Implementation:InternLM Lmdeploy BaseChatTemplate Messages2prompt

Knowledge Sources	LMDeploy Chat Template
Domains	NLP, Prompt_Engineering
Last Updated	2026-02-07 15:00 GMT

Overview

Concrete tool for converting OpenAI-format message lists into model-specific prompt strings provided by the LMDeploy library.

Description

The BaseChatTemplate class provides prompt formatting methods that apply model-specific tokens (system prefix, user/assistant markers, end-of-turn tokens) to raw messages. It uses a registry system (mmengine MODELS) where each model family registers its template, and lookup happens automatically based on model name.

Usage

This is called automatically during Pipeline inference. Import directly when debugging prompt formatting or building custom prompt engineering workflows outside the standard pipeline.

Code Reference

Source Location

• Repository: lmdeploy
• File: lmdeploy/model.py
• Lines: L110-191 (BaseChatTemplate class), L776-793 (get_chat_template)

Signature

class BaseChatTemplate:
    def __init__(self, system='', meta_instruction='', eosys='',
                 user='', eoh='', assistant='', eoa='',
                 separator='', tool='', eotool='',
                 capability='chat', stop_words=None, **kwargs):
        ...

    def messages2prompt(self, messages, sequence_start=True, **kwargs) -> str:
        """Convert OpenAI-format messages to a prompt string."""
        ...

    def get_prompt(self, prompt, sequence_start=True) -> str:
        """Wrap a plain string prompt with chat template."""
        ...

Import

from lmdeploy.model import BaseChatTemplate, get_chat_template

I/O Contract

Inputs

Name	Type	Required	Description
messages	str or List[dict]	Yes	Plain string or OpenAI-format message list [{"role": "user", "content": "..."}]
sequence_start	bool	No	Whether this is the start of a conversation (default: True)

Outputs

Name	Type	Description
prompt	str	Formatted prompt string with model-specific tokens applied

Usage Examples

Direct Template Usage

from lmdeploy.model import get_chat_template

# Get the template for InternLM2
template = get_chat_template('internlm2')

# Format OpenAI-style messages
messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "What is 2+2?"}
]
prompt = template.messages2prompt(messages)
print(prompt)
# Output includes <|im_start|>system\n...<|im_end|>\n<|im_start|>user\n...<|im_end|>\n<|im_start|>assistant\n

Related Pages

Implements Principle

• Principle:InternLM_Lmdeploy_Chat_Template_Formatting

Requires Environment

• Environment:InternLM_Lmdeploy_Python_Dependencies