Enhancements and New Feature Implementation: Fix rag and new tools #188
Conversation
packages/napthaai/customs/prediction_request_rag/prediction_request_rag.py
Show resolved
Hide resolved
|
@Adamantios @jmoreira-valory I have also changed the default temp for prediction_request, prediciton_request_sme and claude to 0. |
packages/napthaai/customs/prediction_url_cot/prediction_url_cot.py
Outdated
Show resolved
Hide resolved
|
|
||
| results = result_dict | ||
| results = json.dumps(results) | ||
| return results, prediction_prompt, None, counter_callback |
There was a problem hiding this comment.
Fix to match the interface.
packages/napthaai/customs/prediction_request_reasoning/prediction_request_reasoning.py
Outdated
Show resolved
Hide resolved
packages/napthaai/customs/prediction_request_reasoning/prediction_request_reasoning.py
Outdated
Show resolved
Hide resolved
There was a problem hiding this comment.
Please, fix the interface of run methods according to https://github.com/valory-xyz/mech/blob/main/README.md#build-your-own, both at definition, and at any return statement within.
Note that second output value is an Optional[Dict[str, Any]], and most times it is wrongly output a string.
A few instances have been marked in the review, but please, verify to cover all of them.
| "gpt-4": 8192, | ||
| } | ||
| ALLOWED_TOOLS = [ | ||
| "prediction-request-rag", | ||
| ] | ||
| TOOL_TO_ENGINE = {tool: "gpt-3.5-turbo" for tool in ALLOWED_TOOLS} | ||
| TOOL_TO_ENGINE = {tool: "gpt-3.5-turbo-0125" for tool in ALLOWED_TOOLS} |
There was a problem hiding this comment.
When running this tool, we get the following error:
ValueError: Model gpt-3.5-turbo-0125 not supported.
| ALLOWED_TOOLS = [ | ||
| "prediction-request-reasoning", | ||
| ] | ||
| TOOL_TO_ENGINE = {tool: "gpt-4-0125-preview" for tool in ALLOWED_TOOLS} |
There was a problem hiding this comment.
Same issue here:
ValueError: Model gpt-4-0125-preview not supported.
|
I think this should fix the issue @0xArdi |
# Conflicts: # packages/packages.json
Chore/bump 0.14.10
This PR encompasses the following updates:
Bug Fixes for Prediction-Request RAG Tool: We've identified and resolved several issues affecting the prediction-request RAG tool, enhancing its stability and performance. These fixes address bugs that were impacting the tool's accuracy.
Conversion of Market Reasoning Tool: The market reasoning tool has been doing very well on the benchmark. We have converted that into a tool for the trader under
prediciton_request_reasoning.New Tool: Chain of Thought Document Analysis: We have had some success with a new tool implementing COT on document/article retrieved from single URL. If a prediction is not reached, we will use the next url. This tool also performs well in the benchmark. This is under
prediction_url_cot. On benchmark it scored0.68.All can be found under napthaai.
Please review the changes and let us know if there is anything.