Skip to content

Evaluate frozen V121 embeddings only #2

Evaluate frozen V121 embeddings only

Evaluate frozen V121 embeddings only #2

name: Trace Ace V121 Eval Only
on:
pull_request:
branches: [agent/v121-cache-batch2]
paths:
- '.github/workflows/trace-ace-v121-eval-only.yml'
workflow_dispatch:
concurrency:
group: trace-ace-v121-eval-only
cancel-in-progress: false
env:
PREPARED_RUN_ID: '32400309220'
EMBEDDING_RUN_ID: '32402681183'
jobs:
evaluate:
runs-on: ubuntu-24.04
timeout-minutes: 20
permissions:
actions: read
contents: read
steps:
- uses: actions/checkout@v4
- uses: actions/setup-python@v5
with:
python-version: '3.12'
cache: pip
- name: Install evaluation dependencies
run: python -m pip install --disable-pip-version-check numpy pandas scipy scikit-learn fastembed==0.8.0
- name: Download exact frozen V121 prepared artifact
uses: actions/download-artifact@v4
with:
name: v121-prepared
path: v121_prepared
repository: heathsanchez/mathgraph
run-id: ${{ env.PREPARED_RUN_ID }}
github-token: ${{ secrets.GITHUB_TOKEN }}
- name: Verify frozen manifest
run: |
test -f v121_prepared/manifest.json
grep -q 'b1612f9fe4558680e468afb2a2452b75c603c244934fe62f7345feee68a61bc1' v121_prepared/manifest.json
- name: Download all 128 frozen embedding shards from original run
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
run: |
mkdir -p v121_embedding_shards
for i in $(seq 0 127); do
gh run download "$EMBEDDING_RUN_ID" -n "v121-embeddings-$i" -D "v121_embedding_shards/$i"
done
test "$(find v121_embedding_shards -name 'v121_embeddings_shard_*.npz' | wc -l)" -eq 128
- name: Evaluate unchanged frozen V121 precommit
run: |
cd competitions/trace_the_ace
python v121_staged_transport.py evaluate --dir ../../v121_prepared \
--embeddings ../../v121_embedding_shards --out ../../v121_pretrained_semantic_residual.json
- name: Show decision
run: cat v121_pretrained_semantic_residual.json
- uses: actions/upload-artifact@v4
with:
name: trace-ace-v121-pretrained-semantic-residual
path: v121_pretrained_semantic_residual.json
retention-days: 14