File size: 3,445 Bytes
d7b3a74 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 | # Release Notes Version 1.1
These notes summarize local changes since the last commit.
## Highlights
- Added a full async pipeline via `ShinkaEvolveRunner` for concurrent proposal generation and evaluation.
- Introduced prompt co-evolution (system prompt archive, prompt mutation, prompt fitness tracking).
- Expanded island and parent sampling logic, including dynamic island spawning on stagnation.
- Refactored LLM + embedding stack into provider-based modules.
- Major WebUI refresh with dashboard + compare views and new API endpoints.
## Cost Budgeting (`max_api_costs`)
- `max_api_costs` is now a first-class runtime budget guard in evolution runners.
- Budget checks use a **committed cost** model:
- realized DB costs (`api_costs`, `embed_cost`, `novelty_cost`, `meta_cost`)
- plus estimated cost of in-flight work
- Once the budget is reached, new proposals stop and the runner drains ongoing jobs.
- If `num_generations` is omitted, `max_api_costs` is required to bound the run.
## Added
- Async runtime and helpers:
- `shinka/core/async_runner.py`
- `shinka/core/async_summarizer.py`
- `shinka/core/async_novelty_judge.py`
- `shinka/database/async_dbase.py`
- Prompt evolution system:
- `shinka/core/prompt_evolver.py`
- `shinka/database/prompt_dbase.py`
- `shinka/prompts/prompts_prompt_evo.py`
- Fix-mode prompt path for incorrect-only populations:
- `shinka/prompts/prompts_fix.py`
- Island sampling strategies:
- `shinka/database/island_sampler.py` (`uniform`, `equal`, `proportional`, `weighted`)
- New plotting modules:
- `shinka/plots/plot_costs.py`
- `shinka/plots/plot_evals.py`
- `shinka/plots/plot_time.py`
- `shinka/plots/plot_llm.py`
- New docs:
- `docs/async_evolution.md`
- `docs/design/dynamic_evolve_markers.md`
- `docs/design/evaluation_cascades.md`
- New example:
- `examples/game_2048/*`
## Changed
- `ShinkaEvolveRunner` now includes:
- stronger resume behavior (meta memory + bandit state persistence)
- fix-mode sampling fallback when no correct program exists
- richer metadata and cost accounting
- Database model expanded with:
- dynamic island spawning controls
- island selection strategy config
- `system_prompt_id` lineage field
- LLM/embedding refactor:
- `shinka/llm/models/*` replaced by `shinka/llm/providers/*`
- embeddings moved to `shinka/embed/*`
- Evaluation wrapper (`shinka/core/wrap_eval.py`) now supports:
- per-run process parallelism via `run_workers` (with optional `max_workers_cap`)
- deterministic result ordering and clearer worker/serialization error surfacing
- early stopping (`bayesian`, `ci`, `hybrid`)
- early stopping remains sequential-only (`run_workers=1`)
- optional plot artifact generation
- NaN/Inf score guards
- WebUI backend and frontend expanded:
- new endpoints for summary/count/details/prompts/stats/plots
- improved WAL retry behavior under active writes
- added `shinka/webui/index.html` and `shinka/webui/compare.html`
## Packaging and Dependencies
- `pyproject.toml` updates:
- `google-generativeai` -> `google-genai`
- added `psutil`
- pinned `httpx==0.27`
- setuptools packaging config updated
## Tests Added
- `tests/test_async_complexity_1000.py`
- `tests/test_bandit_persistence.py`
- `tests/test_dynamic_islands.py`
- `tests/test_island_sampler.py`
- `tests/test_prompt_evolution.py`
- `tests/test_wrap_eval_parallel.py`
|