Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
Python agentaianthropicclaude-codecompressioncontext-engineeringcontext-windowcursorfastapilangchainllmmcpopenaiprompt-engineeringproxypythonragtoken-optimizationtokenstypescript
ci: add experimental intelligent+parallel pipeline (ci-fast.yml)
Tejas Chopra8062e7f7f5b8
diff --git a/.github/workflows/ci-fast.yml b/.github/workflows/ci-fast.ymlnew file mode 100644index 000000000..32bf0071b--- /dev/null+++ b/.github/workflows/ci-fast.yml@@ -0,0 +1,155 @@+name: CI Fast (experimental)++# Intelligent + parallel CI, running ALONGSIDE the existing ci.yml so it can be+# validated/measured without touching required status checks. Once it's proven+# green + faster, fold it into ci.yml and update branch protection.+#+# Design:+# * changes — paths-filter: skip the heavy work for docs-only PRs.+# * build-wheel — compile the Rust extension ONCE (maturin), share via artifact.+# * lint — ruff + mypy, once (3.12).+# * test — 4 parallel shards (pytest-split), each a fresh runner VM so+# the suite's shared-state tests (repo-root db, port 8787) don't+# collide. CPU-only torch + cached HF model keep installs fast.++on:+ pull_request:+ branches: [main]+ workflow_dispatch:++concurrency:+ group: ci-fast-${{ github.ref }}+ cancel-in-progress: true++env:+ PY_VERSION: "3.12"+ # CPU-only torch index — CI runners have no GPU; the default CUDA wheels pull+ # ~2.5 GB of nvidia libs we never use.+ PIP_EXTRA_INDEX_URL: https://download.pytorch.org/whl/cpu++jobs:+ changes:+ runs-on: ubuntu-latest+ outputs:+ code: ${{ steps.filter.outputs.code }}+ steps:+ - uses: actions/checkout@v4+ - uses: dorny/paths-filter@v3+ id: filter+ with:+ filters: |+ code:+ - 'headroom/**'+ - 'crates/**'+ - '**/*.rs'+ - 'pyproject.toml'+ - 'Cargo.toml'+ - 'Cargo.lock'+ - 'tests/**'+ - 'scripts/tests/**'+ - '.github/workflows/ci-fast.yml'++ build-wheel:+ needs: changes+ if: needs.changes.outputs.code == 'true'+ runs-on: ubuntu-latest+ steps:+ - uses: actions/checkout@v4+ - uses: actions/setup-python@v5+ with:+ python-version: ${{ env.PY_VERSION }}+ - uses: dtolnay/[email protected]+ - uses: Swatinem/rust-cache@v2+ with:+ workspaces: ". -> target"+ - name: Build wheel once (maturin)+ run: |+ python -m pip install --upgrade pip maturin+ maturin build --release --out dist --interpreter "python${PY_VERSION}"+ - uses: actions/upload-artifact@v4+ with:+ name: headroom-wheel+ path: dist/*.whl+ retention-days: 1++ lint:+ needs: changes+ if: needs.changes.outputs.code == 'true'+ runs-on: ubuntu-latest+ steps:+ - uses: actions/checkout@v4+ - uses: actions/setup-python@v5+ with:+ python-version: ${{ env.PY_VERSION }}+ - name: Install ruff + mypy+ run: python -m pip install --upgrade pip ruff mypy+ - name: ruff check+ run: ruff check .+ - name: ruff format --check+ run: ruff format --check .+ - name: mypy+ run: mypy headroom --ignore-missing-imports++ test:+ needs: [changes, build-wheel]+ if: needs.changes.outputs.code == 'true'+ runs-on: ubuntu-latest+ strategy:+ fail-fast: false+ matrix:+ shard: [1, 2, 3, 4]+ env:+ FASTEMBED_CACHE_PATH: ${{ github.workspace }}/.fastembed-cache+ steps:+ - uses: actions/checkout@v4+ - uses: actions/setup-python@v5+ with:+ python-version: ${{ env.PY_VERSION }}++ - name: Cache pip+ uses: actions/cache@v4+ with:+ path: ~/.cache/pip+ key: ${{ runner.os }}-pipfast-${{ env.PY_VERSION }}-${{ hashFiles('pyproject.toml') }}+ restore-keys: ${{ runner.os }}-pipfast-${{ env.PY_VERSION }}-++ - name: Cache HuggingFace + fastembed models+ uses: actions/cache@v4+ with:+ path: |+ ~/.cache/huggingface+ ${{ github.workspace }}/.fastembed-cache+ key: ${{ runner.os }}-models-allMiniLM-v1++ - name: Download prebuilt wheel+ uses: actions/download-artifact@v4+ with:+ name: headroom-wheel+ path: dist++ - name: Install (CPU torch + prebuilt wheel + dev deps, no cargo rebuild)+ run: |+ python -m pip install --upgrade pip+ # Install torch CPU explicitly first so the editable/extra resolve reuses it.+ pip install torch --index-url https://download.pytorch.org/whl/cpu+ WHEEL="$(ls dist/*.whl)"+ pip install "${WHEEL}[dev]"+ pip install pytest-split+ python -c "from headroom._core import DiffCompressor; print('headroom._core OK')"++ - name: Pre-fetch embedding model (resilient to HF rate limits)+ env:+ HF_TOKEN: ${{ secrets.HF_TOKEN }}+ HF_HUB_DISABLE_TELEMETRY: "1"+ run: |+ for i in 1 2 3 4 5; do+ if python -c "from sentence_transformers import SentenceTransformer; SentenceTransformer('all-MiniLM-L6-v2')"; then exit 0; fi+ echo "::warning::model fetch attempt $i failed; backing off"; sleep $((i * 20))+ done+ echo "::error::could not fetch all-MiniLM-L6-v2 from HuggingFace"; exit 1++ - name: Run test shard ${{ matrix.shard }}/4+ run: |+ pytest tests scripts/tests \+ --splits 4 --group ${{ matrix.shard }} \+ --tb=short -q