Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

imp: update of the tool prompt #152

Draft
wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

imp: update of the tool prompt #152

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
63 changes: 48 additions & 15 deletions mkdocs/conceptual/prompts/instructions.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,26 +1,59 @@
# Instructions Prompt

The Instruction Prompt explains the [Tool Agent](../cheshire_cat/agent.md#tool-chain) how to format its reasoning.
The Instruction Prompt explains the [Tool Agent](../cheshire_cat/agent.md#tool-chain) how to format its reasoning.
The [Tool Agent](../cheshire_cat/agent.md) uses a [chain](https://docs.langchain.com/docs/components/chains/llm-chain)
to decide *when* and *which* [tool](../plugins.md) is the most appropriate to fulfill the user's needs.

By default, it is set to Langchain [instructions format](https://api.python.langchain.com/en/latest/agents/langchain.agents.conversational.base.ConversationalAgent.html?highlight=prompt%20format_instruction)
which looks like this:

```python
instructions = """
To use a tool, please use the following format:

Thought: Do I need to use a tool? Yes
Action: the action to take, should be one of [{tool_names}]
Action Input: the input to the action
Observation: the result of the action

When you have a response to say to the Human, or if you do not need to use a tool, you MUST use the format:

Thought: Do I need to use a tool? No

AI: [your response here]"""
instructions = """Create a JSON action to complete the action sequence, with the correct "action" and "action_input" to help the Human.
You can use one of these actions:
{tools}
- "final_answer": Use this action to finish or no relevant action is available. Input is always null.

## To add an action, use only the following format:
\```json
{{
"action": // str - The name of the action to take, should be one of [{tool_names}, "final_answer"]
"action_input": // str or null - The input to the action
}}
\```

{examples}

## Action output
After each action there will be an action output in this format:
\```json
{{
"action_output": // output of the preceding action
}}
\```

## Final answer / no action available
When you have a final answer (or no tools are relevant), use the following format:
\```json
{{
"action": "final_answer",
"action_input": null
}}
\```

## Conversation with Human:
{chat_history}

## Actions sequence used until now:
{agent_scratchpad}

## Next action:
\```json
"""
```

where the placeholder `{tool_names}` is replaced with the list of Python [tools](../plugins.md) retrieved from the [procedural memory](../memory/long_term_memory.md).
where:

- the placeholder `{tools}` is replaced with the list of Python [tools](../plugins.md) retrieved from the [procedural memory](../memory/long_term_memory.md);
- the `{examples}` is replaced with some example actions given to the model;
- the `{chat_history}` is replaced with the chat history with the user;
- the `{agent_scratchpad}` is replaced with the agent scratchpad information about the actions' sequence used until that iteration.