Remove function calling #102
Conversation
…ocess_json_response()
gpt-3_5 returns technologies such as "Backend: Node.js with Mongo database (Mongoose)" codellama throws an error due to missing `choices`
| @@ -51,7 +51,7 @@ https://github.com/Pythagora-io/gpt-pilot/assets/10895136/0495631b-511e-451b-93d | |||
| # 🔌 Requirements | |||
|
|
|||
|
|
|||
| - **Python** | |||
| - **Python >= 3.10** | |||
There was a problem hiding this comment.
why python 3.10? I am using 3.9.6 on my current macbook and it works fine
There was a problem hiding this comment.
I ran into an (minor) issue with typing and figured that Python 3.10 came out in 2021 so it would not be likely to be used anymore.
utils/function_calling.py:5: in <module>
JsonType = str | int | float | bool | None | list["JsonType"] | dict[str, "JsonType"]
E TypeError: unsupported operand type(s) for |: 'type' and 'type'
|
|
||
| # TODO: I don't think llm_response would ever be 'DONE'? |
There was a problem hiding this comment.
you are right, once it is done it will return 'INSTALLED' which is defined in 'development/env_setup/install_next_technology.prompt'. This should be updated and maybe best way would be to put it in const/llm.py with other llm constants
There was a problem hiding this comment.
but also have in mind that this feature is disabled at the moment (return on line 312) so it is not important to fix it now
| cli_response, response = execute_command(convo.agent.project, command, timeout) | ||
| if response is None: | ||
| response = convo.send_message('dev_ops/ran_command.prompt', | ||
| # TODO: Prompt mentions `command` could be `INSTALLED` or `NOT_INSTALLED`, where is this handled? |
There was a problem hiding this comment.
for devops agent while installing tech dependencies response will be INSTALLED or NOT_INSTALLED but for developer agent, or in other words commands executed while coding eg. npm run start, it can return DONE or NEEDS_DEBUGGING.
pilot/utils/function_calling.py
Outdated
| return | ||
|
|
||
| model: str = gpt_data['model'] | ||
| is_llama = 'llama' in model or 'anthropic' in model |
There was a problem hiding this comment.
why is it different for llama compared to all other llms, or do we have to add implementation for each llm individually?
There was a problem hiding this comment.
I need to rename this to is_instruct. Wrapping in [INST] seemed to have a negative affect for some models.
pilot/utils/function_calling.py
Outdated
| model: str = gpt_data['model'] | ||
| is_llama = 'llama' in model or 'anthropic' in model | ||
|
|
||
| # if model == 'gpt-4': |
There was a problem hiding this comment.
why is this commented part here?
|
I still need to fix the |
Need to fix prompts for GPT-4
See also notes in #99