EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Unify the model type checking (#1905)

Browse Source
This commit is contained in:
Lianmin Zheng
2024-11-03 12:25:39 -08:00
committed by GitHub
parent c17c578108
commit 0abbf289a8
13 changed files with 146 additions and 160 deletions
Download Patch File Download Diff File Expand all files Collapse all files

87
python/sglang/lang/chat_template.py
View File

@@ -116,6 +116,23 @@ register_chat_template(
)
)
register_chat_template(
ChatTemplate(
name="chatml-llava",
default_system_prompt="You are a helpful assistant.",
role_prefix_and_suffix={
"system": ("<|im_start|>system\n", "<|im_end|>\n"),
"user": ("<|im_start|>user\n", "<|im_end|>\n"),
"assistant": ("<|im_start|>assistant\n", "<|im_end|>\n"),
},
style=ChatTemplateStyle.PLAIN,
stop_str=("<|im_end|>",),
image_token="<image>\n",
)
)
# There is default system prompt for qwen
# reference: https://modelscope.cn/models/qwen/Qwen2-72B-Instruct/file/view/master?fileName=tokenizer_config.json&status=1
# The chat template is: "{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}"
@@ -149,22 +166,6 @@ register_chat_template(
)
)
register_chat_template(
ChatTemplate(
name="chatml-llava",
default_system_prompt="You are a helpful assistant.",
role_prefix_and_suffix={
"system": ("<|im_start|>system\n", "<|im_end|>\n"),
"user": ("<|im_start|>user\n", "<|im_end|>\n"),
"assistant": ("<|im_start|>assistant\n", "<|im_end|>\n"),
},
style=ChatTemplateStyle.PLAIN,
stop_str=("<|im_end|>",),
image_token="<image>\n",
)
)
# Reference: https://github.com/lm-sys/FastChat/blob/main/docs/vicuna_weights_version.md#prompt-template
register_chat_template(
ChatTemplate(
@@ -182,21 +183,6 @@ register_chat_template(
)
)
# Reference: https://modelscope.cn/models/01ai/Yi-1.5-34B-Chat/file/view/master?fileName=tokenizer_config.json&status=1
register_chat_template(
ChatTemplate(
name="yi-1.5",
default_system_prompt=None,
role_prefix_and_suffix={
"system": ("", ""),
"user": ("<|im_start|>user\n", "<|im_end|>\n<|im_start|>assistant\n"),
"assistant": ("", "<|im_end|>\n"),
},
style=ChatTemplateStyle.PLAIN,
stop_str=("<|im_end|>",),
)
)
register_chat_template(
ChatTemplate(
name="llama-2-chat",
@@ -233,6 +219,45 @@ register_chat_template(
)
)
# The difference between "llama-3-instruct-llava" and "llama-3-instruct" is that llava uses a different image_token.
register_chat_template(
ChatTemplate(
name="llama-3-instruct-llava",
default_system_prompt=None,
role_prefix_and_suffix={
"system": (
"<|start_header_id|>system<|end_header_id|>\n\n",
"<|eot_id|>",
),
"user": (
"<|start_header_id|>user<|end_header_id|>\n\n",
"<|eot_id|>",
),
"assistant": (
"<|start_header_id|>assistant<|end_header_id|>\n\n",
"<|eot_id|>",
),
},
stop_str=("<|eot_id|>",),
image_token="<image>\n",
)
)
# Reference: https://modelscope.cn/models/01ai/Yi-1.5-34B-Chat/file/view/master?fileName=tokenizer_config.json&status=1
register_chat_template(
ChatTemplate(
name="yi-1.5",
default_system_prompt=None,
role_prefix_and_suffix={
"system": ("", ""),
"user": ("<|im_start|>user\n", "<|im_end|>\n<|im_start|>assistant\n"),
"assistant": ("", "<|im_end|>\n"),
},
style=ChatTemplateStyle.PLAIN,
stop_str=("<|im_end|>",),
)
)
# Reference: https://github.com/01-ai/Yi/tree/main/VL#major-difference-with-llava
register_chat_template(
ChatTemplate(