek-ai/deepseek-coder-33b-base",
DownloadSource.MODELSCOPE: "deepseek-ai/deepseek-coder-33b-base", DownloadSource.MODELSCOPE: "deepseek-ai/deepseek-coder-33b-base",
}, },
"DeepSeekCoder-6.7B-Instruct": { "DeepSeek-Coder-6.7B-Instruct": {
DownloadSource.DEFAULT: "deepseek-ai/deepseek-coder-6.7b-instruct", DownloadSource.DEFAULT: "deepseek-ai/deepseek-coder-6.7b-instruct",
DownloadSource.MODELSCOPE: "deepseek-ai/deepseek-coder-6.7b-instruct", DownloadSource.MODELSCOPE: "deepseek-ai/deepseek-coder-6.7b-instruct",
}, },
"DeepSeekCoder-7B-Instruct": { "DeepSeek-Coder-7B-Instruct": {
DownloadSource.DEFAULT: "deepseek-ai/deepseek-coder-7b-instruct-v1.5", DownloadSource.DEFAULT: "deepseek-ai/deepseek-coder-7b-instruct-v1.5",
}, },
"DeepSeekCoder-33B-Instruct": { "DeepSeek-Coder-33B-Instruct": {
DownloadSource.DEFAULT: "deepseek-ai/deepseek-coder-33b-instruct", DownloadSource.DEFAULT: "deepseek-ai/deepseek-coder-33b-instruct",
DownloadSource.MODELSCOPE: "deepseek-ai/deepseek-coder-33b-instruct", DownloadSource.MODELSCOPE: "deepseek-ai/deepseek-coder-33b-instruct",
}, },
@ -477,6 +472,16 @@ register_model_group(
) )
register_model_group(
models={
"EXAONE-3.0-7.8B-Instruct": {
DownloadSource.DEFAULT: "LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct",
},
},
template="exaone",
)
register_model_group( register_model_group(
models={ models={
"Falcon-7B": { "Falcon-7B": {
@ -624,13 +629,6 @@ register_model_group(
DownloadSource.DEFAULT: "internlm/internlm2-chat-20b", DownloadSource.DEFAULT: "internlm/internlm2-chat-20b",
DownloadSource.MODELSCOPE: "Shanghai_AI_Laboratory/internlm2-chat-20b", DownloadSource.MODELSCOPE: "Shanghai_AI_Laboratory/internlm2-chat-20b",
}, },
},
template="intern2",
)
register_model_group(
models={
"InternLM2.5-1.8B": { "InternLM2.5-1.8B": {
DownloadSource.DEFAULT: "internlm/internlm2_5-1_8b", DownloadSource.DEFAULT: "internlm/internlm2_5-1_8b",
DownloadSource.MODELSCOPE: "Shanghai_AI_Laboratory/internlm2_5-1_8b", DownloadSource.MODELSCOPE: "Shanghai_AI_Laboratory/internlm2_5-1_8b",
@ -686,19 +684,19 @@ register_model_group(
register_model_group( register_model_group(
models={ models={
"LLaMA-7B": { "Llama-7B": {
DownloadSource.DEFAULT: "huggyllama/llama-7b", DownloadSource.DEFAULT: "huggyllama/llama-7b",
DownloadSource.MODELSCOPE: "skyline2006/llama-7b", DownloadSource.MODELSCOPE: "skyline2006/llama-7b",
}, },
"LLaMA-13B": { "Llama-13B": {
DownloadSource.DEFAULT: "huggyllama/llama-13b", DownloadSource.DEFAULT: "huggyllama/llama-13b",
DownloadSource.MODELSCOPE: "skyline2006/llama-13b", DownloadSource.MODELSCOPE: "skyline2006/llama-13b",
}, },
"LLaMA-30B": { "Llama-30B": {
DownloadSource.DEFAULT: "huggyllama/llama-30b", DownloadSource.DEFAULT: "huggyllama/llama-30b",
DownloadSource.MODELSCOPE: "skyline2006/llama-30b", DownloadSource.MODELSCOPE: "skyline2006/llama-30b",
}, },
"LLaMA-65B": { "Llama-65B": {
DownloadSource.DEFAULT: "huggyllama/llama-65b", DownloadSource.DEFAULT: "huggyllama/llama-65b",
DownloadSource.MODELSCOPE: "skyline2006/llama-65b", DownloadSource.MODELSCOPE: "skyline2006/llama-65b",
}, },
@ -708,27 +706,27 @@ register_model_group(
register_model_group( register_model_group(
models={ models={
"LLaMA2-7B": { "Llama-2-7B": {
DownloadSource.DEFAULT: "meta-llama/Llama-2-7b-hf", DownloadSource.DEFAULT: "meta-llama/Llama-2-7b-hf",
DownloadSource.MODELSCOPE: "modelscope/Llama-2-7b-ms", DownloadSource.MODELSCOPE: "modelscope/Llama-2-7b-ms",
}, },
"LLaMA2-13B": { "Llama-2-13B": {
DownloadSource.DEFAULT: "meta-llama/Llama-2-13b-hf", DownloadSource.DEFAULT: "meta-llama/Llama-2-13b-hf",
DownloadSource.MODELSCOPE: "modelscope/Llama-2-13b-ms", DownloadSource.MODELSCOPE: "modelscope/Llama-2-13b-ms",
}, },
"LLaMA2-70B": { "Llama-2-70B": {
DownloadSource.DEFAULT: "meta-llama/Llama-2-70b-hf", DownloadSource.DEFAULT: "meta-llama/Llama-2-70b-hf",
DownloadSource.MODELSCOPE: "modelscope/Llama-2-70b-ms", DownloadSource.MODELSCOPE: "modelscope/Llama-2-70b-ms",
}, },
"LLaMA2-7B-Chat": { "Llama-2-7B-Chat": {
DownloadSource.DEFAULT: "meta-llama/Llama-2-7b-chat-hf", DownloadSource.DEFAULT: "meta-llama/Llama-2-7b-chat-hf",
DownloadSource.MODELSCOPE: "modelscope/Llama-2-7b-chat-ms", DownloadSource.MODELSCOPE: "modelscope/Llama-2-7b-chat-ms",
}, },
"LLaMA2-13B-Chat": { "Llama-2-13B-Chat": {
DownloadSource.DEFAULT: "meta-llama/Llama-2-13b-chat-hf", DownloadSource.DEFAULT: "meta-llama/Llama-2-13b-chat-hf",
DownloadSource.MODELSCOPE: "modelscope/Llama-2-13b-chat-ms", DownloadSource.MODELSCOPE: "modelscope/Llama-2-13b-chat-ms",
}, },
"LLaMA2-70B-Chat": { "Llama-2-70B-Chat": {
DownloadSource.DEFAULT: "meta-llama/Llama-2-70b-chat-hf", DownloadSource.DEFAULT: "meta-llama/Llama-2-70b-chat-hf",
DownloadSource.MODELSCOPE: "modelscope/Llama-2-70b-chat-ms", DownloadSource.MODELSCOPE: "modelscope/Llama-2-70b-chat-ms",
}, },
@ -739,60 +737,77 @@ register_model_group(
register_model_group( register_model_group(
models={ models={
"LLaMA3-8B": { "Llama-3-8B": {
DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3-8B", DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3-8B",
DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3-8B", DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3-8B",
}, },
"LLaMA3-70B": { "Llama-3-70B": {
DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3-70B", DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3-70B",
DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3-70B", DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3-70B",
}, },
"LLaMA3-8B-Instruct": { "Llama-3-8B-Instruct": {
DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3-8B-Instruct", DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3-8B-Instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3-8B-Instruct", DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3-8B-Instruct",
}, },
"LLaMA3-70B-Instruct": { "Llama-3-70B-Instruct": {
DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3-70B-Instruct", DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3-70B-Instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3-70B-Instruct", DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3-70B-Instruct",
}, },
"LLaMA3-8B-Chinese-Chat": { "Llama-3-8B-Chinese-Chat": {
DownloadSource.DEFAULT: "shenzhi-wang/Llama3-8B-Chinese-Chat", DownloadSource.DEFAULT: "shenzhi-wang/Llama3-8B-Chinese-Chat",
DownloadSource.MODELSCOPE: "LLM-Research/Llama3-8B-Chinese-Chat", DownloadSource.MODELSCOPE: "LLM-Research/Llama3-8B-Chinese-Chat",
}, },
"LLaMA3-70B-Chinese-Chat": { "Llama-3-70B-Chinese-Chat": {
DownloadSource.DEFAULT: "shenzhi-wang/Llama3-70B-Chinese-Chat", DownloadSource.DEFAULT: "shenzhi-wang/Llama3-70B-Chinese-Chat",
}, },
}, "Llama-3.1-8B": {
template="llama3",
)
register_model_group(
models={
"LLaMA3.1-8B": {
DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3.1-8B", DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3.1-8B",
DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3.1-8B", DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3.1-8B",
}, },
"LLaMA3.1-70B": { "Llama-3.1-70B": {
DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3.1-70B", DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3.1-70B",
DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3.1-70B", DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3.1-70B",
}, },
"LLaMA3.1-405B": { "Llama-3.1-405B": {
DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3.1-405B", DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3.1-405B",
DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3.1-405B", DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3.1-405B",
}, },
"LLaMA3.1-8B-Instruct": { "Llama-3.1-8B-Instruct": {
DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3.1-8B-Instruct", DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3.1-8B-Instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3.1-8B-Instruct", DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3.1-8B-Instruct",
}, },
"LLaMA3.1-70B-Instruct": { "Llama-3.1-70B-Instruct": {
DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3.1-70B-Instruct", DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3.1-70B-Instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3.1-70B-Instruct", DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3.1-70B-Instruct",
}, },
"LLaMA3.1-405B-Instruct": { "Llama-3.1-405B-Instruct": {
DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3.1-405B-Instruct", DownloadSource.DEFAULT: "meta-llama/Meta-Llama-3.1-405B-Instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3.1-405B-Instruct", DownloadSource.MODELSCOPE: "LLM-Research/Meta-Llama-3.1-405B-Instruct",
}, },
"Llama-3.1-8B-Chinese-Chat": {
DownloadSource.DEFAULT: "shenzhi-wang/Llama3.1-8B-Chinese-Chat",
DownloadSource.MODELSCOPE: "XD_AI/Llama3.1-8B-Chinese-Chat",
},
"Llama-3.1-70B-Chinese-Chat": {
DownloadSource.DEFAULT: "shenzhi-wang/Llama3.1-70B-Chinese-Chat",
DownloadSource.MODELSCOPE: "XD_AI/Llama3.1-70B-Chinese-Chat",
},
"Llama-3.2-1B": {
DownloadSource.DEFAULT: "meta-llama/Llama-3.2-1B",
DownloadSource.MODELSCOPE: "LLM-Research/Llama-3.2-1B",
},
"Llama-3.2-3B": {
DownloadSource.DEFAULT: "meta-llama/Llama-3.2-3B",
DownloadSource.MODELSCOPE: "LLM-Research/Llama-3.2-3B",
},
"Llama-3.2-1B-Instruct": {
DownloadSource.DEFAULT: "meta-llama/Llama-3.2-1B-Instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Llama-3.2-1B-Instruct",
},
"Llama-3.2-3B-Instruct": {
DownloadSource.DEFAULT: "meta-llama/Llama-3.2-3B-Instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Llama-3.2-3B-Instruct",
},
}, },
template="llama3", template="llama3",
) )
@ -800,11 +815,13 @@ register_model_group(
register_model_group( register_model_group(
models={ models={
"LLaVA1.5-7B-Chat": { "LLaVA-1.5-7B-Chat": {
DownloadSource.DEFAULT: "llava-hf/llava-1.5-7b-hf", DownloadSource.DEFAULT: "llava-hf/llava-1.5-7b-hf",
DownloadSource.MODELSCOPE: "swift/llava-1.5-7b-hf",
}, },
"LLaVA1.5-13B-Chat": { "LLaVA-1.5-13B-Chat": {
DownloadSource.DEFAULT: "llava-hf/llava-1.5-13b-hf", DownloadSource.DEFAULT: "llava-hf/llava-1.5-13b-hf",
DownloadSource.MODELSCOPE: "swift/llava-1.5-13b-hf",
}, },
}, },
template="llava", template="llava",
@ -812,6 +829,117 @@ register_model_group(
) )
register_model_group(
models={
"LLaVA-NeXT-7B-Chat": {
DownloadSource.DEFAULT: "llava-hf/llava-v1.6-vicuna-7b-hf",
DownloadSource.MODELSCOPE: "swift/llava-v1.6-vicuna-7b-hf",
},
"LLaVA-NeXT-13B-Chat": {
DownloadSource.DEFAULT: "llava-hf/llava-v1.6-vicuna-13b-hf",
DownloadSource.MODELSCOPE: "swift/llava-v1.6-vicuna-13b-hf",
},
},
template="llava_next",
vision=True,
)
register_model_group(
models={
"LLaVA-NeXT-Mistral-7B-Chat": {
DownloadSource.DEFAULT: "llava-hf/llava-v1.6-mistral-7b-hf",
DownloadSource.MODELSCOPE: "swift/llava-v1.6-mistral-7b-hf",
},
},
template="llava_next_mistral",
vision=True,
)
register_model_group(
models={
"LLaVA-NeXT-Llama3-8B-Chat": {
DownloadSource.DEFAULT: "llava-hf/llama3-llava-next-8b-hf",
DownloadSource.MODELSCOPE: "swift/llama3-llava-next-8b-hf",
},
},
template="llava_next_llama3",
vision=True,
)
register_model_group(
models={
"LLaVA-NeXT-34B-Chat": {
DownloadSource.DEFAULT: "llava-hf/llava-v1.6-34b-hf",
DownloadSource.MODELSCOPE: "LLM-Research/llava-v1.6-34b-hf",
},
},
template="llava_next_yi",
vision=True,
)
register_model_group(
models={
"LLaVA-NeXT-72B-Chat": {
DownloadSource.DEFAULT: "llava-hf/llava-next-72b-hf",
DownloadSource.MODELSCOPE: "AI-ModelScope/llava-next-72b-hf",
},
"LLaVA-NeXT-110B-Chat": {
DownloadSource.DEFAULT: "llava-hf/llava-next-110b-hf",
DownloadSource.MODELSCOPE: "AI-ModelScope/llava-next-110b-hf",
},
},
template="llava_next_qwen",
vision=True,
)
register_model_group(
models={
"LLaVA-NeXT-Video-7B-Chat": {
DownloadSource.DEFAULT: "llava-hf/LLaVA-NeXT-Video-7B-hf",
DownloadSource.MODELSCOPE: "swift/LLaVA-NeXT-Video-7B-hf",
},
"LLaVA-NeXT-Video-7B-DPO-Chat": {
DownloadSource.DEFAULT: "llava-hf/LLaVA-NeXT-Video-7B-DPO-hf",
DownloadSource.MODELSCOPE: "swift/LLaVA-NeXT-Video-7B-DPO-hf",
},
},
template="llava_next_video",
vision=True,
)
register_model_group(
models={
"LLaVA-NeXT-Video-7B-32k-Chat": {
DownloadSource.DEFAULT: "llava-hf/LLaVA-NeXT-Video-7B-32K-hf",
DownloadSource.MODELSCOPE: "swift/LLaVA-NeXT-Video-7B-32K-hf",
},
},
template="llava_next_video_mistral",
vision=True,
)
register_model_group(
models={
"LLaVA-NeXT-Video-34B-Chat": {
DownloadSource.DEFAULT: "llava-hf/LLaVA-NeXT-Video-34B-hf",
DownloadSource.MODELSCOPE: "swift/LLaVA-NeXT-Video-34B-hf",
},
"LLaVA-NeXT-Video-34B-DPO-Chat": {
DownloadSource.DEFAULT: "llava-hf/LLaVA-NeXT-Video-34B-DPO-hf",
},
},
template="llava_next_video_yi",
vision=True,
)
register_model_group( register_model_group(
models={ models={
"MiniCPM-2B-SFT-Chat": { "MiniCPM-2B-SFT-Chat": {
@ -1005,27 +1133,27 @@ register_model_group(
register_model_group( register_model_group(
models={ models={
"Phi3-4B-4k-Instruct": { "Phi-3-4B-4k-Instruct": {
DownloadSource.DEFAULT: "microsoft/Phi-3-mini-4k-instruct", DownloadSource.DEFAULT: "microsoft/Phi-3-mini-4k-instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Phi-3-mini-4k-instruct", DownloadSource.MODELSCOPE: "LLM-Research/Phi-3-mini-4k-instruct",
}, },
"Phi3-4B-128k-Instruct": { "Phi-3-4B-128k-Instruct": {
DownloadSource.DEFAULT: "microsoft/Phi-3-mini-128k-instruct", DownloadSource.DEFAULT: "microsoft/Phi-3-mini-128k-instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Phi-3-mini-128k-instruct", DownloadSource.MODELSCOPE: "LLM-Research/Phi-3-mini-128k-instruct",
}, },
"Phi3-7B-8k-Instruct": { "Phi-3-7B-8k-Instruct": {
DownloadSource.DEFAULT: "microsoft/Phi-3-small-8k-instruct", DownloadSource.DEFAULT: "microsoft/Phi-3-small-8k-instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Phi-3-small-8k-instruct", DownloadSource.MODELSCOPE: "LLM-Research/Phi-3-small-8k-instruct",
}, },
"Phi3-7B-128k-Instruct": { "Phi-3-7B-128k-Instruct": {
DownloadSource.DEFAULT: "microsoft/Phi-3-small-128k-instruct", DownloadSource.DEFAULT: "microsoft/Phi-3-small-128k-instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Phi-3-small-128k-instruct", DownloadSource.MODELSCOPE: "LLM-Research/Phi-3-small-128k-instruct",
}, },
"Phi3-14B-8k-Instruct": { "Phi-3-14B-8k-Instruct": {
DownloadSource.DEFAULT: "microsoft/Phi-3-medium-4k-instruct", DownloadSource.DEFAULT: "microsoft/Phi-3-medium-4k-instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Phi-3-medium-4k-instruct", DownloadSource.MODELSCOPE: "LLM-Research/Phi-3-medium-4k-instruct",
}, },
"Phi3-14B-128k-Instruct": { "Phi-3-14B-128k-Instruct": {
DownloadSource.DEFAULT: "microsoft/Phi-3-medium-128k-instruct", DownloadSource.DEFAULT: "microsoft/Phi-3-medium-128k-instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Phi-3-medium-128k-instruct", DownloadSource.MODELSCOPE: "LLM-Research/Phi-3-medium-128k-instruct",
}, },
@ -1068,35 +1196,35 @@ register_model_group(
DownloadSource.DEFAULT: "Qwen/Qwen-72B-Chat", DownloadSource.DEFAULT: "Qwen/Qwen-72B-Chat",
DownloadSource.MODELSCOPE: "qwen/Qwen-72B-Chat", DownloadSource.MODELSCOPE: "qwen/Qwen-72B-Chat",
}, },
"Qwen-1.8B-int8-Chat": { "Qwen-1.8B-Chat-Int8": {
DownloadSource.DEFAULT: "Qwen/Qwen-1_8B-Chat-Int8", DownloadSource.DEFAULT: "Qwen/Qwen-1_8B-Chat-Int8",
DownloadSource.MODELSCOPE: "qwen/Qwen-1_8B-Chat-Int8", DownloadSource.MODELSCOPE: "qwen/Qwen-1_8B-Chat-Int8",
}, },
"Qwen-1.8B-int4-Chat": { "Qwen-1.8B-Chat-Int4": {
DownloadSource.DEFAULT: "Qwen/Qwen-1_8B-Chat-Int4", DownloadSource.DEFAULT: "Qwen/Qwen-1_8B-Chat-Int4",
DownloadSource.MODELSCOPE: "qwen/Qwen-1_8B-Chat-Int4", DownloadSource.MODELSCOPE: "qwen/Qwen-1_8B-Chat-Int4",
}, },
"Qwen-7B-int8-Chat": { "Qwen-7B-Chat-Int8": {
DownloadSource.DEFAULT: "Qwen/Qwen-7B-Chat-Int8", DownloadSource.DEFAULT: "Qwen/Qwen-7B-Chat-Int8",
DownloadSource.MODELSCOPE: "qwen/Qwen-7B-Chat-Int8", DownloadSource.MODELSCOPE: "qwen/Qwen-7B-Chat-Int8",
}, },
"Qwen-7B-int4-Chat": { "Qwen-7B-Chat-Int4": {
DownloadSource.DEFAULT: "Qwen/Qwen-7B-Chat-Int4", DownloadSource.DEFAULT: "Qwen/Qwen-7B-Chat-Int4",
DownloadSource.MODELSCOPE: "qwen/Qwen-7B-Chat-Int4", DownloadSource.MODELSCOPE: "qwen/Qwen-7B-Chat-Int4",
}, },
"Qwen-14B-int8-Chat": { "Qwen-14B-Chat-Int8": {
DownloadSource.DEFAULT: "Qwen/Qwen-14B-Chat-Int8", DownloadSource.DEFAULT: "Qwen/Qwen-14B-Chat-Int8",
DownloadSource.MODELSCOPE: "qwen/Qwen-14B-Chat-Int8", DownloadSource.MODELSCOPE: "qwen/Qwen-14B-Chat-Int8",
}, },
"Qwen-14B-int4-Chat": { "Qwen-14B-Chat-Int4": {
DownloadSource.DEFAULT: "Qwen/Qwen-14B-Chat-Int4", DownloadSource.DEFAULT: "Qwen/Qwen-14B-Chat-Int4",
DownloadSource.MODELSCOPE: "qwen/Qwen-14B-Chat-Int4", DownloadSource.MODELSCOPE: "qwen/Qwen-14B-Chat-Int4",
}, },
"Qwen-72B-int8-Chat": { "Qwen-72B-Chat-Int8": {
DownloadSource.DEFAULT: "Qwen/Qwen-72B-Chat-Int8", DownloadSource.DEFAULT: "Qwen/Qwen-72B-Chat-Int8",
DownloadSource.MODELSCOPE: "qwen/Qwen-72B-Chat-Int8", DownloadSource.MODELSCOPE: "qwen/Qwen-72B-Chat-Int8",
}, },
"Qwen-72B-int4-Chat": { "Qwen-72B-Chat-Int4": {
DownloadSource.DEFAULT: "Qwen/Qwen-72B-Chat-Int4", DownloadSource.DEFAULT: "Qwen/Qwen-72B-Chat-Int4",
DownloadSource.MODELSCOPE: "qwen/Qwen-72B-Chat-Int4", DownloadSource.MODELSCOPE: "qwen/Qwen-72B-Chat-Int4",
}, },
@ -1179,75 +1307,75 @@ register_model_group(
DownloadSource.DEFAULT: "Qwen/Qwen1.5-MoE-A2.7B-Chat", DownloadSource.DEFAULT: "Qwen/Qwen1.5-MoE-A2.7B-Chat",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-MoE-A2.7B-Chat", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-MoE-A2.7B-Chat",
}, },
"Qwen1.5-0.5B-int8-Chat": { "Qwen1.5-0.5B-Chat-GPTQ-Int8": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-0.5B-Chat-GPTQ-Int8", DownloadSource.DEFAULT: "Qwen/Qwen1.5-0.5B-Chat-GPTQ-Int8",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-0.5B-Chat-GPTQ-Int8", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-0.5B-Chat-GPTQ-Int8",
}, },
"Qwen1.5-0.5B-int4-Chat": { "Qwen1.5-0.5B-Chat-AWQ": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-0.5B-Chat-AWQ", DownloadSource.DEFAULT: "Qwen/Qwen1.5-0.5B-Chat-AWQ",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-0.5B-Chat-AWQ", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-0.5B-Chat-AWQ",
}, },
"Qwen1.5-1.8B-int8-Chat": { "Qwen1.5-1.8B-Chat-GPTQ-Int8": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-1.8B-Chat-GPTQ-Int8", DownloadSource.DEFAULT: "Qwen/Qwen1.5-1.8B-Chat-GPTQ-Int8",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-1.8B-Chat-GPTQ-Int8", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-1.8B-Chat-GPTQ-Int8",
}, },
"Qwen1.5-1.8B-int4-Chat": { "Qwen1.5-1.8B-Chat-AWQ": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-1.8B-Chat-AWQ", DownloadSource.DEFAULT: "Qwen/Qwen1.5-1.8B-Chat-AWQ",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-1.8B-Chat-AWQ", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-1.8B-Chat-AWQ",
}, },
"Qwen1.5-4B-int8-Chat": { "Qwen1.5-4B-Chat-GPTQ-Int8": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-4B-Chat-GPTQ-Int8", DownloadSource.DEFAULT: "Qwen/Qwen1.5-4B-Chat-GPTQ-Int8",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-4B-Chat-GPTQ-Int8", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-4B-Chat-GPTQ-Int8",
}, },
"Qwen1.5-4B-int4-Chat": { "Qwen1.5-4B-Chat-AWQ": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-4B-Chat-AWQ", DownloadSource.DEFAULT: "Qwen/Qwen1.5-4B-Chat-AWQ",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-4B-Chat-AWQ", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-4B-Chat-AWQ",
}, },
"Qwen1.5-7B-int8-Chat": { "Qwen1.5-7B-Chat-GPTQ-Int8": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-7B-Chat-GPTQ-Int8", DownloadSource.DEFAULT: "Qwen/Qwen1.5-7B-Chat-GPTQ-Int8",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-7B-Chat-GPTQ-Int8", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-7B-Chat-GPTQ-Int8",
}, },
"Qwen1.5-7B-int4-Chat": { "Qwen1.5-7B-Chat-AWQ": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-7B-Chat-AWQ", DownloadSource.DEFAULT: "Qwen/Qwen1.5-7B-Chat-AWQ",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-7B-Chat-AWQ", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-7B-Chat-AWQ",
}, },
"Qwen1.5-14B-int8-Chat": { "Qwen1.5-14B-Chat-GPTQ-Int8": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-14B-Chat-GPTQ-Int8", DownloadSource.DEFAULT: "Qwen/Qwen1.5-14B-Chat-GPTQ-Int8",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-14B-Chat-GPTQ-Int8", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-14B-Chat-GPTQ-Int8",
}, },
"Qwen1.5-14B-int4-Chat": { "Qwen1.5-14B-Chat-AWQ": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-14B-Chat-AWQ", DownloadSource.DEFAULT: "Qwen/Qwen1.5-14B-Chat-AWQ",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-14B-Chat-AWQ", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-14B-Chat-AWQ",
}, },
"Qwen1.5-32B-int4-Chat": { "Qwen1.5-32B-Chat-AWQ": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-32B-Chat-AWQ", DownloadSource.DEFAULT: "Qwen/Qwen1.5-32B-Chat-AWQ",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-32B-Chat-AWQ", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-32B-Chat-AWQ",
}, },
"Qwen1.5-72B-int8-Chat": { "Qwen1.5-72B-Chat-GPTQ-Int8": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-72B-Chat-GPTQ-Int8", DownloadSource.DEFAULT: "Qwen/Qwen1.5-72B-Chat-GPTQ-Int8",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-72B-Chat-GPTQ-Int8", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-72B-Chat-GPTQ-Int8",
}, },
"Qwen1.5-72B-int4-Chat": { "Qwen1.5-72B-Chat-AWQ": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-72B-Chat-AWQ", DownloadSource.DEFAULT: "Qwen/Qwen1.5-72B-Chat-AWQ",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-72B-Chat-AWQ", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-72B-Chat-AWQ",
}, },
"Qwen1.5-110B-int4-Chat": { "Qwen1.5-110B-Chat-AWQ": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-110B-Chat-AWQ", DownloadSource.DEFAULT: "Qwen/Qwen1.5-110B-Chat-AWQ",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-110B-Chat-AWQ", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-110B-Chat-AWQ",
}, },
"Qwen1.5-MoE-A2.7B-int4-Chat": { "Qwen1.5-MoE-A2.7B-Chat-GPTQ-Int4": {
DownloadSource.DEFAULT: "Qwen/Qwen1.5-MoE-A2.7B-Chat-GPTQ-Int4", DownloadSource.DEFAULT: "Qwen/Qwen1.5-MoE-A2.7B-Chat-GPTQ-Int4",
DownloadSource.MODELSCOPE: "qwen/Qwen1.5-MoE-A2.7B-Chat-GPTQ-Int4", DownloadSource.MODELSCOPE: "qwen/Qwen1.5-MoE-A2.7B-Chat-GPTQ-Int4",
}, },
"Qwen1.5-Code-7B": { "CodeQwen1.5-7B": {
DownloadSource.DEFAULT: "Qwen/CodeQwen1.5-7B", DownloadSource.DEFAULT: "Qwen/CodeQwen1.5-7B",
DownloadSource.MODELSCOPE: "qwen/CodeQwen1.5-7B", DownloadSource.MODELSCOPE: "qwen/CodeQwen1.5-7B",
}, },
"Qwen1.5-Code-7B-Chat": { "CodeQwen1.5-7B-Chat": {
DownloadSource.DEFAULT: "Qwen/CodeQwen1.5-7B-Chat", DownloadSource.DEFAULT: "Qwen/CodeQwen1.5-7B-Chat",
DownloadSource.MODELSCOPE: "qwen/CodeQwen1.5-7B-Chat", DownloadSource.MODELSCOPE: "qwen/CodeQwen1.5-7B-Chat",
}, },
"Qwen1.5-Code-7B-int4-Chat": { "CodeQwen1.5-7B-Chat-AWQ": {
DownloadSource.DEFAULT: "Qwen/CodeQwen1.5-7B-Chat-AWQ", DownloadSource.DEFAULT: "Qwen/CodeQwen1.5-7B-Chat-AWQ",
DownloadSource.MODELSCOPE: "qwen/CodeQwen1.5-7B-Chat-AWQ", DownloadSource.MODELSCOPE: "qwen/CodeQwen1.5-7B-Chat-AWQ",
}, },
@ -1568,51 +1696,51 @@ register_model_group(
register_model_group( register_model_group(
models={ models={
"Qwen2VL-2B-Instruct": { "Qwen2-VL-2B-Instruct": {
DownloadSource.DEFAULT: "Qwen/Qwen2-VL-2B-Instruct", DownloadSource.DEFAULT: "Qwen/Qwen2-VL-2B-Instruct",
DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-2B-Instruct", DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-2B-Instruct",
}, },
"Qwen2VL-7B-Instruct": { "Qwen2-VL-7B-Instruct": {
DownloadSource.DEFAULT: "Qwen/Qwen2-VL-7B-Instruct", DownloadSource.DEFAULT: "Qwen/Qwen2-VL-7B-Instruct",
DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-7B-Instruct", DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-7B-Instruct",
}, },
"Qwen2VL-72B-Instruct": { "Qwen2-VL-72B-Instruct": {
DownloadSource.DEFAULT: "Qwen/Qwen2-VL-72B-Instruct", DownloadSource.DEFAULT: "Qwen/Qwen2-VL-72B-Instruct",
DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-72B-Instruct", DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-72B-Instruct",
}, },
"Qwen2VL-2B-Instruct-GPTQ-Int8": { "Qwen2-VL-2B-Instruct-GPTQ-Int8": {
DownloadSource.DEFAULT: "Qwen/Qwen2-VL-2B-Instruct-GPTQ-Int8", DownloadSource.DEFAULT: "Qwen/Qwen2-VL-2B-Instruct-GPTQ-Int8",
DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-2B-Instruct-GPTQ-Int8", DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-2B-Instruct-GPTQ-Int8",
}, },
"Qwen2VL-2B-Instruct-GPTQ-Int4": { "Qwen2-VL-2B-Instruct-GPTQ-Int4": {
DownloadSource.DEFAULT: "Qwen/Qwen2-VL-2B-Instruct-GPTQ-Int4", DownloadSource.DEFAULT: "Qwen/Qwen2-VL-2B-Instruct-GPTQ-Int4",
DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-2B-Instruct-GPTQ-Int4", DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-2B-Instruct-GPTQ-Int4",
}, },
"Qwen2VL-2B-Instruct-AWQ": { "Qwen2-VL-2B-Instruct-AWQ": {
DownloadSource.DEFAULT: "Qwen/Qwen2-VL-2B-Instruct-AWQ", DownloadSource.DEFAULT: "Qwen/Qwen2-VL-2B-Instruct-AWQ",
DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-2B-Instruct-AWQ", DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-2B-Instruct-AWQ",
}, },
"Qwen2VL-7B-Instruct-GPTQ-Int8": { "Qwen2-VL-7B-Instruct-GPTQ-Int8": {
DownloadSource.DEFAULT: "Qwen/Qwen2-VL-7B-Instruct-GPTQ-Int8", DownloadSource.DEFAULT: "Qwen/Qwen2-VL-7B-Instruct-GPTQ-Int8",
DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-7B-Instruct-GPTQ-Int8", DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-7B-Instruct-GPTQ-Int8",
}, },
"Qwen2VL-7B-Instruct-GPTQ-Int4": { "Qwen2-VL-7B-Instruct-GPTQ-Int4": {
DownloadSource.DEFAULT: "Qwen/Qwen2-VL-7B-Instruct-GPTQ-Int4", DownloadSource.DEFAULT: "Qwen/Qwen2-VL-7B-Instruct-GPTQ-Int4",
DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-7B-Instruct-GPTQ-Int4", DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-7B-Instruct-GPTQ-Int4",
}, },
"Qwen2VL-7B-Instruct-AWQ": { "Qwen2-VL-7B-Instruct-AWQ": {
DownloadSource.DEFAULT: "Qwen/Qwen2-VL-7B-Instruct-AWQ", DownloadSource.DEFAULT: "Qwen/Qwen2-VL-7B-Instruct-AWQ",
DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-7B-Instruct-AWQ", DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-7B-Instruct-AWQ",
}, },
"Qwen2VL-72B-Instruct-GPTQ-Int8": { "Qwen2-VL-72B-Instruct-GPTQ-Int8": {
DownloadSource.DEFAULT: "Qwen/Qwen2-VL-72B-Instruct-GPTQ-Int8", DownloadSource.DEFAULT: "Qwen/Qwen2-VL-72B-Instruct-GPTQ-Int8",
DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-72B-Instruct-GPTQ-Int8", DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-72B-Instruct-GPTQ-Int8",
}, },
"Qwen2VL-72B-Instruct-GPTQ-Int4": { "Qwen2-VL-72B-Instruct-GPTQ-Int4": {
DownloadSource.DEFAULT: "Qwen/Qwen2-VL-72B-Instruct-GPTQ-Int4", DownloadSource.DEFAULT: "Qwen/Qwen2-VL-72B-Instruct-GPTQ-Int4",
DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-72B-Instruct-GPTQ-Int4", DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-72B-Instruct-GPTQ-Int4",
}, },
"Qwen2VL-72B-Instruct-AWQ": { "Qwen2-VL-72B-Instruct-AWQ": {
DownloadSource.DEFAULT: "Qwen/Qwen2-VL-72B-Instruct-AWQ", DownloadSource.DEFAULT: "Qwen/Qwen2-VL-72B-Instruct-AWQ",
DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-72B-Instruct-AWQ", DownloadSource.MODELSCOPE: "qwen/Qwen2-VL-72B-Instruct-AWQ",
}, },
@ -1689,11 +1817,11 @@ register_model_group(
register_model_group( register_model_group(
models={ models={
"Vicuna1.5-7B-Chat": { "Vicuna-v1.5-7B-Chat": {
DownloadSource.DEFAULT: "lmsys/vicuna-7b-v1.5", DownloadSource.DEFAULT: "lmsys/vicuna-7b-v1.5",
DownloadSource.MODELSCOPE: "Xorbits/vicuna-7b-v1.5", DownloadSource.MODELSCOPE: "Xorbits/vicuna-7b-v1.5",
}, },
"Vicuna1.5-13B-Chat": { "Vicuna-v1.5-13B-Chat": {
DownloadSource.DEFAULT: "lmsys/vicuna-13b-v1.5", DownloadSource.DEFAULT: "lmsys/vicuna-13b-v1.5",
DownloadSource.MODELSCOPE: "Xorbits/vicuna-13b-v1.5", DownloadSource.MODELSCOPE: "Xorbits/vicuna-13b-v1.5",
}, },
@ -1702,6 +1830,17 @@ register_model_group(
) )
register_model_group(
models={
"Video-LLaVA-7B-Chat": {
DownloadSource.DEFAULT: "LanguageBind/Video-LLaVA-7B-hf",
},
},
template="video_llava",
vision=True,
)
register_model_group( register_model_group(
models={ models={
"XuanYuan-6B": { "XuanYuan-6B": {
@ -1712,7 +1851,7 @@ register_model_group(
DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan-70B", DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan-70B",
DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan-70B", DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan-70B",
}, },
"XuanYuan-2-70B": { "XuanYuan2-70B": {
DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan2-70B", DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan2-70B",
DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan2-70B", DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan2-70B",
}, },
@ -1724,31 +1863,31 @@ register_model_group(
DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan-70B-Chat", DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan-70B-Chat",
DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan-70B-Chat", DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan-70B-Chat",
}, },
"XuanYuan-2-70B-Chat": { "XuanYuan2-70B-Chat": {
DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan2-70B-Chat", DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan2-70B-Chat",
DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan2-70B-Chat", DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan2-70B-Chat",
}, },
"XuanYuan-6B-int8-Chat": { "XuanYuan-6B-Chat-8bit": {
DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan-6B-Chat-8bit", DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan-6B-Chat-8bit",
DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan-6B-Chat-8bit", DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan-6B-Chat-8bit",
}, },
"XuanYuan-6B-int4-Chat": { "XuanYuan-6B-Chat-4bit": {
DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan-6B-Chat-4bit", DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan-6B-Chat-4bit",
DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan-6B-Chat-4bit", DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan-6B-Chat-4bit",
}, },
"XuanYuan-70B-int8-Chat": { "XuanYuan-70B-Chat-8bit": {
DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan-70B-Chat-8bit", DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan-70B-Chat-8bit",
DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan-70B-Chat-8bit", DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan-70B-Chat-8bit",
}, },
"XuanYuan-70B-int4-Chat": { "XuanYuan-70B-Chat-4bit": {
DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan-70B-Chat-4bit", DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan-70B-Chat-4bit",
DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan-70B-Chat-4bit", DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan-70B-Chat-4bit",
}, },
"XuanYuan-2-70B-int8-Chat": { "XuanYuan2-70B-Chat-8bit": {
DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan2-70B-Chat-8bit", DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan2-70B-Chat-8bit",
DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan2-70B-Chat-8bit", DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan2-70B-Chat-8bit",
}, },
"XuanYuan-2-70B-int4-Chat": { "XuanYuan2-70B-Chat-4bit": {
DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan2-70B-Chat-4bit", DownloadSource.DEFAULT: "Duxiaoman-DI/XuanYuan2-70B-Chat-4bit",
DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan2-70B-Chat-4bit", DownloadSource.MODELSCOPE: "Duxiaoman-DI/XuanYuan2-70B-Chat-4bit",
}, },
@ -1853,19 +1992,19 @@ register_model_group(
DownloadSource.DEFAULT: "01-ai/Yi-34B-Chat", DownloadSource.DEFAULT: "01-ai/Yi-34B-Chat",
DownloadSource.MODELSCOPE: "01ai/Yi-34B-Chat", DownloadSource.MODELSCOPE: "01ai/Yi-34B-Chat",
}, },
"Yi-6B-int8-Chat": { "Yi-6B-Chat-8bits": {
DownloadSource.DEFAULT: "01-ai/Yi-6B-Chat-8bits", DownloadSource.DEFAULT: "01-ai/Yi-6B-Chat-8bits",
DownloadSource.MODELSCOPE: "01ai/Yi-6B-Chat-8bits", DownloadSource.MODELSCOPE: "01ai/Yi-6B-Chat-8bits",
}, },
"Yi-6B-int4-Chat": { "Yi-6B-Chat-4bits": {
DownloadSource.DEFAULT: "01-ai/Yi-6B-Chat-4bits", DownloadSource.DEFAULT: "01-ai/Yi-6B-Chat-4bits",
DownloadSource.MODELSCOPE: "01ai/Yi-6B-Chat-4bits", DownloadSource.MODELSCOPE: "01ai/Yi-6B-Chat-4bits",
}, },
"Yi-34B-int8-Chat": { "Yi-34B-Chat-8bits": {
DownloadSource.DEFAULT: "01-ai/Yi-34B-Chat-8bits", DownloadSource.DEFAULT: "01-ai/Yi-34B-Chat-8bits",
DownloadSource.MODELSCOPE: "01ai/Yi-34B-Chat-8bits", DownloadSource.MODELSCOPE: "01ai/Yi-34B-Chat-8bits",
}, },
"Yi-34B-int4-Chat": { "Yi-34B-Chat-4bits": {
DownloadSource.DEFAULT: "01-ai/Yi-34B-Chat-4bits", DownloadSource.DEFAULT: "01-ai/Yi-34B-Chat-4bits",
DownloadSource.MODELSCOPE: "01ai/Yi-34B-Chat-4bits", DownloadSource.MODELSCOPE: "01ai/Yi-34B-Chat-4bits",
}, },
@ -1916,10 +2055,10 @@ register_model_group(
register_model_group( register_model_group(
models={ models={
"YiVL-6B-Chat": { "Yi-VL-6B-Chat": {
DownloadSource.DEFAULT: "BUAADreamer/Yi-VL-6B-hf", DownloadSource.DEFAULT: "BUAADreamer/Yi-VL-6B-hf",
}, },
"YiVL-34B-Chat": { "Yi-VL-34B-Chat": {
DownloadSource.DEFAULT: "BUAADreamer/Yi-VL-34B-hf", DownloadSource.DEFAULT: "BUAADreamer/Yi-VL-34B-hf",
}, },
}, },

30
src/llamafactory/hparams/model_args.py
View File

@ -15,7 +15,7 @@
# See the License for the specific language governing permissions and # See the License for the specific language governing permissions and
# limitations under the License. # limitations under the License.
from dataclasses import asdict, dataclass, field, fields from dataclasses import dataclass, field, fields
from typing import Any, Dict, Literal, Optional, Union from typing import Any, Dict, Literal, Optional, Union
import torch import torch
@ -308,20 +308,18 @@ class ModelArguments(QuantizationArguments, ProcessorArguments, ExportArguments,
if self.export_quantization_bit is not None and self.export_quantization_dataset is None: if self.export_quantization_bit is not None and self.export_quantization_dataset is None:
raise ValueError("Quantization dataset is necessary for exporting.") raise ValueError("Quantization dataset is necessary for exporting.")
def to_dict(self) -> Dict[str, Any]:
return asdict(self)
@classmethod @classmethod
def copyfrom(cls, old_arg: "Self", **kwargs) -> "Self": def copyfrom(cls, source: "Self", **kwargs) -> "Self":
arg_dict = old_arg.to_dict() init_args, lazy_args = {}, {}
arg_dict.update(**kwargs) for attr in fields(source):
for attr in fields(cls): if attr.init:
if not attr.init: init_args[attr.name] = getattr(source, attr.name)
arg_dict.pop(attr.name) else:
lazy_args[attr.name] = getattr(source, attr.name)
new_arg = cls(**arg_dict) init_args.update(kwargs)
new_arg.compute_dtype = old_arg.compute_dtype result = cls(**init_args)
new_arg.device_map = old_arg.device_map for name, value in lazy_args.items():
new_arg.model_max_length = old_arg.model_max_length setattr(result, name, value)
new_arg.block_diag_attn = old_arg.block_diag_attn
return new_arg return result

4
src/llamafactory/hparams/parser.py
View File

@ -57,7 +57,7 @@ def _parse_args(parser: "HfArgumentParser", args: Optional[Dict[str, Any]] = Non
if args is not None: if args is not None:
return parser.parse_dict(args) return parser.parse_dict(args)
if len(sys.argv) == 2 and sys.argv[1].endswith(".yaml"): if len(sys.argv) == 2 and (sys.argv[1].endswith(".yaml") or sys.argv[1].endswith(".yml")):
return parser.parse_yaml_file(os.path.abspath(sys.argv[1])) return parser.parse_yaml_file(os.path.abspath(sys.argv[1]))
if len(sys.argv) == 2 and sys.argv[1].endswith(".json"): if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
@ -123,7 +123,7 @@ def _check_extra_dependencies(
require_version("mixture-of-depth>=1.1.6", "To fix: pip install mixture-of-depth>=1.1.6") require_version("mixture-of-depth>=1.1.6", "To fix: pip install mixture-of-depth>=1.1.6")
if model_args.infer_backend == "vllm": if model_args.infer_backend == "vllm":
require_version("vllm>=0.4.3,<=0.6.0", "To fix: pip install vllm>=0.4.3,<=0.6.0") require_version("vllm>=0.4.3,<=0.6.2", "To fix: pip install vllm>=0.4.3,<=0.6.2")
if finetuning_args.use_galore: if finetuning_args.use_galore:
require_version("galore_torch", "To fix: pip install galore_torch") require_version("galore_torch", "To fix: pip install galore_torch")

16
src/llamafactory/model/loader.py
View File

@ -21,12 +21,12 @@ from trl import AutoModelForCausalLMWithValueHead
from ..extras.logging import get_logger from ..extras.logging import get_logger
from ..extras.misc import count_parameters, skip_check_imports, try_download_model_from_ms from ..extras.misc import count_parameters, skip_check_imports, try_download_model_from_ms
from .adapter import init_adapter from .adapter import init_adapter
from .model_utils.liger_kernel import apply_liger_kernel
from .model_utils.misc import register_autoclass from .model_utils.misc import register_autoclass
from .model_utils.mod import convert_pretrained_model_to_mod, load_mod_pretrained_model from .model_utils.mod import convert_pretrained_model_to_mod, load_mod_pretrained_model
from .model_utils.unsloth import load_unsloth_pretrained_model from .model_utils.unsloth import load_unsloth_pretrained_model
from .model_utils.valuehead import load_valuehead_params from .model_utils.valuehead import load_valuehead_params
from .model_utils.visual import get_image_seqlen from .patcher import patch_config, patch_model, patch_processor, patch_tokenizer, patch_valuehead_model
from .patcher import patch_config, patch_model, patch_tokenizer, patch_valuehead_model
if TYPE_CHECKING: if TYPE_CHECKING:
@ -61,7 +61,7 @@ def _get_init_kwargs(model_args: "ModelArguments") -> Dict[str, Any]:
def load_tokenizer(model_args: "ModelArguments") -> "TokenizerModule": def load_tokenizer(model_args: "ModelArguments") -> "TokenizerModule":
r""" r"""
Loads pretrained tokenizer. Loads pretrained tokenizer and optionally loads processor.
Note: including inplace operation of model_args. Note: including inplace operation of model_args.
""" """
@ -96,15 +96,9 @@ def load_tokenizer(model_args: "ModelArguments") -> "TokenizerModule":
logger.warning("New tokens have been added, changed `resize_vocab` to True.") logger.warning("New tokens have been added, changed `resize_vocab` to True.")
patch_tokenizer(tokenizer) patch_tokenizer(tokenizer)
try: try:
processor = AutoProcessor.from_pretrained(model_args.model_name_or_path, **init_kwargs) processor = AutoProcessor.from_pretrained(model_args.model_name_or_path, **init_kwargs)
setattr(processor, "tokenizer", tokenizer) patch_processor(processor, config, tokenizer, model_args)
setattr(processor, "image_seqlen", get_image_seqlen(config))
setattr(processor, "image_resolution", model_args.image_resolution)
setattr(processor, "video_resolution", model_args.video_resolution)
setattr(processor, "video_fps", model_args.video_fps)
setattr(processor, "video_maxlen", model_args.video_maxlen)
except Exception as e: except Exception as e:
logger.warning("Failed to load processor. Error: {}".format(e)) logger.warning("Failed to load processor. Error: {}".format(e))
processor = None processor = None
@ -138,6 +132,7 @@ def load_model(
init_kwargs = _get_init_kwargs(model_args) init_kwargs = _get_init_kwargs(model_args)
config = load_config(model_args) config = load_config(model_args)
patch_config(config, tokenizer, model_args, init_kwargs, is_trainable) patch_config(config, tokenizer, model_args, init_kwargs, is_trainable)
apply_liger_kernel(config, model_args, is_trainable, require_logits=(finetuning_args.stage not in ["pt", "sft"]))
model = None model = None
lazy_load = False lazy_load = False
@ -158,7 +153,6 @@ def load_model(
load_class = AutoModelForVision2Seq load_class = AutoModelForVision2Seq
else: else:
load_class = AutoModelForCausalLM load_class = AutoModelForCausalLM
if model_args.train_from_scratch: if model_args.train_from_scratch:
model = load_class.from_config(config) model = load_class.from_config(config)
else: else:

7
src/llamafactory/model/model_utils/attention.py
View File

@ -37,10 +37,11 @@ def configure_attn_implementation(
if is_flash_attn_2_available(): if is_flash_attn_2_available():
require_version("transformers>=4.42.4", "To fix: pip install transformers>=4.42.4") require_version("transformers>=4.42.4", "To fix: pip install transformers>=4.42.4")
require_version("flash_attn>=2.6.3", "To fix: pip install flash_attn>=2.6.3") require_version("flash_attn>=2.6.3", "To fix: pip install flash_attn>=2.6.3")
logger.warning("Gemma-2 should use flash attention 2, change `flash_attn` to fa2.") if model_args.flash_attn != "fa2":
model_args.flash_attn = "fa2" logger.warning("Gemma-2 should use flash attention 2, change `flash_attn` to fa2.")
model_args.flash_attn = "fa2"
else: else:
logger.warning("Gemma-2 should use eager attention, change `flash_attn` to disabled.") logger.warning("FlashAttention-2 is not installed, use eager attention.")
model_args.flash_attn = "disabled" model_args.flash_attn = "disabled"
elif model_args.flash_attn == "sdpa": elif model_args.flash_attn == "sdpa":
logger.warning("Gemma-2 should use soft-capping attention, while the SDPA attention does not support it.") logger.warning("Gemma-2 should use soft-capping attention, while the SDPA attention does not support it.")

16
src/llamafactory/model/model_utils/liger_kernel.py
View File

@ -12,6 +12,7 @@
# See the License for the specific language governing permissions and # See the License for the specific language governing permissions and
# limitations under the License. # limitations under the License.
import inspect
from typing import TYPE_CHECKING from typing import TYPE_CHECKING
from ...extras.logging import get_logger from ...extras.logging import get_logger
@ -26,7 +27,12 @@ if TYPE_CHECKING:
logger = get_logger(__name__) logger = get_logger(__name__)
def configure_liger_kernel(config: "PretrainedConfig", model_args: "ModelArguments", is_trainable: bool) -> None: def apply_liger_kernel(
config: "PretrainedConfig",
model_args: "ModelArguments",
is_trainable: bool,
require_logits: bool,
) -> None:
if not is_trainable or not model_args.enable_liger_kernel: if not is_trainable or not model_args.enable_liger_kernel:
return return
@ -51,5 +57,11 @@ def configure_liger_kernel(config: "PretrainedConfig", model_args: "ModelArgumen
logger.warning("Current model does not support liger kernel.") logger.warning("Current model does not support liger kernel.")
return return
apply_liger_kernel() if require_logits and "fused_linear_cross_entropy" in inspect.signature(apply_liger_kernel).parameters:
logger.info("Current training stage does not support chunked cross entropy.")
kwargs = {"fused_linear_cross_entropy": False}
else:
kwargs = {}
apply_liger_kernel(**kwargs)
logger.info("Liger kernel has been applied to the model.") logger.info("Liger kernel has been applied to the model.")

2
src/llamafactory/model/model_utils/misc.py
View File

@ -34,7 +34,7 @@ def find_all_linear_modules(model: "PreTrainedModel", freeze_vision_tower: bool)
forbidden_modules.add("output_layer") forbidden_modules.add("output_layer")
elif model_type == "internlm2": elif model_type == "internlm2":
forbidden_modules.add("output") forbidden_modules.add("output")
elif model_type in ["llava", "paligemma"]: elif model_type in ["llava", "llava_next", "llava_next_video", "paligemma", "video_llava"]:
forbidden_modules.add("multi_modal_projector") forbidden_modules.add("multi_modal_projector")
elif model_type == "qwen2_vl": elif model_type == "qwen2_vl":
forbidden_modules.add("merger") forbidden_modules.add("merger")

32
src/llamafactory/model/model_utils/visual.py
View File

@ -92,7 +92,7 @@ def autocast_projector_dtype(model: "PreTrainedModel", model_args: "ModelArgumen
if getattr(model, "quantization_method", None): if getattr(model, "quantization_method", None):
model_type = getattr(model.config, "model_type", None) model_type = getattr(model.config, "model_type", None)
if model_type in ["llava", "paligemma"]: if model_type in ["llava", "llava_next", "llava_next_video", "paligemma", "video_llava"]:
mm_projector: "torch.nn.Module" = getattr(model, "multi_modal_projector") mm_projector: "torch.nn.Module" = getattr(model, "multi_modal_projector")
elif model_type == "qwen2_vl": elif model_type == "qwen2_vl":
mm_projector: "torch.nn.Module" = getattr(getattr(model, "visual"), "merger") mm_projector: "torch.nn.Module" = getattr(getattr(model, "visual"), "merger")
@ -108,7 +108,13 @@ def configure_visual_model(config: "PretrainedConfig") -> None:
Patches VLMs before loading them. Patches VLMs before loading them.
""" """
model_type = getattr(config, "model_type", None) model_type = getattr(config, "model_type", None)
if model_type == "llava": # required for ds zero3 and valuehead models if model_type in [
"llava",
"llava_next",
"llava_next_video",
"paligemma",
"video_llava",
]: # required for ds zero3 and valuehead models
setattr(config, "hidden_size", getattr(config.text_config, "hidden_size", None)) setattr(config, "hidden_size", getattr(config.text_config, "hidden_size", None))
if getattr(config, "is_yi_vl_derived_model", None): if getattr(config, "is_yi_vl_derived_model", None):
@ -122,7 +128,7 @@ def get_forbidden_modules(config: "PretrainedConfig", finetuning_args: "Finetuni
""" """
model_type = getattr(config, "model_type", None) model_type = getattr(config, "model_type", None)
forbidden_modules = set() forbidden_modules = set()
if model_type in ["llava", "paligemma"]: if model_type in ["llava", "llava_next", "llava_next_video", "paligemma", "video_llava"]:
if finetuning_args.freeze_vision_tower: if finetuning_args.freeze_vision_tower:
forbidden_modules.add("vision_tower") forbidden_modules.add("vision_tower")
@ -150,12 +156,28 @@ def get_image_seqlen(config: "PretrainedConfig") -> int:
image_seqlen += 1 image_seqlen += 1
elif model_type == "paligemma": elif model_type == "paligemma":
image_seqlen = config.vision_config.num_image_tokens image_seqlen = config.vision_config.num_image_tokens
elif model_type == "qwen2_vl": # variable length else:
image_seqlen = -1 image_seqlen = -1
return image_seqlen return image_seqlen
def get_patch_size(config: "PretrainedConfig") -> int:
r"""
Computes the patch size of the vit.
"""
patch_size = getattr(config.vision_config, "patch_size", -1)
return patch_size
def get_vision_feature_select_strategy(config: "PretrainedConfig") -> int:
r"""
Get the vision_feature_select_strategy.
"""
vision_feature_select_strategy = getattr(config, "vision_feature_select_strategy", "default")
return vision_feature_select_strategy
def patch_target_modules( def patch_target_modules(
config: "PretrainedConfig", finetuning_args: "FinetuningArguments", target_modules: Sequence[str] config: "PretrainedConfig", finetuning_args: "FinetuningArguments", target_modules: Sequence[str]
) -> Union[str, List[str]]: ) -> Union[str, List[str]]:
@ -164,7 +186,7 @@ def patch_target_modules(
""" """
model_type = getattr(config, "model_type", None) model_type = getattr(config, "model_type", None)
if finetuning_args.freeze_vision_tower: if finetuning_args.freeze_vision_tower:
if model_type in ["llava", "paligemma"]: if model_type in ["llava", "llava_next", "llava_next_video", "paligemma", "video_llava"]:
return "^(?!.*vision_tower).*(?:{}).*".format("|".join(target_modules)) return "^(?!.*vision_tower).*(?:{}).*".format("|".join(target_modules))
elif model_type == "qwen2_vl": elif model_type == "qwen2_vl":
return "^(?!.*visual).*(?:{}).*".format("|".join(target_modules)) return "^(?!.*visual).*(?:{}).*".format("|".join(target_modules))

31
src/llamafactory/model/patcher.py
View File

@ -27,18 +27,23 @@ from ..extras.misc import infer_optim_dtype
from .model_utils.attention import configure_attn_implementation, print_attn_implementation from .model_utils.attention import configure_attn_implementation, print_attn_implementation
from .model_utils.checkpointing import prepare_model_for_training from .model_utils.checkpointing import prepare_model_for_training
from .model_utils.embedding import resize_embedding_layer from .model_utils.embedding import resize_embedding_layer
from .model_utils.liger_kernel import configure_liger_kernel
from .model_utils.longlora import configure_longlora from .model_utils.longlora import configure_longlora
from .model_utils.moe import add_z3_leaf_module, configure_moe from .model_utils.moe import add_z3_leaf_module, configure_moe
from .model_utils.packing import configure_packing from .model_utils.packing import configure_packing
from .model_utils.quantization import configure_quantization from .model_utils.quantization import configure_quantization
from .model_utils.rope import configure_rope from .model_utils.rope import configure_rope
from .model_utils.valuehead import prepare_valuehead_model from .model_utils.valuehead import prepare_valuehead_model
from .model_utils.visual import autocast_projector_dtype, configure_visual_model from .model_utils.visual import (
autocast_projector_dtype,
configure_visual_model,
get_image_seqlen,
get_patch_size,
get_vision_feature_select_strategy,
)
if TYPE_CHECKING: if TYPE_CHECKING:
from transformers import PretrainedConfig, PreTrainedTokenizer from transformers import PretrainedConfig, PreTrainedTokenizer, ProcessorMixin
from trl import AutoModelForCausalLMWithValueHead from trl import AutoModelForCausalLMWithValueHead
from ..hparams import ModelArguments from ..hparams import ModelArguments
@ -52,6 +57,22 @@ def patch_tokenizer(tokenizer: "PreTrainedTokenizer") -> None:
tokenizer._pad = MethodType(PreTrainedTokenizerBase._pad, tokenizer) tokenizer._pad = MethodType(PreTrainedTokenizerBase._pad, tokenizer)
def patch_processor(
processor: "ProcessorMixin",
config: "PretrainedConfig",
tokenizer: "PreTrainedTokenizer",
model_args: "ModelArguments",
) -> None:
setattr(processor, "tokenizer", tokenizer)
setattr(processor, "image_seqlen", get_image_seqlen(config))
setattr(processor, "image_resolution", model_args.image_resolution)
setattr(processor, "patch_size", get_patch_size(config))
setattr(processor, "video_resolution", model_args.video_resolution)
setattr(processor, "video_fps", model_args.video_fps)
setattr(processor, "video_maxlen", model_args.video_maxlen)
setattr(processor, "vision_feature_select_strategy", get_vision_feature_select_strategy(config))
def patch_config( def patch_config(
config: "PretrainedConfig", config: "PretrainedConfig",
tokenizer: "PreTrainedTokenizer", tokenizer: "PreTrainedTokenizer",
@ -71,7 +92,6 @@ def patch_config(
configure_attn_implementation(config, model_args, is_trainable) configure_attn_implementation(config, model_args, is_trainable)
configure_rope(config, model_args, is_trainable) configure_rope(config, model_args, is_trainable)
configure_liger_kernel(config, model_args, is_trainable)
configure_longlora(config, model_args, is_trainable) configure_longlora(config, model_args, is_trainable)
configure_quantization(config, tokenizer, model_args, init_kwargs) configure_quantization(config, tokenizer, model_args, init_kwargs)
configure_moe(config, model_args, is_trainable) configure_moe(config, model_args, is_trainable)
@ -90,6 +110,9 @@ def patch_config(
if getattr(config, "model_type", None) == "qwen2" and is_trainable and model_args.flash_attn == "fa2": if getattr(config, "model_type", None) == "qwen2" and is_trainable and model_args.flash_attn == "fa2":
setattr(config, "use_cache", False) # qwen2 does not support use_cache when using flash attn setattr(config, "use_cache", False) # qwen2 does not support use_cache when using flash attn
if "LlavaLlamaForCausalLM" in getattr(config, "architectures", []):
raise ValueError("Please download llava models with hf-compatible format: https://huggingface.co/llava-hf")
# deepspeed zero3 is not compatible with low_cpu_mem_usage # deepspeed zero3 is not compatible with low_cpu_mem_usage
init_kwargs["low_cpu_mem_usage"] = model_args.low_cpu_mem_usage and (not is_deepspeed_zero3_enabled()) init_kwargs["low_cpu_mem_usage"] = model_args.low_cpu_mem_usage and (not is_deepspeed_zero3_enabled())

18
src/llamafactory/webui/common.py
View File

@ -115,13 +115,6 @@ def get_model_path(model_name: str) -> str:
return model_path return model_path
def get_prefix(model_name: str) -> str:
r"""
Gets the prefix of the model name to obtain the model family.
"""
return model_name.split("-")[0]
def get_model_info(model_name: str) -> Tuple[str, str]: def get_model_info(model_name: str) -> Tuple[str, str]:
r""" r"""
Gets the necessary information of this model. Gets the necessary information of this model.
@ -137,21 +130,14 @@ def get_template(model_name: str) -> str:
r""" r"""
Gets the template name if the model is a chat model. Gets the template name if the model is a chat model.
""" """
if ( return DEFAULT_TEMPLATE.get(model_name, "default")
model_name
and any(suffix in model_name for suffix in ("-Chat", "-Instruct"))
and get_prefix(model_name) in DEFAULT_TEMPLATE
):
return DEFAULT_TEMPLATE[get_prefix(model_name)]
return "default"
def get_visual(model_name: str) -> bool: def get_visual(model_name: str) -> bool:
r""" r"""
Judges if the model is a vision language model. Judges if the model is a vision language model.
""" """
return get_prefix(model_name) in VISION_MODELS return model_name in VISION_MODELS
def list_checkpoints(model_name: str, finetuning_type: str) -> "gr.Dropdown": def list_checkpoints(model_name: str, finetuning_type: str) -> "gr.Dropdown":

39
tests/data/test_mm_plugin.py
View File

@ -136,6 +136,32 @@ def test_llava_plugin():
_check_plugin(**check_inputs) _check_plugin(**check_inputs)
def test_llava_next_plugin():
tokenizer, processor = _load_tokenizer_module(model_name_or_path="llava-hf/llava-v1.6-vicuna-7b-hf")
llava_next_plugin = get_mm_plugin(name="llava_next", image_token="<image>")
check_inputs = {"plugin": llava_next_plugin, "tokenizer": tokenizer, "processor": processor}
image_seqlen = 1176
check_inputs["expected_mm_messages"] = [
{key: value.replace("<image>", "<image>" * image_seqlen) for key, value in message.items()}
for message in MM_MESSAGES
]
check_inputs["expected_mm_inputs"] = _get_mm_inputs(processor)
_check_plugin(**check_inputs)
def test_llava_next_video_plugin():
tokenizer, processor = _load_tokenizer_module(model_name_or_path="llava-hf/LLaVA-NeXT-Video-7B-hf")
llava_next_video_plugin = get_mm_plugin(name="llava_next_video", image_token="<image>", video_token="<video>")
check_inputs = {"plugin": llava_next_video_plugin, "tokenizer": tokenizer, "processor": processor}
image_seqlen = 1176
check_inputs["expected_mm_messages"] = [
{key: value.replace("<image>", "<image>" * image_seqlen) for key, value in message.items()}
for message in MM_MESSAGES
]
check_inputs["expected_mm_inputs"] = _get_mm_inputs(processor)
_check_plugin(**check_inputs)
@pytest.mark.skipif(not HF_TOKEN, reason="Gated model.") @pytest.mark.skipif(not HF_TOKEN, reason="Gated model.")
def test_paligemma_plugin(): def test_paligemma_plugin():
tokenizer, processor = _load_tokenizer_module(model_name_or_path="google/paligemma-3b-pt-224") tokenizer, processor = _load_tokenizer_module(model_name_or_path="google/paligemma-3b-pt-224")
@ -167,3 +193,16 @@ def test_qwen2_vl_plugin():
] ]
check_inputs["expected_mm_inputs"] = _get_mm_inputs(processor) check_inputs["expected_mm_inputs"] = _get_mm_inputs(processor)
_check_plugin(**check_inputs) _check_plugin(**check_inputs)
def test_video_llava_plugin():
tokenizer, processor = _load_tokenizer_module(model_name_or_path="LanguageBind/Video-LLaVA-7B-hf")
video_llava_plugin = get_mm_plugin(name="video_llava", image_token="<image>", video_token="<video>")
check_inputs = {"plugin": video_llava_plugin, "tokenizer": tokenizer, "processor": processor}
image_seqlen = 256
check_inputs["expected_mm_messages"] = [
{key: value.replace("<image>", "<image>" * image_seqlen) for key, value in message.items()}
for message in MM_MESSAGES
]
check_inputs["expected_mm_inputs"] = _get_mm_inputs(processor)
_check_plugin(**check_inputs)