Danny search and rescue #14
Conversation
… one class for both, ActionLoc
…t_poignancy run_gpt_prompt_summarize_conversation
| Takes in "Answer: {name}" and reduces to just name. | ||
| Also hanldes an input of {name} | ||
| ''' | ||
| name: str |
There was a problem hiding this comment.
Maybe we can call this "area" rather than "name" to give the LLM a bit more of a hint?
There was a problem hiding this comment.
I agree, that could work (not sure if I change that or if you)
| output = safe_generate_response(prompt, gpt_param, 5, fail_safe, | ||
| __func_validate, __func_clean_up, verbose=False) | ||
| #output = safe_generate_response(prompt, gpt_param, 5, fail_safe,__func_validate, __func_clean_up, verbose=False) | ||
| output = generate_structured_response(prompt, gpt_param, ActionLoc ,5, fail_safe,__func_validate, __func_clean_up, verbose=False) |
| @@ -991,7 +1012,16 @@ def get_fail_safe(persona): | |||
|
|
|||
| return output, [output, prompt, gpt_param, prompt_input, fail_safe] | |||
|
|
|||
| class ObjDesc(BaseModel): | |||
| desc: str | |||
There was a problem hiding this comment.
Maybe expand this to "description"?
| @field_validator("desc") | ||
| def max_token_limit(cls, value): | ||
| # Split text by whitespace to count words (tokens) | ||
| tokens = value.split() | ||
| if len(tokens) > 100: | ||
| raise ValueError("Text exceeds the maximum limit of 100 tokens.") | ||
| return value |
There was a problem hiding this comment.
I don't think we need a validator here for token limit. We already tell the LLM to give us a max of 100 down below, so we shouldn't be getting more than 100 anyway
There was a problem hiding this comment.
Yeah we can take that out
| output = ChatGPT_safe_generate_response(prompt, example_output, special_instruction, 3, fail_safe, | ||
| __chat_func_validate, __chat_func_clean_up, True) | ||
| #output = ChatGPT_safe_generate_response(prompt, example_output, special_instruction, 3, fail_safe,__chat_func_validate, __chat_func_clean_up, True) | ||
| output = generate_structured_response(prompt, gpt_param, ObjDesc ,5, fail_safe,__chat_func_validate, __chat_func_clean_up, verbose=False) |
There was a problem hiding this comment.
We shouldn't replace ChatGPT_safe_generate_response with generate_structured_response, as they have slightly different arguments. Instead we should use the Structured Outputs ChatGPT_safe_generate_response conversion that @LemmyTron is working on, so this can wait for that PR to be merged
| class DecideToReact(BaseModel): | ||
| ''' | ||
| Should be a decision 1,2, or 3 | ||
| ''' | ||
| decision: int | ||
|
|
There was a problem hiding this comment.
When there's only a small set of specific values that are valid, you can use an Enum (short for enumerated) to constrain the model field: https://docs.pydantic.dev/2.0/usage/types/enums/
Something like this might work:
class DecideToReactEnum(IntEnum):
one = 1
two = 2
three = 3
class DecideToReact(BaseModel):
decision: DecideToReactEnum
| def run_gpt_generate_safety_score(persona, comment, test_input=None, verbose=False): | ||
| class SafetyScore(BaseModel): | ||
| #safety score should range 1-10 | ||
| output: int |
There was a problem hiding this comment.
Could call this "score" or "safety_score"?
| if isinstance(gpt_response.output, int) and 1 <= gpt_response.output <= 10: | ||
| return gpt_response.output | ||
| raise ValueError("Output is not a valid integer between 1 and 10") |
There was a problem hiding this comment.
Good validation and error logic!
| output = ChatGPT_safe_generate_response_OLD(prompt, 3, fail_safe, | ||
| __chat_func_validate, __chat_func_clean_up, verbose) | ||
| #output = ChatGPT_safe_generate_response_OLD(prompt, 3, fail_safe,__chat_func_validate, __chat_func_clean_up, verbose) | ||
| output = generate_structured_response( | ||
| prompt, | ||
| gpt_param, | ||
| SafetyScore, | ||
| 3, | ||
| fail_safe, | ||
| __chat_func_validate, | ||
| __chat_func_clean_up | ||
| ) |
There was a problem hiding this comment.
Let's also wait on this one since it was using ChatGPT_safe_generate_response_OLD
Just look at differences for my functions in run_gpt_prompt as I implement structured output