parakeet_web/.github/workflows/test.yml at main · thiswillbeyourgithub/parakeet_web · GitHub

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
# CI test gate for pull requests. Mirrors the local pre-push hook (the fast
# tiers) and adds the heavy tier-3 E2E as a separate job.
#
# NOTE: to make these checks actually BLOCK a merge, a repo admin must enable
# them as required status checks under Settings -> Branches -> branch protection
# (the workflow alone runs the checks but does not enforce them).
#
# Generated with Claude Code.
name: tests

on:
  pull_request:
  push:
    branches: [master, fork]

jobs:
  fast:
    # Tier 1 (unit) + tier 2 (http integration). Fast, no model download.
    name: unit + http
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: 22
      # Tier-1 imports parakeet.js (needs onnxruntime-web); tier-2 spawns the
      # signaling server (needs express + ws). Each package vendors its own lock.
      - name: Install app deps (onnxruntime-web)
        run: npm ci
        working-directory: app
      - name: Install signaling deps (express, ws)
        run: npm ci
        working-directory: signaling
      - name: Run unit + http tiers
        run: npm test

  e2e:
    # Tier 3: full-transcription happy path in headless Chromium. Heavy (~1.7 GB
    # of weights: both int8 encoders, the decoder, and the diarization pair), so
    # they are cached across runs.
    name: e2e (full transcription)
    runs-on: ubuntu-latest
    timeout-minutes: 30
    env:
      PARAKEET_E2E_MODEL_DIR: ${{ github.workspace }}/.e2e-models
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: 22

      - name: Install root deps (Playwright)
        run: npm ci
      - name: Install UI deps
        run: npm ci
        working-directory: app/ui
      - name: Build UI
        run: npm run build
        working-directory: app/ui

      - name: Cache int8 model weights
        uses: actions/cache@v4
        with:
          path: ${{ env.PARAKEET_E2E_MODEL_DIR }}
          # Keyed on the fetch list itself: editing scripts/fetch-e2e-models.mjs
          # (new file, repo/revision change, optional -> required flip) mints a
          # fresh cache so newly listed weights get baked in. Optional entries
          # that 404'd at save time are retried every run (cheap HEAD-sized
          # miss) until a list edit re-keys the cache.
          key: parakeet-e2e-models-${{ hashFiles('scripts/fetch-e2e-models.mjs') }}
      - name: Fetch model weights (cache miss only)
        run: npm run e2e:models

      # The batching numeric-equivalence gate (batched encode is byte-identical
      # to standalone encode) is a tier-1 unit test, but it needs the real int8
      # weights and so SELF-SKIPS in the fast job (which downloads no model).
      # Run it here, where the weights already exist, so a batching regression
      # cannot merge green. Uses the vendored ORT (no extra deps); modelDir comes
      # from PARAKEET_E2E_MODEL_DIR; fetch-e2e-models nests by repo, so the test
      # descends into the ASR repo folder itself (asrRootIn) rather than
      # assuming a flat root, which would leave it self-skipping and green.
      - name: Batching equivalence gate (needs weights)
        run: node --test test/unit/encode-batch-equivalence.test.mjs

      - name: Install Playwright Chromium
        run: npx playwright install --with-deps chromium
      - name: Run E2E
        run: npm run test:e2e