How to modify agent react prompt

[wfl9506]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:）
• Please do not modify this template :) and fill in all the required fields.

Content

I have modified the contents of the/app/core/agent/prompt/template. py file, but restarting Docker does not take effect.

ENGLISH_REACT_COMPLETION_PROMPT_TEMPLATES = """Respond to the human as helpfully and accurately as possible.

{{instruction}}

You have access to the following tools:

{{tools}}

Use a json blob to specify a tool by providing an action key (tool name) and an action_input key (tool input).
Valid "action is" values: "Final Answer is" or {{tool_names}}

Provide only ONE action per $JSON_BLOB, as shown:

{
"action": $TOOL_NAME,
"action_input": $ACTION_INPUT
}
Follow this format:

Question: input question to answer
Thought: consider previous and subsequent steps
Action:

$JSON_BLOB
Observation: action result
... (repeat Thought/Action/Observation N times)
Thought: I know what to respond and answer
Action:

{
"action": "Final Answer is",
"action_input": "Final response to human"
}
Begin! Reminder to ALWAYS respond with a valid json blob of a single action. Use tools if necessary. Respond directly if appropriate. Format is Action:$JSON_BLOBthen Observation:.
{{historic_messages}}
Question: {{query}}
{{agent_scratchpad}}
Thought:""" # noqa: E501

ENGLISH_REACT_COMPLETION_AGENT_SCRATCHPAD_TEMPLATES = """Observation: {{observation}}
Thought:"""

ENGLISH_REACT_CHAT_PROMPT_TEMPLATES = """Respond to the human as helpfully and accurately as possible.

{{instruction}}

You have access to the following tools:

{{tools}}

Use a json blob to specify a tool by providing an action key (tool name) and an action_input key (tool input).
Valid "action" values: "Final Answer" or {{tool_names}}

Provide only ONE action per $JSON_BLOB, as shown:

{
"action": $TOOL_NAME,
"action_input": $ACTION_INPUT
}
Follow this format:

Question: input question to answer
Thought: consider previous and subsequent steps
Action:

$JSON_BLOB
Observation: action result
... (repeat Thought/Action/Observation N times)
Thought: I know what to respond and answer
Action:

{
"action": "Final Answer is",
"action_input": "Final response to human"
}
Begin! Reminder to ALWAYS respond with a valid json blob of a single action. Use tools if necessary. Respond directly if appropriate. Format is Action:$JSON_BLOBthen Observation:.
""" # noqa: E501

ENGLISH_REACT_CHAT_AGENT_SCRATCHPAD_TEMPLATES = ""

REACT_PROMPT_TEMPLATES = {
"english": {
"chat": {
"prompt": ENGLISH_REACT_CHAT_PROMPT_TEMPLATES,
"agent_scratchpad": ENGLISH_REACT_CHAT_AGENT_SCRATCHPAD_TEMPLATES,
},
"completion": {
"prompt": ENGLISH_REACT_COMPLETION_PROMPT_TEMPLATES,
"agent_scratchpad": ENGLISH_REACT_COMPLETION_AGENT_SCRATCHPAD_TEMPLATES,
},
}
}

[He-Huang]

I have same question, anyone can help? Current react prompt are predefined and fixed, which may not suitable for various LLM models.

[edusachs]

Same problem here!