423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-08-02 19:52:50 +08:00
Code Issues Packages Projects Releases Wiki Activity

improve template, add phi4 model

Browse Source 
Former-commit-id: ae16ea755d581a5a288fb55f12481215f369b255
This commit is contained in:
hiyouga 2025-01-09 18:27:20 +00:00
parent b777fed171
commit b471def13d
5 changed files with 147 additions and 125 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
README.md
View File

@ -88,14 +88,16 @@ Compared to ChatGLM's [P-Tuning](https://github.com/THUDM/ChatGLM2-6B/tree/main/
## Changelog
[25/01/10] We supported fine-tuning the **[Phi-4](https://huggingface.co/microsoft/phi-4)** model.
[24/12/21] We supported using **[SwanLab](https://github.com/SwanHubX/SwanLab)** for experiment tracking and visualization. See [this section](#use-swanlab-logger) for details.
[24/11/27] We supported fine-tuning the **[Skywork-o1](https://huggingface.co/Skywork/Skywork-o1-Open-Llama-3.1-8B)** model and the **[OpenO1](https://huggingface.co/datasets/O1-OPEN/OpenO1-SFT)** dataset.
[24/10/09] We supported downloading pre-trained models and datasets from the **[Modelers Hub](https://modelers.cn/models)**. See [this tutorial](#download-from-modelers-hub) for usage.
<details><summary>Full Changelog</summary>
[24/10/09] We supported downloading pre-trained models and datasets from the **[Modelers Hub](https://modelers.cn/models)**. See [this tutorial](#download-from-modelers-hub) for usage.
[24/09/19] We supported fine-tuning the **[Qwen2.5](https://qwenlm.github.io/blog/qwen2.5/)** models.
[24/08/30] We supported fine-tuning the **[Qwen2-VL](https://qwenlm.github.io/blog/qwen2-vl/)** models. Thank [@simonJJJ](https://github.com/simonJJJ)'s PR.
@ -211,8 +213,9 @@ Compared to ChatGLM's [P-Tuning](https://github.com/THUDM/ChatGLM2-6B/tree/main/
| [OLMo](https://huggingface.co/allenai) | 1B/7B | - |
| [PaliGemma/PaliGemma2](https://huggingface.co/google) | 3B/10B/28B | paligemma |
| [Phi-1.5/Phi-2](https://huggingface.co/microsoft) | 1.3B/2.7B | - |
| [Phi-3](https://huggingface.co/microsoft) | 4B/14B | phi |
| [Phi-3/Phi-3.5](https://huggingface.co/microsoft) | 4B/14B | phi |
| [Phi-3-small](https://huggingface.co/microsoft) | 7B | phi_small |
| [Phi-4](https://huggingface.co/microsoft) | 14B | phi4 |
| [Pixtral](https://huggingface.co/mistralai) | 12B | pixtral |
| [Qwen/QwQ (1-2.5) (Code/Math/MoE)](https://huggingface.co/Qwen) | 0.5B/1.5B/3B/7B/14B/32B/72B/110B | qwen |
| [Qwen2-VL/QVQ](https://huggingface.co/Qwen) | 2B/7B/72B | qwen2_vl |
@ -762,7 +765,7 @@ If you have a project that should be incorporated, please contact via email or c
This repository is licensed under the [Apache-2.0 License](LICENSE).
Please follow the model licenses to use the corresponding model weights: [Baichuan 2](https://huggingface.co/baichuan-inc/Baichuan2-7B-Base/blob/main/Community%20License%20for%20Baichuan%202%20Model.pdf) / [BLOOM](https://huggingface.co/spaces/bigscience/license) / [ChatGLM3](https://github.com/THUDM/ChatGLM3/blob/main/MODEL_LICENSE) / [Command R](https://cohere.com/c4ai-cc-by-nc-license) / [DeepSeek](https://github.com/deepseek-ai/DeepSeek-LLM/blob/main/LICENSE-MODEL) / [Falcon](https://huggingface.co/tiiuae/falcon-180B/blob/main/LICENSE.txt) / [Gemma](https://ai.google.dev/gemma/terms) / [GLM-4](https://huggingface.co/THUDM/glm-4-9b/blob/main/LICENSE) / [GPT-2](https://github.com/openai/gpt-2/blob/master/LICENSE) / [Granite](LICENSE) / [Index](https://huggingface.co/IndexTeam/Index-1.9B/blob/main/LICENSE) / [InternLM2](https://github.com/InternLM/InternLM#license) / [Llama](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) / [Llama 2 (LLaVA-1.5)](https://ai.meta.com/llama/license/) / [Llama 3](https://llama.meta.com/llama3/license/) / [MiniCPM](https://github.com/OpenBMB/MiniCPM/blob/main/MiniCPM%20Model%20License.md) / [Mistral/Mixtral/Pixtral](LICENSE) / [OLMo](LICENSE) / [Phi-1.5/Phi-2](https://huggingface.co/microsoft/phi-1_5/resolve/main/Research%20License.docx) / [Phi-3](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/blob/main/LICENSE) / [Qwen](https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT) / [Skywork](https://huggingface.co/Skywork/Skywork-13B-base/blob/main/Skywork%20Community%20License.pdf) / [StarCoder 2](https://huggingface.co/spaces/bigcode/bigcode-model-license-agreement) / [TeleChat2](https://huggingface.co/Tele-AI/telechat-7B/blob/main/TeleChat%E6%A8%A1%E5%9E%8B%E7%A4%BE%E5%8C%BA%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE.pdf) / [XVERSE](https://github.com/xverse-ai/XVERSE-13B/blob/main/MODEL_LICENSE.pdf) / [Yi](https://huggingface.co/01-ai/Yi-6B/blob/main/LICENSE) / [Yi-1.5](LICENSE) / [Yuan 2](https://github.com/IEIT-Yuan/Yuan-2.0/blob/main/LICENSE-Yuan)
Please follow the model licenses to use the corresponding model weights: [Baichuan 2](https://huggingface.co/baichuan-inc/Baichuan2-7B-Base/blob/main/Community%20License%20for%20Baichuan%202%20Model.pdf) / [BLOOM](https://huggingface.co/spaces/bigscience/license) / [ChatGLM3](https://github.com/THUDM/ChatGLM3/blob/main/MODEL_LICENSE) / [Command R](https://cohere.com/c4ai-cc-by-nc-license) / [DeepSeek](https://github.com/deepseek-ai/DeepSeek-LLM/blob/main/LICENSE-MODEL) / [Falcon](https://huggingface.co/tiiuae/falcon-180B/blob/main/LICENSE.txt) / [Gemma](https://ai.google.dev/gemma/terms) / [GLM-4](https://huggingface.co/THUDM/glm-4-9b/blob/main/LICENSE) / [GPT-2](https://github.com/openai/gpt-2/blob/master/LICENSE) / [Granite](LICENSE) / [Index](https://huggingface.co/IndexTeam/Index-1.9B/blob/main/LICENSE) / [InternLM2](https://github.com/InternLM/InternLM#license) / [Llama](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) / [Llama 2 (LLaVA-1.5)](https://ai.meta.com/llama/license/) / [Llama 3](https://llama.meta.com/llama3/license/) / [MiniCPM](https://github.com/OpenBMB/MiniCPM/blob/main/MiniCPM%20Model%20License.md) / [Mistral/Mixtral/Pixtral](LICENSE) / [OLMo](LICENSE) / [Phi-1.5/Phi-2](https://huggingface.co/microsoft/phi-1_5/resolve/main/Research%20License.docx) / [Phi-3/Phi-4](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/blob/main/LICENSE) / [Qwen](https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT) / [Skywork](https://huggingface.co/Skywork/Skywork-13B-base/blob/main/Skywork%20Community%20License.pdf) / [StarCoder 2](https://huggingface.co/spaces/bigcode/bigcode-model-license-agreement) / [TeleChat2](https://huggingface.co/Tele-AI/telechat-7B/blob/main/TeleChat%E6%A8%A1%E5%9E%8B%E7%A4%BE%E5%8C%BA%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE.pdf) / [XVERSE](https://github.com/xverse-ai/XVERSE-13B/blob/main/MODEL_LICENSE.pdf) / [Yi](https://huggingface.co/01-ai/Yi-6B/blob/main/LICENSE) / [Yi-1.5](LICENSE) / [Yuan 2](https://github.com/IEIT-Yuan/Yuan-2.0/blob/main/LICENSE-Yuan)
## Citation

11
README_zh.md
View File

@ -89,14 +89,16 @@ https://github.com/user-attachments/assets/e6ce34b0-52d5-4f3e-a830-592106c4c272
## 更新日志
[25/01/10] 我们支持了 **[Phi-4](https://huggingface.co/microsoft/phi-4)** 模型的微调。
[24/12/21] 我们支持了使用 **[SwanLab](https://github.com/SwanHubX/SwanLab)** 跟踪与可视化实验。详细用法请参考 [此部分](#使用-swanlab-面板)。
[24/11/27] 我们支持了 **[Skywork-o1](https://huggingface.co/Skywork/Skywork-o1-Open-Llama-3.1-8B)** 模型的微调和 **[OpenO1](https://huggingface.co/datasets/O1-OPEN/OpenO1-SFT)** 数据集。
[24/10/09] 我们支持了从 **[魔乐社区](https://modelers.cn/models)** 下载预训练模型和数据集。详细用法请参照 [此教程](#从魔乐社区下载)。
<details><summary>展开日志</summary>
[24/10/09] 我们支持了从 **[魔乐社区](https://modelers.cn/models)** 下载预训练模型和数据集。详细用法请参照 [此教程](#从魔乐社区下载)。
[24/09/19] 我们支持了 **[Qwen2.5](https://qwenlm.github.io/blog/qwen2.5/)** 模型的微调。
[24/08/30] 我们支持了 **[Qwen2-VL](https://qwenlm.github.io/blog/qwen2-vl/)** 模型的微调。感谢 [@simonJJJ](https://github.com/simonJJJ) 的 PR。
@ -212,8 +214,9 @@ https://github.com/user-attachments/assets/e6ce34b0-52d5-4f3e-a830-592106c4c272
| [OLMo](https://huggingface.co/allenai) | 1B/7B | - |
| [PaliGemma/PaliGemma2](https://huggingface.co/google) | 3B/10B/28B | paligemma |
| [Phi-1.5/Phi-2](https://huggingface.co/microsoft) | 1.3B/2.7B | - |
| [Phi-3](https://huggingface.co/microsoft) | 4B/14B | phi |
| [Phi-3/Phi-3.5](https://huggingface.co/microsoft) | 4B/14B | phi |
| [Phi-3-small](https://huggingface.co/microsoft) | 7B | phi_small |
| [Phi-4](https://huggingface.co/microsoft) | 14B | phi4 |
| [Pixtral](https://huggingface.co/mistralai) | 12B | pixtral |
| [Qwen/QwQ (1-2.5) (Code/Math/MoE)](https://huggingface.co/Qwen) | 0.5B/1.5B/3B/7B/14B/32B/72B/110B | qwen |
| [Qwen2-VL/QVQ](https://huggingface.co/Qwen) | 2B/7B/72B | qwen2_vl |
@ -763,7 +766,7 @@ swanlab_run_name: test_run # 可选
本仓库的代码依照 [Apache-2.0](LICENSE) 协议开源。
使用模型权重时，请遵循对应的模型协议：[Baichuan 2](https://huggingface.co/baichuan-inc/Baichuan2-7B-Base/blob/main/Community%20License%20for%20Baichuan%202%20Model.pdf) / [BLOOM](https://huggingface.co/spaces/bigscience/license) / [ChatGLM3](https://github.com/THUDM/ChatGLM3/blob/main/MODEL_LICENSE) / [Command R](https://cohere.com/c4ai-cc-by-nc-license) / [DeepSeek](https://github.com/deepseek-ai/DeepSeek-LLM/blob/main/LICENSE-MODEL) / [Falcon](https://huggingface.co/tiiuae/falcon-180B/blob/main/LICENSE.txt) / [Gemma](https://ai.google.dev/gemma/terms) / [GLM-4](https://huggingface.co/THUDM/glm-4-9b/blob/main/LICENSE) / [GPT-2](https://github.com/openai/gpt-2/blob/master/LICENSE) / [Granite](LICENSE) / [Index](https://huggingface.co/IndexTeam/Index-1.9B/blob/main/LICENSE) / [InternLM2](https://github.com/InternLM/InternLM#license) / [Llama](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) / [Llama 2 (LLaVA-1.5)](https://ai.meta.com/llama/license/) / [Llama 3](https://llama.meta.com/llama3/license/) / [MiniCPM](https://github.com/OpenBMB/MiniCPM/blob/main/MiniCPM%20Model%20License.md) / [Mistral/Mixtral/Pixtral](LICENSE) / [OLMo](LICENSE) / [Phi-1.5/Phi-2](https://huggingface.co/microsoft/phi-1_5/resolve/main/Research%20License.docx) / [Phi-3](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/blob/main/LICENSE) / [Qwen](https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT) / [Skywork](https://huggingface.co/Skywork/Skywork-13B-base/blob/main/Skywork%20Community%20License.pdf) / [StarCoder 2](https://huggingface.co/spaces/bigcode/bigcode-model-license-agreement) / [TeleChat2](https://huggingface.co/Tele-AI/telechat-7B/blob/main/TeleChat%E6%A8%A1%E5%9E%8B%E7%A4%BE%E5%8C%BA%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE.pdf) / [XVERSE](https://github.com/xverse-ai/XVERSE-13B/blob/main/MODEL_LICENSE.pdf) / [Yi](https://huggingface.co/01-ai/Yi-6B/blob/main/LICENSE) / [Yi-1.5](LICENSE) / [Yuan 2](https://github.com/IEIT-Yuan/Yuan-2.0/blob/main/LICENSE-Yuan)
使用模型权重时，请遵循对应的模型协议：[Baichuan 2](https://huggingface.co/baichuan-inc/Baichuan2-7B-Base/blob/main/Community%20License%20for%20Baichuan%202%20Model.pdf) / [BLOOM](https://huggingface.co/spaces/bigscience/license) / [ChatGLM3](https://github.com/THUDM/ChatGLM3/blob/main/MODEL_LICENSE) / [Command R](https://cohere.com/c4ai-cc-by-nc-license) / [DeepSeek](https://github.com/deepseek-ai/DeepSeek-LLM/blob/main/LICENSE-MODEL) / [Falcon](https://huggingface.co/tiiuae/falcon-180B/blob/main/LICENSE.txt) / [Gemma](https://ai.google.dev/gemma/terms) / [GLM-4](https://huggingface.co/THUDM/glm-4-9b/blob/main/LICENSE) / [GPT-2](https://github.com/openai/gpt-2/blob/master/LICENSE) / [Granite](LICENSE) / [Index](https://huggingface.co/IndexTeam/Index-1.9B/blob/main/LICENSE) / [InternLM2](https://github.com/InternLM/InternLM#license) / [Llama](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) / [Llama 2 (LLaVA-1.5)](https://ai.meta.com/llama/license/) / [Llama 3](https://llama.meta.com/llama3/license/) / [MiniCPM](https://github.com/OpenBMB/MiniCPM/blob/main/MiniCPM%20Model%20License.md) / [Mistral/Mixtral/Pixtral](LICENSE) / [OLMo](LICENSE) / [Phi-1.5/Phi-2](https://huggingface.co/microsoft/phi-1_5/resolve/main/Research%20License.docx) / [Phi-3/Phi-4](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/blob/main/LICENSE) / [Qwen](https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT) / [Skywork](https://huggingface.co/Skywork/Skywork-13B-base/blob/main/Skywork%20Community%20License.pdf) / [StarCoder 2](https://huggingface.co/spaces/bigcode/bigcode-model-license-agreement) / [TeleChat2](https://huggingface.co/Tele-AI/telechat-7B/blob/main/TeleChat%E6%A8%A1%E5%9E%8B%E7%A4%BE%E5%8C%BA%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE.pdf) / [XVERSE](https://github.com/xverse-ai/XVERSE-13B/blob/main/MODEL_LICENSE.pdf) / [Yi](https://huggingface.co/01-ai/Yi-6B/blob/main/LICENSE) / [Yi-1.5](LICENSE) / [Yuan 2](https://github.com/IEIT-Yuan/Yuan-2.0/blob/main/LICENSE-Yuan)
## 引用

144
src/llamafactory/data/template.py
View File

@ -44,7 +44,6 @@ class Template:
format_function: "Formatter"
format_observation: "Formatter"
format_tools: "Formatter"
format_separator: "Formatter"
format_prefix: "Formatter"
default_system: str
stop_words: List[str]
@ -113,9 +112,6 @@ class Template:
tool_text = self.format_tools.apply(content=tools)[0] if tools else ""
elements += self.format_system.apply(content=(system + tool_text))
if i > 0 and i % 2 == 0:
elements += self.format_separator.apply()
if message["role"] == Role.USER.value:
elements += self.format_user.apply(content=message["content"], idx=str(i // 2))
elif message["role"] == Role.ASSISTANT.value:
@ -180,9 +176,6 @@ class Llama2Template(Template):
tool_text = self.format_tools.apply(content=tools)[0] if tools else ""
system_text = self.format_system.apply(content=(system + tool_text))[0]
if i > 0 and i % 2 == 0:
elements += self.format_separator.apply()
if message["role"] == Role.USER.value:
elements += self.format_user.apply(content=system_text + message["content"])
elif message["role"] == Role.ASSISTANT.value:
@ -210,7 +203,6 @@ def _register_template(
format_function: Optional["Formatter"] = None,
format_observation: Optional["Formatter"] = None,
format_tools: Optional["Formatter"] = None,
format_separator: Optional["Formatter"] = None,
format_prefix: Optional["Formatter"] = None,
default_system: str = "",
stop_words: Sequence[str] = [],
@ -224,34 +216,28 @@ def _register_template(
To add the following chat template:
```
[HUMAN]:
user prompt here
[AI]:
model response here
[HUMAN]:
user prompt here
[AI]:
model response here
<s><user>user prompt here
<model>model response here</s>
<user>user prompt here
<model>model response here</s>
```
The corresponding code should be:
```
_register_template(
name="custom",
format_user=StringFormatter(slots=["[HUMAN]:\n{{content}}\n[AI]:\n"]),
format_separator=EmptyFormatter(slots=["\n\n"]),
efficient_eos=True,
format_user=StringFormatter(slots=["<user>{{content}}\n<model>"]),
format_assistant=StringFormatter(slots=["{{content}}</s>\n"]),
format_prefix=EmptyFormatter("<s>"),
)
```
"""
template_class = Llama2Template if any(k in name for k in ("llama2", "mistral")) else Template
template_class = Llama2Template if any(k in name for k in ("llama2", "mistral", "pixtral")) else Template
default_slots = ["{{content}}"] if efficient_eos else ["{{content}}", {"eos_token"}]
default_user_formatter = StringFormatter(slots=["{{content}}"])
default_assistant_formatter = StringFormatter(slots=default_slots)
default_function_formatter = FunctionFormatter(slots=default_slots, tool_format="default")
default_tool_formatter = ToolFormatter(tool_format="default")
default_separator_formatter = EmptyFormatter()
default_prefix_formatter = EmptyFormatter()
TEMPLATES[name] = template_class(
format_user=format_user or default_user_formatter,
@ -260,7 +246,6 @@ def _register_template(
format_function=format_function or default_function_formatter,
format_observation=format_observation or format_user or default_user_formatter,
format_tools=format_tools or default_tool_formatter,
format_separator=format_separator or default_separator_formatter,
format_prefix=format_prefix or default_prefix_formatter,
default_system=default_system,
stop_words=stop_words,
@ -344,9 +329,7 @@ def _get_jinja_template(template: "Template", tokenizer: "PreTrainedTokenizer")
jinja_template += "{{ " + user_message + " }}"
jinja_template += "{% elif message['role'] == 'assistant' %}"
assistant_message = _convert_slots_to_jinja(
template.format_assistant.apply() + template.format_separator.apply(), tokenizer
)
assistant_message = _convert_slots_to_jinja(template.format_assistant.apply(), tokenizer)
jinja_template += "{{ " + assistant_message + " }}"
jinja_template += "{% endif %}"
jinja_template += "{% endfor %}"
@ -411,7 +394,7 @@ def get_template_and_fix_tokenizer(tokenizer: "PreTrainedTokenizer", data_args:
_register_template(
name="alpaca",
format_user=StringFormatter(slots=["### Instruction:\n{{content}}\n\n### Response:\n"]),
format_separator=EmptyFormatter(slots=["\n\n"]),
format_assistant=StringFormatter(slots=["{{content}}", {"eos_token"}, "\n\n"]),
default_system=(
"Below is an instruction that describes a task. "
"Write a response that appropriately completes the request.\n\n"
@ -423,13 +406,13 @@ _register_template(
_register_template(
name="aquila",
format_user=StringFormatter(slots=["Human: {{content}}###Assistant:"]),
format_separator=EmptyFormatter(slots=["###"]),
format_assistant=StringFormatter(slots=["{{content}}###"]),
format_system=StringFormatter(slots=["System: {{content}}###"]),
default_system=(
"A chat between a curious human and an artificial intelligence assistant. "
"The assistant gives helpful, detailed, and polite answers to the human's questions."
),
stop_words=["</s>"],
efficient_eos=True,
)
@ -459,7 +442,7 @@ _register_template(
_register_template(
name="belle",
format_user=StringFormatter(slots=["Human: {{content}}\n\nBelle: "]),
format_separator=EmptyFormatter(slots=["\n\n"]),
format_assistant=StringFormatter(slots=["{{content}}", {"eos_token"}, "\n\n"]),
format_prefix=EmptyFormatter(slots=[{"bos_token"}]),
)
@ -481,7 +464,6 @@ _register_template(
_register_template(
name="chatglm2",
format_user=StringFormatter(slots=["[Round {{idx}}]\n\n问：{{content}}\n\n答："]),
format_separator=EmptyFormatter(slots=["\n\n"]),
format_prefix=EmptyFormatter(slots=[{"token": "[gMASK]"}, {"token": "sop"}]),
efficient_eos=True,
)
@ -506,9 +488,9 @@ _register_template(
_register_template(
name="chatml",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\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"]),
stop_words=["<|im_end|>", "<|im_start|>"],
replace_eos=True,
replace_jinja_template=True,
@ -519,9 +501,9 @@ _register_template(
_register_template(
name="chatml_de",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\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="Du bist ein freundlicher und hilfsbereiter KI-Assistent.",
stop_words=["<|im_end|>", "<|im_start|>"],
replace_eos=True,
@ -574,9 +556,11 @@ _register_template(
)
# copied from chatml template
_register_template(
name="cpm3",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\n"]),
format_system=StringFormatter(slots=["<|im_start|>system\n{{content}}<|im_end|>\n"]),
format_prefix=EmptyFormatter(slots=[{"bos_token"}]),
stop_words=["<|im_end|>"],
@ -587,9 +571,9 @@ _register_template(
_register_template(
name="dbrx",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\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 DBRX, created by Databricks. You were last updated in December 2023. "
"You answer questions based on information available up to that point.\n"
@ -606,7 +590,6 @@ _register_template(
"ABOUT YOURSELF UNLESS THE INFORMATION IS DIRECTLY PERTINENT TO THE USER'S QUERY."
),
stop_words=["<|im_end|>"],
replace_eos=True,
)
@ -628,8 +611,7 @@ _register_template(
_register_template(
name="deepseekcoder",
format_user=StringFormatter(slots=["### Instruction:\n{{content}}\n### Response:"]),
format_assistant=StringFormatter(slots=["\n{{content}}\n<|EOT|>"]),
format_separator=EmptyFormatter(slots=["\n"]),
format_assistant=StringFormatter(slots=["\n{{content}}\n<|EOT|>\n"]),
format_prefix=EmptyFormatter(slots=[{"bos_token"}]),
default_system=(
"You are an AI programming assistant, utilizing the DeepSeek Coder model, "
@ -643,8 +625,8 @@ _register_template(
_register_template(
name="default",
format_user=StringFormatter(slots=["Human: {{content}}\nAssistant:"]),
format_system=StringFormatter(slots=["{{content}}\n"]),
format_separator=EmptyFormatter(slots=["\n"]),
format_assistant=StringFormatter(slots=["{{content}}", {"eos_token"}, "\n"]),
format_system=StringFormatter(slots=["System: {{content}}\n"]),
)
@ -657,22 +639,22 @@ _register_template(
_register_template(
name="exaone",
format_user=StringFormatter(slots=["[|user|]{{content}}\n[|assistant|]"]),
format_assistant=StringFormatter(slots=["{{content}}", {"eos_token"}, "\n"]),
format_system=StringFormatter(slots=["[|system|]{{content}}[|endofturn|]\n"]),
format_separator=EmptyFormatter(slots=["\n"]),
)
_register_template(
name="falcon",
format_user=StringFormatter(slots=["User: {{content}}\nFalcon:"]),
format_separator=EmptyFormatter(slots=["\n"]),
format_assistant=StringFormatter(slots=["{{content}}\n"]),
efficient_eos=True,
)
_register_template(
name="fewshot",
format_separator=EmptyFormatter(slots=["\n\n"]),
format_assistant=StringFormatter(slots=["{{content}}\n\n"]),
efficient_eos=True,
)
@ -680,12 +662,11 @@ _register_template(
_register_template(
name="gemma",
format_user=StringFormatter(slots=["<start_of_turn>user\n{{content}}<end_of_turn>\n<start_of_turn>model\n"]),
format_assistant=StringFormatter(slots=["{{content}}<end_of_turn>\n"]),
format_observation=StringFormatter(
slots=["<start_of_turn>tool\n{{content}}<end_of_turn>\n<start_of_turn>model\n"]
),
format_separator=EmptyFormatter(slots=["<end_of_turn>\n"]),
format_prefix=EmptyFormatter(slots=[{"bos_token"}]),
efficient_eos=True,
)
@ -710,8 +691,8 @@ _register_template(
"<|start_of_role|>user<|end_of_role|>{{content}}<|end_of_text|>\n<|start_of_role|>assistant<|end_of_role|>"
]
),
format_assistant=StringFormatter(slots=["{{content}}<|end_of_text|>\n"]),
format_system=StringFormatter(slots=["<|start_of_role|>system<|end_of_role|>{{content}}<|end_of_text|>\n"]),
format_separator=EmptyFormatter(slots=["\n"]),
)
@ -726,22 +707,20 @@ _register_template(
_register_template(
name="intern",
format_user=StringFormatter(slots=["<|User|>:{{content}}\n<|Bot|>:"]),
format_assistant=StringFormatter(slots=["{{content}}<eoa>\n"]),
format_system=StringFormatter(slots=["<|System|>:{{content}}\n"]),
format_separator=EmptyFormatter(slots=["<eoa>\n"]),
format_prefix=EmptyFormatter(slots=[{"bos_token"}]),
stop_words=["<eoa>"],
efficient_eos=True, # internlm tokenizer cannot set eos_token_id
)
_register_template(
name="intern2",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\n"]),
format_system=StringFormatter(slots=["<|im_start|>system\n{{content}}<|im_end|>\n"]),
format_separator=EmptyFormatter(slots=["<|im_end|>\n"]),
format_prefix=EmptyFormatter(slots=[{"bos_token"}]),
stop_words=["<|im_end|>"],
efficient_eos=True, # internlm2 tokenizer cannot set eos_token_id
)
@ -872,6 +851,7 @@ _register_template(
name="llava_next_mistral",
format_user=StringFormatter(slots=["[INST] {{content}}[/INST]"]),
format_assistant=StringFormatter(slots=[" {{content}}", {"eos_token"}]),
format_system=StringFormatter(slots=["{{content}}\n\n"]),
format_function=FunctionFormatter(slots=["[TOOL_CALLS] ", "{{content}}", {"eos_token"}], tool_format="mistral"),
format_observation=StringFormatter(slots=["""[TOOL_RESULTS] {"content": {{content}}}[/TOOL_RESULTS]"""]),
format_tools=ToolFormatter(tool_format="mistral"),
@ -884,16 +864,15 @@ _register_template(
_register_template(
name="llava_next_qwen",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\n"]),
format_system=StringFormatter(slots=["<|im_start|>system\n{{content}}<|im_end|>\n"]),
format_function=FunctionFormatter(slots=["{{content}}", "<|im_end|>"], tool_format="qwen"),
format_function=FunctionFormatter(slots=["{{content}}<|im_end|>\n"], tool_format="qwen"),
format_observation=StringFormatter(
slots=["<|im_start|>user\n<tool_response>\n{{content}}\n</tool_response><|im_end|>\n<|im_start|>assistant\n"]
),
format_tools=ToolFormatter(tool_format="qwen"),
format_separator=EmptyFormatter(slots=["\n"]),
default_system="You are a helpful assistant.",
stop_words=["<|im_end|>"],
replace_eos=True,
mm_plugin=get_mm_plugin(name="llava_next", image_token="<image>"),
)
@ -902,10 +881,9 @@ _register_template(
_register_template(
name="llava_next_yi",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\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>"),
)
@ -927,6 +905,7 @@ _register_template(
name="llava_next_video_mistral",
format_user=StringFormatter(slots=["[INST] {{content}}[/INST]"]),
format_assistant=StringFormatter(slots=[" {{content}}", {"eos_token"}]),
format_system=StringFormatter(slots=["{{content}}\n\n"]),
format_function=FunctionFormatter(slots=["[TOOL_CALLS] ", "{{content}}", {"eos_token"}], tool_format="mistral"),
format_observation=StringFormatter(slots=["""[TOOL_RESULTS] {"content": {{content}}}[/TOOL_RESULTS]"""]),
format_tools=ToolFormatter(tool_format="mistral"),
@ -939,10 +918,9 @@ _register_template(
_register_template(
name="llava_next_video_yi",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\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>"),
)
@ -951,16 +929,15 @@ _register_template(
_register_template(
name="marco",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\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=(
"你是一个经过良好训练的AI助手，你的名字是Marco-o1.由阿里国际数字商业集团的AI Business创造.\n## 重要！！！！！\n"
"当你回答问题时，你的思考应该在<Thought>内完成，<Output>内输出你的结果。\n"
"<Thought>应该尽可能是英文，但是有2个特例，一个是对原文中的引用，另一个是是数学应该使用markdown格式，<Output>内的输出需要遵循用户输入的语言。\n"
),
stop_words=["<|im_end|>"],
replace_eos=True,
)
@ -968,6 +945,7 @@ _register_template(
name="mistral",
format_user=StringFormatter(slots=["[INST] {{content}}[/INST]"]),
format_assistant=StringFormatter(slots=[" {{content}}", {"eos_token"}]),
format_system=StringFormatter(slots=["{{content}}\n\n"]),
format_function=FunctionFormatter(slots=["[TOOL_CALLS] ", "{{content}}", {"eos_token"}], tool_format="mistral"),
format_observation=StringFormatter(slots=["""[TOOL_RESULTS] {"content": {{content}}}[/TOOL_RESULTS]"""]),
format_tools=ToolFormatter(tool_format="mistral"),
@ -1001,7 +979,6 @@ _register_template(
),
format_prefix=EmptyFormatter(slots=[{"bos_token"}]),
stop_words=["<|eot_id|>"],
replace_eos=True,
)
@ -1009,9 +986,9 @@ _register_template(
_register_template(
name="opencoder",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\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 OpenCoder, created by OpenCoder Team.",
stop_words=["<|im_end|>"],
)
@ -1028,12 +1005,11 @@ _register_template(
_register_template(
name="paligemma",
format_user=StringFormatter(slots=["<start_of_turn>user\n{{content}}<end_of_turn>\n<start_of_turn>model\n"]),
format_assistant=StringFormatter(slots=["{{content}}<end_of_turn>\n"]),
format_observation=StringFormatter(
slots=["<start_of_turn>tool\n{{content}}<end_of_turn>\n<start_of_turn>model\n"]
),
format_separator=EmptyFormatter(slots=["<end_of_turn>\n"]),
format_prefix=EmptyFormatter(slots=[{"bos_token"}]),
efficient_eos=True,
mm_plugin=get_mm_plugin(name="paligemma", image_token="<image>"),
)
@ -1041,28 +1017,37 @@ _register_template(
_register_template(
name="phi",
format_user=StringFormatter(slots=["<|user|>\n{{content}}<|end|>\n<|assistant|>\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|end|>\n"]),
format_system=StringFormatter(slots=["<|system|>\n{{content}}<|end|>\n"]),
format_separator=EmptyFormatter(slots=["\n"]),
format_prefix=EmptyFormatter(slots=[{"bos_token"}]),
stop_words=["<|end|>"],
replace_eos=True,
)
_register_template(
name="phi_small",
format_user=StringFormatter(slots=["<|user|>\n{{content}}<|end|>\n<|assistant|>\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|end|>\n"]),
format_system=StringFormatter(slots=["<|system|>\n{{content}}<|end|>\n"]),
format_separator=EmptyFormatter(slots=["\n"]),
format_prefix=EmptyFormatter(slots=[{"<|endoftext|>"}]),
stop_words=["<|end|>"],
replace_eos=True,
)
_register_template(
name="phi4",
format_user=StringFormatter(
slots=["<|im_start|>user<|im_sep|>{{content}}<|im_end|><|im_start|>assistant<|im_sep|>"]
),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>"]),
format_system=StringFormatter(slots=["<|im_start|>system<|im_sep|>{{content}}<|im_end|>"]),
stop_words=["<|im_end|>"],
)
_register_template(
name="pixtral",
format_user=StringFormatter(slots=["[INST]{{content}}[/INST]"]),
format_system=StringFormatter(slots=["{{content}}\n\n"]),
format_prefix=EmptyFormatter(slots=[{"bos_token"}]),
mm_plugin=get_mm_plugin(name="pixtral", image_token="[IMG]"),
)
@ -1072,13 +1057,13 @@ _register_template(
_register_template(
name="qwen",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\n"]),
format_system=StringFormatter(slots=["<|im_start|>system\n{{content}}<|im_end|>\n"]),
format_function=FunctionFormatter(slots=["{{content}}", "<|im_end|>"], tool_format="qwen"),
format_function=FunctionFormatter(slots=["{{content}}<|im_end|>\n"], tool_format="qwen"),
format_observation=StringFormatter(
slots=["<|im_start|>user\n<tool_response>\n{{content}}\n</tool_response><|im_end|>\n<|im_start|>assistant\n"]
),
format_tools=ToolFormatter(tool_format="qwen"),
format_separator=EmptyFormatter(slots=["\n"]),
default_system="You are a helpful assistant.",
stop_words=["<|im_end|>"],
)
@ -1088,13 +1073,13 @@ _register_template(
_register_template(
name="qwen2_vl",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\n"]),
format_system=StringFormatter(slots=["<|im_start|>system\n{{content}}<|im_end|>\n"]),
format_function=FunctionFormatter(slots=["{{content}}", "<|im_end|>"], tool_format="qwen"),
format_function=FunctionFormatter(slots=["{{content}}<|im_end|>\n"], tool_format="qwen"),
format_observation=StringFormatter(
slots=["<|im_start|>user\n<tool_response>\n{{content}}\n</tool_response><|im_end|>\n<|im_start|>assistant\n"]
),
format_tools=ToolFormatter(tool_format="qwen"),
format_separator=EmptyFormatter(slots=["\n"]),
default_system="You are a helpful assistant.",
stop_words=["<|im_end|>"],
mm_plugin=get_mm_plugin(name="qwen2_vl", image_token="<|image_pad|>", video_token="<|video_pad|>"),
@ -1104,8 +1089,8 @@ _register_template(
_register_template(
name="sailor",
format_user=StringFormatter(slots=["<|im_start|>question\n{{content}}<|im_end|>\n<|im_start|>answer\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\n"]),
format_system=StringFormatter(slots=["<|im_start|>system\n{{content}}<|im_end|>\n"]),
format_separator=EmptyFormatter(slots=["\n"]),
default_system=(
"You are an AI assistant named Sailor created by Sea AI Lab. "
"Your answer should be friendly, unbiased, faithful, informative and detailed."
@ -1157,10 +1142,9 @@ _register_template(
_register_template(
name="starchat",
format_user=StringFormatter(slots=["<|user|>\n{{content}}<|end|>\n<|assistant|>"]),
format_assistant=StringFormatter(slots=["{{content}}<|end|>\n"]),
format_system=StringFormatter(slots=["<|system|>\n{{content}}<|end|>\n"]),
format_separator=EmptyFormatter(slots=["\n"]),
stop_words=["<|end|>"],
replace_eos=True,
)
@ -1223,8 +1207,8 @@ _register_template(
_register_template(
name="yayi",
format_user=StringFormatter(slots=[{"token": "<|Human|>"}, ":\n{{content}}\n\n", {"token": "<|YaYi|>"}, ":"]),
format_assistant=StringFormatter(slots=["{{content}}\n\n"]),
format_system=StringFormatter(slots=[{"token": "<|System|>"}, ":\n{{content}}\n\n"]),
format_separator=EmptyFormatter(slots=["\n\n"]),
default_system=(
"You are a helpful, respectful and honest assistant named YaYi "
"developed by Beijing Wenge Technology Co.,Ltd. "
@ -1244,17 +1228,16 @@ _register_template(
_register_template(
name="yi",
format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\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,
)
_register_template(
name="yi_vl",
format_user=StringFormatter(slots=["### Human: {{content}}\n### Assistant:"]),
format_separator=EmptyFormatter(slots=["\n"]),
format_assistant=StringFormatter(slots=["{{content}}\n"]),
default_system=(
"This is a chat between an inquisitive human and an AI assistant. "
"Assume the role of the AI assistant. Read all the images carefully, "
@ -1271,9 +1254,8 @@ _register_template(
_register_template(
name="yuan",
format_user=StringFormatter(slots=["{{content}}", {"token": "<sep>"}]),
format_separator=EmptyFormatter(slots=["\n"]),
format_assistant=StringFormatter(slots=["{{content}}<eod>\n"]),
stop_words=["<eod>"],
replace_eos=True,
)
@ -1288,5 +1270,5 @@ _register_template(
_register_template(
name="ziya",
format_user=StringFormatter(slots=["<human>:{{content}}\n<bot>:"]),
format_separator=EmptyFormatter(slots=["\n"]),
format_assistant=StringFormatter(slots=["{{content}}\n"]),
)

19
src/llamafactory/extras/constants.py
View File

@ -1413,6 +1413,14 @@ register_model_group(
DownloadSource.DEFAULT: "microsoft/Phi-3-medium-128k-instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Phi-3-medium-128k-instruct",
},
"Phi-3.5-4B-instruct": {
DownloadSource.DEFAULT: "microsoft/Phi-3.5-mini-instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Phi-3.5-mini-instruct",
},
"Phi-3.5-MoE-42B-A6.6B-instruct": {
DownloadSource.DEFAULT: "microsoft/Phi-3.5-MoE-instruct",
DownloadSource.MODELSCOPE: "LLM-Research/Phi-3.5-MoE-instruct",
},
},
template="phi",
)
@ -1433,6 +1441,17 @@ register_model_group(
)
register_model_group(
models={
"Phi-4-14B-Instruct": {
DownloadSource.DEFAULT: "microsoft/phi-4",
DownloadSource.MODELSCOPE: "LLM-Research/phi-4",
},
},
template="phi4",
)
register_model_group(
models={
"Pixtral-12B-Instruct": {

85
tests/data/test_template.py
View File

@ -13,7 +13,7 @@
# limitations under the License.
import os
from typing import TYPE_CHECKING, List, Sequence
from typing import TYPE_CHECKING, Sequence
import pytest
from transformers import AutoTokenizer
@ -42,39 +42,36 @@ MESSAGES = [
def _check_tokenization(
tokenizer: "PreTrainedTokenizer", batch_input_ids: Sequence[Sequence[int]], batch_text: Sequence[str]
) -> None:
r"""
Checks token ids and texts.
encode(text) == token_ids
decode(token_ids) == text
"""
for input_ids, text in zip(batch_input_ids, batch_text):
assert input_ids == tokenizer.encode(text, add_special_tokens=False)
assert tokenizer.encode(text, add_special_tokens=False) == input_ids
assert tokenizer.decode(input_ids) == text
def _check_single_template(
model_id: str, template_name: str, prompt_str: str, answer_str: str, extra_str: str, use_fast: bool
) -> List[str]:
tokenizer = AutoTokenizer.from_pretrained(model_id, use_fast=use_fast, token=HF_TOKEN)
content_str = tokenizer.apply_chat_template(MESSAGES, tokenize=False)
content_ids = tokenizer.apply_chat_template(MESSAGES, tokenize=True)
template = get_template_and_fix_tokenizer(tokenizer, DataArguments(template=template_name))
prompt_ids, answer_ids = template.encode_oneturn(tokenizer, MESSAGES)
assert content_str == prompt_str + answer_str + extra_str
assert content_ids == prompt_ids + answer_ids + tokenizer.encode(extra_str, add_special_tokens=False)
_check_tokenization(tokenizer, (prompt_ids, answer_ids), (prompt_str, answer_str))
return content_ids
def _check_template(model_id: str, template_name: str, prompt_str: str, answer_str: str, extra_str: str = "") -> None:
"""
Checks template for both the slow tokenizer and the fast tokenizer.
def _check_template(model_id: str, template_name: str, prompt_str: str, answer_str: str, use_fast: bool) -> None:
r"""
Checks template.
Args:
model_id: the model id on hugging face hub.
template_name: the template name.
prompt_str: the string corresponding to the prompt part.
answer_str: the string corresponding to the answer part.
extra_str: the extra string in the jinja template of the original tokenizer.
use_fast: whether to use fast tokenizer.
"""
slow_ids = _check_single_template(model_id, template_name, prompt_str, answer_str, extra_str, use_fast=False)
fast_ids = _check_single_template(model_id, template_name, prompt_str, answer_str, extra_str, use_fast=True)
assert slow_ids == fast_ids
tokenizer = AutoTokenizer.from_pretrained(model_id, use_fast=use_fast, token=HF_TOKEN)
content_str = tokenizer.apply_chat_template(MESSAGES, tokenize=False)
content_ids = tokenizer.apply_chat_template(MESSAGES, tokenize=True)
template = get_template_and_fix_tokenizer(tokenizer, DataArguments(template=template_name))
prompt_ids, answer_ids = template.encode_oneturn(tokenizer, MESSAGES)
assert content_str == prompt_str + answer_str
assert content_ids == prompt_ids + answer_ids
_check_tokenization(tokenizer, (prompt_ids, answer_ids), (prompt_str, answer_str))
@pytest.mark.parametrize("use_fast", [True, False])
@ -125,19 +122,21 @@ def test_jinja_template(use_fast: bool):
@pytest.mark.skipif(not HF_TOKEN, reason="Gated model.")
def test_gemma_template():
@pytest.mark.parametrize("use_fast", [True, False])
def test_gemma_template(use_fast: bool):
prompt_str = (
"<bos><start_of_turn>user\nHow are you<end_of_turn>\n"
"<start_of_turn>model\nI am fine!<end_of_turn>\n"
"<start_of_turn>user\n你好<end_of_turn>\n"
"<start_of_turn>model\n"
)
answer_str = "很高兴认识你！"
_check_template("google/gemma-2-9b-it", "gemma", prompt_str, answer_str, extra_str="<end_of_turn>\n")
answer_str = "很高兴认识你！<end_of_turn>\n"
_check_template("google/gemma-2-9b-it", "gemma", prompt_str, answer_str, use_fast)
@pytest.mark.skipif(not HF_TOKEN, reason="Gated model.")
def test_llama3_template():
@pytest.mark.parametrize("use_fast", [True, False])
def test_llama3_template(use_fast: bool):
prompt_str = (
"<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\nHow are you<|eot_id|>"
"<|start_header_id|>assistant<|end_header_id|>\n\nI am fine!<|eot_id|>"
@ -145,10 +144,25 @@ def test_llama3_template():
"<|start_header_id|>assistant<|end_header_id|>\n\n"
)
answer_str = "很高兴认识你！<|eot_id|>"
_check_template("meta-llama/Meta-Llama-3-8B-Instruct", "llama3", prompt_str, answer_str)
_check_template("meta-llama/Meta-Llama-3-8B-Instruct", "llama3", prompt_str, answer_str, use_fast)
def test_qwen_template():
@pytest.mark.parametrize(
"use_fast", [True, pytest.param(False, marks=pytest.mark.xfail(reason="Phi-4 slow tokenizer is broken."))]
)
def test_phi4_template(use_fast: bool):
prompt_str = (
"<|im_start|>user<|im_sep|>How are you<|im_end|>"
"<|im_start|>assistant<|im_sep|>I am fine!<|im_end|>"
"<|im_start|>user<|im_sep|>你好<|im_end|>"
"<|im_start|>assistant<|im_sep|>"
)
answer_str = "很高兴认识你！<|im_end|>"
_check_template("microsoft/phi-4", "phi4", prompt_str, answer_str, use_fast)
@pytest.mark.parametrize("use_fast", [True, False])
def test_qwen_template(use_fast: bool):
prompt_str = (
"<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n"
"<|im_start|>user\nHow are you<|im_end|>\n"
@ -156,17 +170,18 @@ def test_qwen_template():
"<|im_start|>user\n你好<|im_end|>\n"
"<|im_start|>assistant\n"
)
answer_str = "很高兴认识你！<|im_end|>"
_check_template("Qwen/Qwen2-7B-Instruct", "qwen", prompt_str, answer_str, extra_str="\n")
answer_str = "很高兴认识你！<|im_end|>\n"
_check_template("Qwen/Qwen2-7B-Instruct", "qwen", prompt_str, answer_str, use_fast)
@pytest.mark.xfail(reason="The fast tokenizer of Yi model is corrupted.")
def test_yi_template():
@pytest.mark.parametrize("use_fast", [True, False])
@pytest.mark.xfail(reason="Yi tokenizer is broken.")
def test_yi_template(use_fast: bool):
prompt_str = (
"<|im_start|>user\nHow are you<|im_end|>\n"
"<|im_start|>assistant\nI am fine!<|im_end|>\n"
"<|im_start|>user\n你好<|im_end|>\n"
"<|im_start|>assistant\n"
)
answer_str = "很高兴认识你！<|im_end|>"
_check_template("01-ai/Yi-1.5-6B-Chat", "yi", prompt_str, answer_str)
answer_str = "很高兴认识你！<|im_end|>\n"
_check_template("01-ai/Yi-1.5-6B-Chat", "yi", prompt_str, answer_str, use_fast)