Add fixed-15% protein MLM autoresearch and CPU validation

.dockerignore

@@ -1,14 +1,23 @@
-# Data directories - these will be accessed via volume mount
-data/
-results/
-logs_to_keep/
+# Generated data payloads are accessed via a volume mount. Keep the root
+# compatibility modules and the packaged speedrunning_plms.data source code.
+/data/*
+!/data/
+!/data/*.py
+/results/
+/logs_to_keep/
+/runs/
+/targets.local.json
 
 # Cache directories
 .cache/
+.pytest_cache/
 __pycache__/
 *.pyc
 *.pyo
 *.pyd
+*.egg-info/
+build/
+dist/
 
 # Git files
 .git/
@@ -27,6 +36,7 @@ Thumbs.db
 
 # Python virtual environments
 venv/
+.venv/
 env/
 .env
 
@@ -45,4 +55,4 @@ logs/
 
 # Temporary files
 tmp/
-temp/ 
+temp/

.github/workflows/gh-pages.yml

@@ -3,7 +3,7 @@ on:
   push:
     branches: [main]
 
-permissions:          # 👈 add this block (workflow- or job-level)
+permissions:
   contents: write
 
 jobs:
@@ -14,6 +14,6 @@ jobs:
 
       - uses: peaceiris/actions-gh-pages@v4
         with:
-          publish_dir: docs          # folder to publish
-          publish_branch: gh-pages   # default is fine; adjust if you use a different branch
-          github_token: ${{ secrets.PERSONAL_TOKEN }}
+          publish_dir: docs
+          publish_branch: gh-pages
+          github_token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/tests.yml

@@ -0,0 +1,40 @@
+name: CPU tests
+
+on:
+  pull_request:
+  push:
+    branches: [main]
+
+permissions:
+  contents: read
+
+concurrency:
+  group: cpu-tests-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ['3.10', '3.12']
+    steps:
+      - uses: actions/checkout@v7
+        with:
+          persist-credentials: false
+      - uses: actions/setup-python@v7
+        with:
+          python-version: ${{ matrix.python-version }}
+          cache: pip
+          cache-dependency-path: pyproject.toml
+      - name: Install CPU dependencies
+        run: |
+          python -m pip install torch==2.6.0 --index-url https://download.pytorch.org/whl/cpu
+          python -m pip install -e ".[test,evaluation]"
+          python -m pip check
+      - name: Run offline tests
+        run: python -m pytest -q --durations=10
+      - name: Test historical score viewer
+        run: node --test tests/test_hub.cjs

.gitignore

@@ -1,7 +1,14 @@
 omgprot50/
 __pycache__/
+.pytest_cache/
 *.parquet
 *.bin
 /logs
 /experiments/*.yaml
 /.cache
+*.egg-info/
+/runs/
+/targets.local.json
+/.venv/
+/data/*/manifest.json
+/data/*/*.pt

Dockerfile

@@ -1,13 +1,10 @@
-# sudo docker build -t speedrun_plm .
-# sudo docker run --gpus all --shm-size=128g -v ${PWD}:/workspace speedrun_plm torchrun --standalone --nproc_per_node=4 train.py
-# docker run --gpus all -v ${PWD}:/workspace speedrun_plm python train.py --bugfix
-# 1️⃣  CUDA / cuDNN base with no Python
+# docker build -t speedrun_plm .
+# docker run --gpus all -v "${PWD}:/workspace" speedrun_plm python train.py --config experiment.json
 FROM nvidia/cuda:12.8.0-cudnn-devel-ubuntu24.04
 
-# 2️⃣  System prerequisites + Python 3.12
-ENV        DEBIAN_FRONTEND=noninteractive \
-           PYTHON_VERSION=3.12.7 \
-           PATH=/usr/local/bin:$PATH
+ENV DEBIAN_FRONTEND=noninteractive \
+    PYTHON_VERSION=3.12.7 \
+    PATH=/usr/local/bin:$PATH
 
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
@@ -27,34 +24,27 @@ RUN curl -fsSLO https://www.python.org/ftp/python/${PYTHON_VERSION}/Python-${PYT
     ln -s /usr/local/bin/python3.12 /usr/local/bin/python && \
     ln -s /usr/local/bin/pip3.12    /usr/local/bin/pip
 
-# 3️⃣  Location of project code (inside image) – NOT shared with host
 WORKDIR /app
 
-# 4️⃣  Copy requirements first for layer caching
+# Cache dependency installation independently of source changes.
 COPY requirements.txt .
 
 RUN pip install --upgrade pip setuptools && \
-    pip install -r requirements.txt -U && \
-    pip install --force-reinstall torch torchvision --index-url https://download.pytorch.org/whl/cu128 -U && \
-    pip install numpy==1.26.4
-
+    pip install torch --index-url https://download.pytorch.org/whl/cu128 -U && \
+    pip install -r requirements.txt
 
-# 5️⃣  Copy the rest of the source
 COPY . .
 
-# 6️⃣  Change working directory to where the volume will be mounted
+RUN pip install -e ".[test,evaluation]"
+
 WORKDIR /workspace
 
-# ──────────────────────────────────────────────────────────────────────────────
-# 7️⃣  Single persistent host volume (/workspace) for *all* artefacts & caches
-#     Bind-mount it when you run the container:  -v ${PWD}:/workspace
-# ──────────────────────────────────────────────────────────────────────────────
+# Prefer the bind-mounted candidate over the image's installed source.
 ENV PROJECT_ROOT=/workspace \
-    TRANSFORMERS_CACHE=/workspace/.cache/huggingface \
+    PYTHONPATH=/workspace/src \
     HF_HOME=/workspace/.cache/huggingface \
     TORCH_HOME=/workspace/.cache/torch \
     XDG_CACHE_HOME=/workspace/.cache \
-    WANDB_DIR=/workspace/logs \
     TQDM_CACHE=/workspace/.cache/tqdm
 
 RUN mkdir -p \
@@ -65,8 +55,6 @@ RUN mkdir -p \
       /workspace/data \
       /workspace/results
 
-# Declare the volume so other developers know it's intended to persist
 VOLUME ["/workspace"]
 
-# 8️⃣  Default command – override in `docker run … python train.py`
-CMD ["bash"]
+CMD ["bash"]

MANIFEST.in

@@ -0,0 +1,11 @@
+include LICENSE
+include README.md
+include requirements.txt
+include prepare.py
+include train.py
+include research.py
+include program.md
+include experiment.json
+recursive-include targets *.json
+recursive-include evaluation *.py *.json
+recursive-include tests *.py