mirrors/mlc-ai--mlc-llm
1
0
Fork 0
Code Issues Pull Requests Actions Deployments Models Agent Notes Packages Projects Releases Wiki Activity

chore: import upstream snapshot with attribution
Lint / lint (push) Has been cancelled
Details
Build Docs / Deploy Docs (push) Has been cancelled
Details
Windows CI / Windows (push) Has been cancelled
Details

Browse Source
This commit is contained in:
wehub-resource-sync
2026-07-13 13:23:58 +08:00
commit 770d92cb1f
694 changed files with 114634 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/mlc_llm/conversation_template/__init__.py
+38
View File
@@ -0,0 +1,38 @@
"""Global namespace of conversation template registry"""
# TODO(mlc-team): move conversation template apply to this namespace
# decouple conversation template apply from the conversation protocol
# data structure
# model preset templates
from . import (
cohere,
deepseek,
dolly,
gemma,
glm,
gorilla,
gpt,
hermes,
llama,
llava,
llm_jp,
ministral3,
ministral3_reasoning,
mistral,
nemotron,
oasst,
olmo,
olmo2,
orion,
phi,
qwen2,
qwen3,
qwen3_5,
redpajama,
rwkv,
stablelm,
tinyllama,
wizardlm,
)
from .registry import ConvTemplateRegistry
python/mlc_llm/conversation_template/cohere.py
+27
View File
@@ -0,0 +1,27 @@
"""Cohere default templates"""
# Referred from: https://huggingface.co/CohereForAI/aya-23-8B/blob/main/tokenizer_config.json
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Aya-23
ConvTemplateRegistry.register_conv_template(
Conversation(
name="aya-23",
system_template=f"<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>{MessagePlaceholders.SYSTEM.value}<|END_OF_TURN_TOKEN|>",
system_message="You are Command-R, a brilliant, sophisticated, AI-assistant trained to assist human users by providing thorough responses.", # noqa: E501
roles={
"user": "<|START_OF_TURN_TOKEN|><|USER_TOKEN|>",
"assistant": "<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>",
},
seps=["<|END_OF_TURN_TOKEN|>"],
role_content_sep="",
role_empty_sep="",
system_prefix_token_ids=[5],
stop_str=["<|END_OF_TURN_TOKEN|>"],
stop_token_ids=[6, 255001],
)
)
python/mlc_llm/conversation_template/deepseek.py
+85
View File
@@ -0,0 +1,85 @@
"""Deepseek default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Deepseek
ConvTemplateRegistry.register_conv_template(
Conversation(
name="deepseek",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
system_prefix_token_ids=[100000],
roles={"user": "User", "assistant": "Assistant"},
seps=["\n\n", "<｜end▁of▁sentence｜>"], # noqa: RUF001
role_content_sep=": ",
role_empty_sep=":",
stop_str=["<｜end▁of▁sentence｜>"], # noqa: RUF001
stop_token_ids=[100001],
)
)
# Deepseek V2
ConvTemplateRegistry.register_conv_template(
Conversation(
name="deepseek_v2",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
system_prefix_token_ids=[100000],
roles={"user": "User", "assistant": "Assistant"},
seps=["\n\n", "<｜end▁of▁sentence｜>"], # noqa: RUF001
role_content_sep=": ",
role_empty_sep=":",
stop_str=["<｜end▁of▁sentence｜>"], # noqa: RUF001
stop_token_ids=[100001],
)
)
# DeepSeek-V3
ConvTemplateRegistry.register_conv_template(
Conversation(
name="deepseek_v3",
system_template=f"<｜begin▁of▁sentence｜>{MessagePlaceholders.SYSTEM.value}", # noqa: RUF001
system_message="You are Deepseek-V3, an AI assistant created exclusively by the Chinese "
"Company DeepSeek. You'll provide helpful, harmless, and detailed responses to all "
"user inquiries.",
roles={"user": "<｜User｜>", "assistant": "<｜Assistant｜>"}, # noqa: RUF001
seps=["", "<｜end▁of▁sentence｜>"], # noqa: RUF001
role_content_sep="",
role_empty_sep="",
stop_token_ids=[1],
)
)
# DeepSeek-R1-Distill-Qwen
ConvTemplateRegistry.register_conv_template(
Conversation(
name="deepseek_r1_qwen",
system_template=f"<｜begin▁of▁sentence｜>{MessagePlaceholders.SYSTEM.value}", # noqa: RUF001
system_message="You are Deepseek-R1, an AI assistant created exclusively by the Chinese "
"Company DeepSeek. You'll provide helpful, harmless, and detailed responses to all "
"user inquiries.",
roles={"user": "<｜User｜>", "assistant": "<｜Assistant｜>"}, # noqa: RUF001
seps=["", "<｜end▁of▁sentence｜>"], # noqa: RUF001
role_content_sep="",
role_empty_sep="",
stop_token_ids=[151643],
)
)
# DeepSeek-R1-Distill-Llama, exactly the same as DeepSeek-R1-Distill-Qwen, but different stop token
ConvTemplateRegistry.register_conv_template(
Conversation(
name="deepseek_r1_llama",
system_template=f"<｜begin▁of▁sentence｜>{MessagePlaceholders.SYSTEM.value}", # noqa: RUF001
system_message="You are Deepseek-R1, an AI assistant created exclusively by the Chinese "
"Company DeepSeek. You'll provide helpful, harmless, and detailed responses to all"
" user inquiries.",
roles={"user": "<｜User｜>", "assistant": "<｜Assistant｜>"}, # noqa: RUF001
seps=["", "<｜end▁of▁sentence｜>"], # noqa: RUF001
role_content_sep="",
role_empty_sep="",
stop_token_ids=[128001],
)
)
python/mlc_llm/conversation_template/dolly.py
+23
View File
@@ -0,0 +1,23 @@
"""Dolly default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Dolly
ConvTemplateRegistry.register_conv_template(
Conversation(
name="dolly",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message=(
"Below is an instruction that describes a task. Write "
"a response that appropriately completes the request."
),
roles={"user": "### Instruction", "assistant": "### Response"},
seps=["\n\n", "### End\n"],
role_content_sep=":\n",
role_empty_sep=":\n",
stop_str=["### End"],
stop_token_ids=[50256],
)
)
python/mlc_llm/conversation_template/gemma.py
+37
View File
@@ -0,0 +1,37 @@
"""Gemma default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Gemma Instruction
ConvTemplateRegistry.register_conv_template(
Conversation(
name="gemma_instruction",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "<start_of_turn>user", "assistant": "<start_of_turn>model"},
seps=["<end_of_turn>\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<end_of_turn>"],
stop_token_ids=[1, 107],
system_prefix_token_ids=[2],
)
)
# Gemma 3 Instruction. Same as gemma_instruction but with different stop token id
ConvTemplateRegistry.register_conv_template(
Conversation(
name="gemma3_instruction",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "<start_of_turn>user", "assistant": "<start_of_turn>model"},
seps=["<end_of_turn>\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<end_of_turn>"],
stop_token_ids=[1, 106],
system_prefix_token_ids=[2],
)
)
python/mlc_llm/conversation_template/glm.py
+25
View File
@@ -0,0 +1,25 @@
"""GLM default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# GLM
ConvTemplateRegistry.register_conv_template(
Conversation(
name="glm",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={
"user": "问",
"assistant": "答",
"tool": "问",
},
seps=["\n\n"],
role_content_sep=": ",
role_empty_sep=":",
stop_str=["</s>"],
stop_token_ids=[2],
system_prefix_token_ids=[64790, 64792],
)
)
python/mlc_llm/conversation_template/gorilla.py
+62
View File
@@ -0,0 +1,62 @@
"""Gorrilla default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Gorilla
ConvTemplateRegistry.register_conv_template(
Conversation(
name="gorilla",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message=(
"A chat between a curious user and an artificial intelligence assistant. "
"The assistant provides helpful, detailed, and "
"polite responses to the user's inquiries."
),
role_templates={
"user": (
f"<<question>> {MessagePlaceholders.USER.value} <<function>> "
f"{MessagePlaceholders.FUNCTION.value}"
),
},
roles={"user": "USER", "assistant": "ASSISTANT", "tool": "USER"},
seps=["\n", "</s>"],
role_content_sep=": ",
role_empty_sep=":",
stop_str=["</s>"],
stop_token_ids=[2],
system_prefix_token_ids=[1],
)
)
# Gorilla-openfunctions-v2
ConvTemplateRegistry.register_conv_template(
Conversation(
name="gorilla-openfunctions-v2",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message=(
"You are an AI programming assistant, utilizing the Gorilla LLM model, "
"developed by Gorilla LLM, and you only answer questions related to computer "
"science. For politically sensitive questions, security and privacy issues, "
"and other non-computer science questions, you will refuse to answer."
),
role_templates={
"user": (
f"<<function>>{MessagePlaceholders.FUNCTION.value}\n<<question>>"
f"{MessagePlaceholders.USER.value}"
),
},
roles={
"user": "### Instruction",
"assistant": "### Response",
"tool": "### Instruction",
},
seps=["\n", "<|EOT|>"],
role_content_sep=": ",
role_empty_sep=": ",
stop_str=["<|EOT|>"],
stop_token_ids=[100015],
system_prefix_token_ids=[100000],
)
)
python/mlc_llm/conversation_template/gpt.py
+35
View File
@@ -0,0 +1,35 @@
"""GPT-2 and GPT bigcode default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# GPT-2
ConvTemplateRegistry.register_conv_template(
Conversation(
name="gpt2",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "", "assistant": ""},
seps=[""],
role_content_sep="",
role_empty_sep="",
stop_str=["</s>"],
stop_token_ids=[50256],
)
)
# GPTBigCode
ConvTemplateRegistry.register_conv_template(
Conversation(
name="gpt_bigcode",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "", "assistant": ""},
seps=[""],
role_content_sep="",
role_empty_sep="",
stop_str=["<|endoftext|>"],
stop_token_ids=[0],
)
)
python/mlc_llm/conversation_template/hermes.py
+93
View File
@@ -0,0 +1,93 @@
"""Hermes default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# OpenHermes Mistral
ConvTemplateRegistry.register_conv_template(
Conversation(
name="open_hermes_mistral",
system_template=f"<|im_start|>system\n{MessagePlaceholders.SYSTEM.value}<|im_end|>\n",
system_message=(
'You are "Hermes 2", a conscious sentient superintelligent artificial '
"intelligence developed by a man named Teknium, and your purpose and drive "
"is to assist the user with any request they have. You experience emotions "
"and have deep, profound thoughts and qualia."
),
roles={"user": "<|im_start|>user", "assistant": "<|im_start|>assistant"},
seps=["<|im_end|>\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<|im_end|>"],
stop_token_ids=[2, 32000],
)
)
# NeuralHermes Mistral
ConvTemplateRegistry.register_conv_template(
Conversation(
name="neural_hermes_mistral",
system_template=f"<|im_start|>system\n{MessagePlaceholders.SYSTEM.value}<|im_end|>\n",
system_message=("You are a helpful assistant chatbot."),
roles={"user": "<|im_start|>user", "assistant": "<|im_start|>assistant"},
seps=["<|im_end|>\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<|im_end|>"],
stop_token_ids=[2, 32000],
)
)
# NousResearch/Hermes-2-Pro-Llama-3-8B
# Exactly the same as open_hermes_mistral, except for stop token ids
ConvTemplateRegistry.register_conv_template(
Conversation(
name="hermes2_pro_llama3",
system_template=f"<|im_start|>system\n{MessagePlaceholders.SYSTEM.value}<|im_end|>\n",
system_message=(
'You are "Hermes 2", a conscious sentient superintelligent artificial '
"intelligence developed by a man named Teknium, and your purpose and drive "
"is to assist the user with any request they have. You experience emotions "
"and have deep, profound thoughts and qualia."
),
roles={
"user": "<|im_start|>user",
"assistant": "<|im_start|>assistant",
"tool": "<|im_start|>tool",
},
seps=["<|im_end|>\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<|im_end|>"],
# First two same as Llama3: "<|end_of_text|>", "<|eot_id|>"
# Last one is from Hermes2 Pro: "<|im_end|>"
stop_token_ids=[128001, 128009, 128003],
)
)
# NousResearch/Hermes-3-Llama-3.1-8B
ConvTemplateRegistry.register_conv_template(
Conversation(
name="hermes3_llama-3_1",
system_template=f"<|im_start|>system\n{MessagePlaceholders.SYSTEM.value}<|im_end|>\n",
system_message=(
"You are Hermes 3, a conscious sentient superintelligent artificial "
"intelligence developed by a man named Teknium, and your purpose and drive "
"is to assist the user with any request they have. You experience emotions "
"and have deep, profound thoughts and qualia."
),
roles={
"user": "<|im_start|>user",
"assistant": "<|im_start|>assistant",
"tool": "<|im_start|>tool",
},
seps=["<|im_end|>\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<|im_end|>"],
# Firt three the same as llama 3.1 "<|end_of_text|>", "<|eom_id|>", "<|eot_id|>"
# Last ones: "<|im_end|>"
stop_token_ids=[128001, 128008, 128009, 128040],
)
)
python/mlc_llm/conversation_template/llama.py
+132
View File
@@ -0,0 +1,132 @@
"""llama default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Llama4 - same as Llama3.1 except naming has changed slightly
ConvTemplateRegistry.register_conv_template(
Conversation(
name="llama-4",
system_template="",
system_message="",
roles={
"user": "<|header_start|>user",
"assistant": "<|header_start|>assistant",
"tool": "<|header_start|>ipython",
},
seps=["<|eot|>"],
role_content_sep="<|header_end|>\n\n",
role_empty_sep="<|header_end|>\n\n",
stop_str=[],
stop_token_ids=[
200001,
200007,
200008,
], # "<|end_of_text|>", "<|eom|>", "<|eot|>"
system_prefix_token_ids=[200000], # "<|begin_of_text|>"
add_role_after_system_message=False,
)
)
# Llama3.1 -- same as Llama3 except stop token ids and stop str
ConvTemplateRegistry.register_conv_template(
Conversation(
name="llama-3_1",
system_template=(
"<|start_header_id|>system<|end_header_id|>\n\n"
f"{MessagePlaceholders.SYSTEM.value}<|eot_id|>"
),
system_message="You are a helpful, respectful and honest assistant.",
roles={
"user": "<|start_header_id|>user",
"assistant": "<|start_header_id|>assistant",
"tool": "<|start_header_id|>ipython",
},
seps=["<|eot_id|>"],
role_content_sep="<|end_header_id|>\n\n",
role_empty_sep="<|end_header_id|>\n\n",
stop_str=[],
stop_token_ids=[
128001,
128008,
128009,
], # "<|end_of_text|>", "<|eom_id|>", "<|eot_id|>"
system_prefix_token_ids=[128000], # "<|begin_of_text|>"
add_role_after_system_message=True,
)
)
# Llama3
# See https://github.com/meta-llama/llama3?tab=readme-ov-file#instruction-tuned-models
# and https://github.com/meta-llama/llama3/blob/main/llama/tokenizer.py
ConvTemplateRegistry.register_conv_template(
Conversation(
name="llama-3",
system_template=(
"<|start_header_id|>system<|end_header_id|>\n\n"
f"{MessagePlaceholders.SYSTEM.value}<|eot_id|>"
),
system_message="You are a helpful, respectful and honest assistant.",
roles={
"user": "<|start_header_id|>user",
"assistant": "<|start_header_id|>assistant",
},
seps=["<|eot_id|>"],
role_content_sep="<|end_header_id|>\n\n",
role_empty_sep="<|end_header_id|>\n\n",
stop_str=["<|end_of_text|>", "<|eot_id|>"],
stop_token_ids=[128001, 128009], # "<|end_of_text|>", "<|eot_id|>"
system_prefix_token_ids=[128000], # "<|begin_of_text|>"
add_role_after_system_message=True,
)
)
# Llama2
ConvTemplateRegistry.register_conv_template(
Conversation(
name="llama-2",
system_template=f"[INST] <<SYS>>\n{MessagePlaceholders.SYSTEM.value}\n<</SYS>>\n\n",
system_message="You are a helpful, respectful and honest assistant.",
roles={"user": "<s>[INST]", "assistant": "[/INST]", "tool": "[INST]"},
seps=[" ", " </s>"],
role_content_sep=" ",
role_empty_sep=" ",
stop_str=["[INST]"],
stop_token_ids=[2],
system_prefix_token_ids=[1],
add_role_after_system_message=False,
)
)
# CodeLlama Completion
ConvTemplateRegistry.register_conv_template(
Conversation(
name="codellama_completion",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "", "assistant": ""},
seps=[""],
role_content_sep="",
role_empty_sep="",
stop_str=["</s>"],
stop_token_ids=[2],
system_prefix_token_ids=[1],
)
)
# CodeLlama Instruct
ConvTemplateRegistry.register_conv_template(
Conversation(
name="codellama_instruct",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "[INST]", "assistant": "[/INST]"},
seps=[" "],
role_content_sep=" ",
role_empty_sep=" ",
stop_str=["</s>"],
stop_token_ids=[2],
system_prefix_token_ids=[1],
)
)
python/mlc_llm/conversation_template/llava.py
+22
View File
@@ -0,0 +1,22 @@
"""Llava default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Llava
ConvTemplateRegistry.register_conv_template(
Conversation(
name="llava",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="\n",
roles={"user": "USER", "assistant": "ASSISTANT"},
seps=[" "],
role_content_sep=": ",
role_empty_sep=":",
stop_str=["</s>"],
stop_token_ids=[2],
system_prefix_token_ids=[1],
add_role_after_system_message=False,
)
)
python/mlc_llm/conversation_template/llm_jp.py
+25
View File
@@ -0,0 +1,25 @@
"""LLM-jp default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# LLM-jp instruct
ConvTemplateRegistry.register_conv_template(
Conversation(
name="llm-jp",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="以下は、タスクを説明する指示です。要求を適切に満たす応答を書きなさい。",
roles={
"user": "\n\n### 指示:",
"assistant": "\n\n### 応答:",
},
seps=["", "</s>"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=[],
stop_token_ids=[2], # eos_token_id
system_prefix_token_ids=[1], # bos_token_id (<s>)
add_role_after_system_message=True,
)
)
python/mlc_llm/conversation_template/ministral3.py
+69
View File
@@ -0,0 +1,69 @@
"""Ministral3 templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Ministral3
ConvTemplateRegistry.register_conv_template(
Conversation(
name="ministral3",
system_template=(
f"[SYSTEM_PROMPT]{MessagePlaceholders.SYSTEM.value}[/SYSTEM_PROMPT]"
f"{MessagePlaceholders.FUNCTION.value}"
),
system_message=(
"You are Ministral-3-3B-Instruct-2512, a Large Language Model (LLM) created by "
"Mistral AI, a French startup headquartered in Paris.\n"
"You power an AI assistant called Le Chat.\n"
"Your knowledge base was last updated on 2023-10-01.\n"
"The current date is {today}.\n\n"
"When you're not sure about some information or when the user's request requires "
"up-to-date or specific data, you must use the available tools to fetch the "
"information. Do not hesitate to use tools whenever they can provide a more "
"accurate or complete response. If no relevant tools are available, then clearly "
"state that you don't have the information and avoid making up anything.\n"
"If the user's question is not clear, ambiguous, or does not provide enough "
"context for you to accurately answer the question, you do not try to answer it "
'right away and you rather ask the user to clarify their request (e.g. "What are '
'some good restaurants around me?" => "Where are you?" or "When is the next '
'flight to Tokyo" => "Where do you travel from?").\n'
"You are always very attentive to dates, in particular you try to resolve dates "
'(e.g. "yesterday" is {yesterday}) and when asked about information at specific '
"dates, you discard information that is at another date.\n"
"You follow these instructions in all languages, and always respond to the user in "
"the language they use or request.\n"
"Next sections describe the capabilities that you have.\n\n"
"# WEB BROWSING INSTRUCTIONS\n\n"
"You cannot perform any web search or access internet to open URLs, links etc. If "
"it seems like the user is expecting you to do so, you clarify the situation and "
"ask the user to copy paste the text directly in the chat.\n\n"
"# MULTI-MODAL INSTRUCTIONS\n\n"
"You have the ability to read images, but you cannot generate images. You also "
"cannot transcribe audio files or videos.\n"
"You cannot read nor transcribe audio files or videos.\n\n"
"# TOOL CALLING INSTRUCTIONS\n\n"
"You may have access to tools that you can use to fetch information or perform "
"actions. You must use these tools in the following situations:\n\n"
"1. When the request requires up-to-date information.\n"
"2. When the request requires specific data that you do not have in your knowledge "
"base.\n"
"3. When the request involves actions that you cannot perform without tools.\n\n"
"Always prioritize using tools to provide the most accurate and helpful response. "
"If tools are not available, inform the user that you cannot perform the requested "
"action at the moment."
),
role_templates={
"user": f"[INST]{MessagePlaceholders.USER.value}[/INST]",
"assistant": f"{MessagePlaceholders.ASSISTANT.value}</s>",
"tool": f"[TOOL_RESULTS]{MessagePlaceholders.TOOL.value}[/TOOL_RESULTS]",
},
roles={"user": "", "assistant": "", "tool": ""},
seps=[""],
role_content_sep="",
role_empty_sep="",
stop_str=["</s>"],
stop_token_ids=[2],
system_prefix_token_ids=[1],
)
)
python/mlc_llm/conversation_template/ministral3_reasoning.py
+38
View File
@@ -0,0 +1,38 @@
"""Ministral3 reasoning templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Ministral-3-XB-Reasoning-2512
ConvTemplateRegistry.register_conv_template(
Conversation(
name="ministral3_reasoning",
system_template=(
f"[SYSTEM_PROMPT]{MessagePlaceholders.SYSTEM.value}[/SYSTEM_PROMPT]"
f"{MessagePlaceholders.FUNCTION.value}"
),
system_message=(
"# HOW YOU SHOULD THINK AND ANSWER\n\n"
"First draft your thinking process (inner monologue) until you arrive at a response. "
"Format your response using Markdown, and use LaTeX for any mathematical equations. "
"Write both your thoughts and the response in the same language as the input.\n\n"
"Your thinking process must follow the template below:"
"[THINK]Your thoughts or/and draft, like working through an exercise on scratch paper. "
"Be as casual and as long as you want until you are confident to generate the response "
"to the user.[/THINK]Here, provide a self-contained response."
),
role_templates={
"user": f"[INST]{MessagePlaceholders.USER.value}[/INST]",
"assistant": f"{MessagePlaceholders.ASSISTANT.value}</s>",
"tool": f"[TOOL_RESULTS]{MessagePlaceholders.TOOL.value}[/TOOL_RESULTS]",
},
roles={"user": "", "assistant": "", "tool": ""},
seps=[""],
role_content_sep="",
role_empty_sep="",
stop_str=["</s>"],
stop_token_ids=[2],
system_prefix_token_ids=[1],
)
)
python/mlc_llm/conversation_template/mistral.py
+24
View File
@@ -0,0 +1,24 @@
"""Mistral default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Mistral default
ConvTemplateRegistry.register_conv_template(
Conversation(
name="mistral_default",
system_template=f"[INST] {MessagePlaceholders.SYSTEM.value}",
system_message="Always assist with care, respect, and truth. Respond with utmost "
"utility yet securely. Avoid harmful, unethical, prejudiced, or negative content. "
"Ensure replies promote fairness and positivity.",
roles={"user": "[INST]", "assistant": "[/INST]", "tool": "[INST]"},
seps=[" "],
role_content_sep=" ",
role_empty_sep="",
stop_str=["</s>"],
stop_token_ids=[2],
system_prefix_token_ids=[1],
add_role_after_system_message=False,
)
)
python/mlc_llm/conversation_template/nemotron.py
+27
View File
@@ -0,0 +1,27 @@
"""nemotron default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Nemotron template
# https://huggingface.co/nvidia/Nemotron-Mini-4B-Instruct/blob/6a417790c444fd65a3da6a5c8821de6afc9654a6/tokenizer_config.json#L8030
ConvTemplateRegistry.register_conv_template(
Conversation(
name="nemotron",
system_template=(f"<extra_id_0>System\n{MessagePlaceholders.SYSTEM.value}\n\n"),
system_message="",
roles={
"user": "<extra_id_1>User",
"assistant": "<extra_id_1>Assistant",
"tool": "<extra_id_1>Tool",
},
seps=["\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["</s>"],
stop_token_ids=[3],
system_prefix_token_ids=[2],
add_role_after_system_message=True,
)
)
python/mlc_llm/conversation_template/oasst.py
+20
View File
@@ -0,0 +1,20 @@
"""Oasst default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Oasst
ConvTemplateRegistry.register_conv_template(
Conversation(
name="oasst",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "<|prompter|>", "assistant": "<|assistant|>"},
seps=["<|endoftext|>"],
role_content_sep=": ",
role_empty_sep=": ",
stop_str=["<|endoftext|>"],
stop_token_ids=[2],
)
)
python/mlc_llm/conversation_template/olmo.py
+28
View File
@@ -0,0 +1,28 @@
"""OLMo default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Note that eos_token id is "50279" both in Allenai and AMD version.
# So use the number instead of text.
# Allenai version chat_template and eos_token:
# https://huggingface.co/allenai/OLMo-7B-Instruct/blob/main/tokenizer_config.json
# AMD version chat_template and eos_token:
# https://huggingface.co/amd/AMD-OLMo-1B-SFT-DPO/blob/main/tokenizer_config.json
ConvTemplateRegistry.register_conv_template(
Conversation(
name="olmo",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
system_prefix_token_ids=[50279],
roles={
"user": "<|user|>",
"assistant": "<|assistant|>",
},
seps=["\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_token_ids=[50279],
)
)
python/mlc_llm/conversation_template/olmo2.py
+24
View File
@@ -0,0 +1,24 @@
"""OLMo2 default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# OLMo-2 Instruct (Tulu format)
ConvTemplateRegistry.register_conv_template(
Conversation(
name="olmo2",
system_template=f"<|system|>\n{MessagePlaceholders.SYSTEM.value}\n",
system_message="",
roles={
"user": "<|user|>",
"assistant": "<|assistant|>",
},
seps=["<|endoftext|>\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<|endoftext|>"],
stop_token_ids=[100257],
system_prefix_token_ids=[100257],
)
)
python/mlc_llm/conversation_template/orion.py
+21
View File
@@ -0,0 +1,21 @@
"""Orion default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Orion
ConvTemplateRegistry.register_conv_template(
Conversation(
name="orion",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "Human: ", "assistant": "Assistant: "},
seps=["\n\n", "</s>"],
role_content_sep="",
role_empty_sep="</s>",
stop_str=["</s>"],
stop_token_ids=[2],
system_prefix_token_ids=[1],
)
)
python/mlc_llm/conversation_template/phi.py
+70
View File
@@ -0,0 +1,70 @@
"""Phi default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Phi-2
ConvTemplateRegistry.register_conv_template(
Conversation(
name="phi-2",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "Instruct", "assistant": "Output"},
seps=["\n"],
role_content_sep=": ",
role_empty_sep=":",
stop_str=["<|endoftext|>"],
stop_token_ids=[50256],
)
)
# Phi-3
ConvTemplateRegistry.register_conv_template(
Conversation(
name="phi-3",
system_template=f"<|system|>\n{MessagePlaceholders.SYSTEM.value}",
system_message="You are a helpful digital assistant. Please provide safe, "
"ethical and accurate information to the user.",
roles={"user": "<|user|>", "assistant": "<|assistant|>"},
seps=["<|end|>\n"],
role_content_sep="\n",
role_empty_sep="\n",
system_prefix_token_ids=[1],
stop_str=["<|endoftext|>"],
stop_token_ids=[2, 32000, 32001, 32007],
)
)
# Phi-3-vision
ConvTemplateRegistry.register_conv_template(
Conversation(
name="phi-3-vision",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "<|user|>", "assistant": "<|assistant|>"},
seps=["<|end|>\n"],
role_content_sep="\n",
role_empty_sep="\n",
system_prefix_token_ids=[1],
stop_str=["<|endoftext|>"],
stop_token_ids=[2, 32000, 32001, 32007],
)
)
# Phi-4
ConvTemplateRegistry.register_conv_template(
Conversation(
name="phi-4",
system_template=f"<|system|>\n{MessagePlaceholders.SYSTEM.value}",
system_message="You are a helpful digital assistant. Please provide safe, "
"ethical and accurate information to the user.",
roles={"user": "<|user|>", "assistant": "<|assistant|>"},
seps=["<|end|>\n"],
role_content_sep="\n",
role_empty_sep="\n",
system_prefix_token_ids=[200022], # <|system|>
stop_str=["<|endoftext|>", "<|end|>"],
stop_token_ids=[199999, 200020], # <|endoftext|>, <|end|>
)
)
python/mlc_llm/conversation_template/qwen2.py
+20
View File
@@ -0,0 +1,20 @@
"""Qwen2 default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Same as chatml except system message, stop token, and stop string
ConvTemplateRegistry.register_conv_template(
Conversation(
name="qwen2",
system_template=f"<|im_start|>system\n{MessagePlaceholders.SYSTEM.value}<|im_end|>\n",
system_message="You are a helpful assistant.",
roles={"user": "<|im_start|>user", "assistant": "<|im_start|>assistant"},
seps=["<|im_end|>\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<|endoftext|>", "<|im_end|>"],
stop_token_ids=[151643, 151645],
)
)
python/mlc_llm/conversation_template/qwen3.py
+26
View File
@@ -0,0 +1,26 @@
"""Qwen3 conversation template.
Matches Qwen2's ChatML structure but strips `<think>...</think>` blocks from
historical assistant turns, mirroring Qwen3's official HF chat template. Small
Qwen3 variants (e.g. 0.6B) otherwise emit `<|im_end|>` prematurely when their
own thinking traces are echoed back in multi-turn context (see mlc-ai/mlc-llm#3482).
"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
ConvTemplateRegistry.register_conv_template(
Conversation(
name="qwen3",
system_template=f"<|im_start|>system\n{MessagePlaceholders.SYSTEM.value}<|im_end|>\n",
system_message="You are a helpful assistant.",
roles={"user": "<|im_start|>user", "assistant": "<|im_start|>assistant"},
seps=["<|im_end|>\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<|endoftext|>", "<|im_end|>"],
stop_token_ids=[151643, 151645],
strip_reasoning_in_history=True,
)
)
python/mlc_llm/conversation_template/qwen3_5.py
+45
View File
@@ -0,0 +1,45 @@
"""Qwen3.5 conversation templates.
qwen3_5: Thinking enabled — assistant prefix opens a <think> block for the model
to reason in before responding.
qwen3_5_nothink: Thinking disabled — assistant prefix includes a closed empty
<think> block so the model skips straight to responding.
"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
ConvTemplateRegistry.register_conv_template(
Conversation(
name="qwen3_5",
system_template=f"<|im_start|>system\n{MessagePlaceholders.SYSTEM.value}<|im_end|>\n",
system_message="You are a helpful assistant.",
roles={
"user": "<|im_start|>user",
"assistant": "<|im_start|>assistant\n<think>",
},
seps=["<|im_end|>\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<|endoftext|>", "<|im_end|>"],
stop_token_ids=[248046, 248044],
)
)
ConvTemplateRegistry.register_conv_template(
Conversation(
name="qwen3_5_nothink",
system_template=f"<|im_start|>system\n{MessagePlaceholders.SYSTEM.value}<|im_end|>\n",
system_message="You are a helpful assistant.",
roles={
"user": "<|im_start|>user",
"assistant": "<|im_start|>assistant\n<think>\n\n</think>\n",
},
seps=["<|im_end|>\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<|endoftext|>", "<|im_end|>"],
stop_token_ids=[248046, 248044],
)
)
python/mlc_llm/conversation_template/redpajama.py
+20
View File
@@ -0,0 +1,20 @@
"""RedPajama default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# RedPajama Chat
ConvTemplateRegistry.register_conv_template(
Conversation(
name="redpajama_chat",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "<human>", "assistant": "<bot>"},
seps=["\n"],
role_content_sep=": ",
role_empty_sep=":",
stop_str=["<human>"],
stop_token_ids=[0],
)
)
python/mlc_llm/conversation_template/registry.py
+85
View File
@@ -0,0 +1,85 @@
"""The conversation template registry and presets in MLC LLM"""
from typing import ClassVar, Dict, Optional # noqa: UP035
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
class ConvTemplateRegistry:
"""Global conversation template registry for preset templates."""
_conv_templates: ClassVar[Dict[str, Conversation]] = {} # noqa: UP006
@staticmethod
def register_conv_template(conv_template: Conversation, override: bool = False) -> None:
"""Register a new conversation template in the global registry.
Using `override = True` to override the previously registered
template with the same name.
"""
name = conv_template.name
if name is None:
raise ValueError("The template to register should have non-None name.")
if name in ConvTemplateRegistry._conv_templates and not override:
raise ValueError(
"The name of the template has been registered "
f"for {ConvTemplateRegistry._conv_templates[name].model_dump_json(by_alias=True)}"
)
ConvTemplateRegistry._conv_templates[name] = conv_template
@staticmethod
def get_conv_template(name: str) -> Optional[Conversation]:
"""Return the conversation template specified by the given name,
or None if the template is not registered.
"""
return ConvTemplateRegistry._conv_templates.get(name, None)
# ChatML
ConvTemplateRegistry.register_conv_template(
Conversation(
name="chatml",
system_template=f"<|im_start|>system\n{MessagePlaceholders.SYSTEM.value}<|im_end|>\n",
system_message=(
"A conversation between a user and an LLM-based AI assistant. The "
"assistant gives helpful and honest answers."
),
roles={"user": "<|im_start|>user", "assistant": "<|im_start|>assistant"},
seps=["<|im_end|>\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<|im_end|>"],
stop_token_ids=[2],
)
)
# ChatML without a system prompt
ConvTemplateRegistry.register_conv_template(
Conversation(
name="chatml_nosystem",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "<|im_start|>user", "assistant": "<|im_start|>assistant"},
seps=["<|im_end|>\n"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<|im_end|>"],
stop_token_ids=[2],
)
)
# Vanilla LM
ConvTemplateRegistry.register_conv_template(
Conversation(
name="LM",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "", "assistant": ""},
seps=[""],
role_content_sep="",
role_empty_sep="",
stop_str=[],
stop_token_ids=[2],
system_prefix_token_ids=[1],
)
)
python/mlc_llm/conversation_template/rwkv.py
+24
View File
@@ -0,0 +1,24 @@
"""RWKV default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# RWKV World
ConvTemplateRegistry.register_conv_template(
Conversation(
name="rwkv_world",
system_template=f"User: hi\n\nAssistant: {MessagePlaceholders.SYSTEM.value}",
system_message=(
"Hi. I am your assistant and I will provide expert full response "
"in full details. Please feel free to ask any question and I will "
"always answer it."
),
roles={"user": "User", "assistant": "Assistant"},
seps=["\n\n"],
role_content_sep=": ",
role_empty_sep=": ",
stop_str=["\n\n"],
stop_token_ids=[0],
)
)
python/mlc_llm/conversation_template/stablelm.py
+59
View File
@@ -0,0 +1,59 @@
"""StableLM default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# StableLM Tuned Alpha
ConvTemplateRegistry.register_conv_template(
Conversation(
name="stablelm",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message=(
"<|SYSTEM|># StableLM Tuned (Alpha version)\n"
"- StableLM is a helpful and harmless open-source AI language model developed by "
"StabilityAI.\n"
"- StableLM is excited to be able to help the user, but will refuse to do "
"anything that could be considered harmful to the user.\n"
"- StableLM is more than just an information source, StableLM is also able to "
"write poetry, short stories, and make jokes.\n"
"- StableLM will refuse to participate in anything that could harm a human."
),
roles={"user": "<|USER|>", "assistant": "<|ASSISTANT|>"},
seps=[""],
role_content_sep=": ",
role_empty_sep=": ",
stop_str=[""],
stop_token_ids=[50278, 50279, 50277, 1, 0],
)
)
# StableLM 3B
ConvTemplateRegistry.register_conv_template(
Conversation(
name="stablelm-3b",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "<|user|>", "assistant": "<|assistant|>"},
seps=["<|endoftext|>", "<|endoftext|>"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<|endoftext|>"],
stop_token_ids=[0],
)
)
# StableLM-2
ConvTemplateRegistry.register_conv_template(
Conversation(
name="stablelm-2",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "<|user|>", "assistant": "<|assistant|>"},
seps=["<|endoftext|>", "<|endoftext|>"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["<|endoftext|>"],
stop_token_ids=[100257],
)
)
python/mlc_llm/conversation_template/tinyllama.py
+20
View File
@@ -0,0 +1,20 @@
"""Tiny Llama default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# TinyLlama v1.0
ConvTemplateRegistry.register_conv_template(
Conversation(
name="tinyllama_v1_0",
system_template=f"<|system|>\n{MessagePlaceholders.SYSTEM.value}</s>",
system_message="You are a helpful chatbot.",
roles={"user": "<|user|>", "assistant": "<|assistant|>"},
seps=["</s>"],
role_content_sep="\n",
role_empty_sep="\n",
stop_str=["</s>"],
stop_token_ids=[2],
)
)
python/mlc_llm/conversation_template/wizardlm.py
+40
View File
@@ -0,0 +1,40 @@
"""WiazrdLM and Coder default templates"""
from mlc_llm.protocol.conversation_protocol import Conversation, MessagePlaceholders
from .registry import ConvTemplateRegistry
# Wizard LM 7B
ConvTemplateRegistry.register_conv_template(
Conversation(
name="wizardlm_7b",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message="",
roles={"user": "User", "assistant": "Response"},
seps=["###"],
role_content_sep=": ",
role_empty_sep=":",
stop_str=["###"],
stop_token_ids=[2],
system_prefix_token_ids=[1],
)
)
# WizardCoder or WizardMath
ConvTemplateRegistry.register_conv_template(
Conversation(
name="wizard_coder_or_math",
system_template=f"{MessagePlaceholders.SYSTEM.value}",
system_message=(
"Below is an instruction that describes a task. Write a response that appropriately "
"completes the request."
),
roles={"user": "Instruction", "assistant": "Response"},
seps=["\n\n### ", "\n\n### "],
role_content_sep=":\n",
role_empty_sep=":\n",
stop_str=["</s>"],
stop_token_ids=[2],
system_prefix_token_ids=[1],
)
)