NegotiationArena
NegotiationArena copied to clipboard
When I try to run "[NegotiationArena] Scenario Creation Tutorial" I got this error at this code block a1 = ChatGPTAgent( model="gpt-4-1106-preview", agent_name=AGENT_ONE, ) a2 = ChatGPTAgent( model="gpt-4-1106-preview", agent_name=AGENT_TWO, ) c...
I just wanted to run example code and I tried to run [NegotiationArena] Explore BuySell Game. But when I run this line of code ti get example states states =...
Bumps [streamlit](https://github.com/streamlit/streamlit) from 1.28.2 to 1.30.0. Release notes Sourced from streamlit's releases. 1.30.0 What's Changed New Features 🎉 Add support for scroll container via the height parameter by @LukasMasuch in...
Currently, modifying a prompt requires a lot of work (basically, the best way is making a new game or reimporting the GameInterface). is there any way to simplify this?
we should avoid reloading the games everytime
if the palyer split the 100 in equal amounts, the outcome we save in the logs -50, 50. Needs to be 50,50.
https://github.com/vinid/rationality-llms/blob/afa933d51a2aa56367f4b09909098cde7b221de5/game/agents/chatgpt.py#L35 should this be checking for `in self.agent_name` instead of `role`?
Issue: Incorrect mapping of tag and content in the from_name_and_tag_to_message function call. This caused each message to double in size, leading to: Exponential growth in total token count and API...
Issue: (Class AlternatingGameEndsOnTag) The end_tag was restricted to ACCEPTING_TAG only, causing the game to continue with invalid conversations when an offer was rejected. This led to incorrect game behavior. Fix:...
Adds support for using Anthropic models through AWS Bedrock