Lee Penkman

autocompaction

8

noticed this kind of thing https://github.com/sst/opencode/issues/1092 where prompts are quickly too long, and this way it autocompacts if its over roughly 80k tokens

## Summary - default to SmolLM3-3B weights - add simple example script for loading SmolLM3 - allow db_models to work without GCP credentials - fix inference_server import error - skip...

## Summary - switch default model paths to Google Gemma - lazily import torch in link_enricher and use Gemma pipeline for captions - update example script to use Gemma -...

## Summary - implement `vllm_inference` helper - integrate vLLM into server with `USE_VLLM` toggle - document vLLM installation and env var - provide CLI helper and tests for new logic...

## Summary - introduce `/api/v1/generate_speech_stream` endpoint returning streamed chunks - support `chunk_words` parameter with environment variable default - add async chunk generator and Python example - document API via new...

## Summary - remove Python 3.11 from the CI test matrix ## Testing - `ruff check .` *(fails: `cgi` and `time` unused in `gameon/facebook.py`)* - `make test` ------ https://chatgpt.com/codex/tasks/task_e_6843a64af0a48333b8f117d0764f48b0

## Summary - separate integration tests and ignore them in pytest.ini - add ruff config and CI workflow using uv - mark heavy tests as integration and skip when deps...

## Summary - separate integration tests and ignore them in pytest.ini - add ruff config and CI workflow using uv - mark heavy tests as integration and skip when deps...

## Summary - add vLLM based inference helper with min-probability and sentence stopping - integrate vLLM into the FastAPI server when available - make `tests/conftest` resilient if `httpx` is missing...