fix: patch streaming API code #1693
Commits on Aug 28, 2024
-
fix: patch stream_steps to properly trigger + return compact JSON ins…
…tead of JSON with newlines
-
-
-
feat: add dummy Message creation at the top of the stream handler so …
…that we can pass a MemGPT message ID back in the chunks of our streaming API (previously we hadn't created a Message so there was no Message ID to pass back by the time the streaming started)
-
-
feat: add FunctionCallDelta to the MemGPT message schema, since this …
…was an unsupported data type with the existing set of schemas
-
feat: support passing ID when doing a dict-to-message Message constru…
…ctor (TODO this is intended to be used to allow creating a message in agent.py that uses the ID that came inside of the ChatCompletionResponse
-
fix: fix issues with process_chunk that broke when we added typing to…
… the chunks/message objects
-
fix: allow process_chunk to take extra message IDs and created_ats to…
… pass through in the chunks from the MemGPT API
-
feat: provide option in the openai stream handler determining whether…
… or not we want to use timestamps/ids from the chunks coming from the server, or from a message we created ahead of time
-
-
fix: fix pretty bad bug where we were always creating new timestamps …
…on the return payload from any MemGPT message POST SSE return
-
-
fix: this was the final step - we create the Message IDs ahead of tim…
…e for streaming, but this enables actually persisting them (previously they were just fake throwaways). NOTE: this isn't really done very well since it has a hack assuming the 'message-' prefix on a ChatCompletionResponse means that we intended to use the .id property on the ChatCompletionResponse in subsequent Message creations
-
-
-
-
refactor: placate pylance and potentially fix pytest error by reverti…
…ng back to duplicated code style
