CodeWeave-PyTorch/.github/codeweave.config at main · HighTech-Innovators/CodeWeave-PyTorch · 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
# External repository to process
EXTERNAL_REPO_NAME=pytorch
EXTERNAL_REPO_URL=https://github.com/HighTech-Innovators/pytorch
EXTERNAL_REPO_BRANCH=main
EXTERNAL_REPO_WORK_BRANCH=copilot-work

# Phase 1 — Book generation: iterative Copilot passes to produce a structured architecture book
PHASE1_MAX_ITERATIONS=10
PHASE1_MODEL_SCHEDULE=claude-opus-4.6:1,claude-sonnet-4.6:2,claude-haiku-4.5

# Phase 2 — ADR generation: iterative Copilot passes to produce Architecture Decision Records
PHASE2_MAX_ITERATIONS=10
PHASE2_MODEL_SCHEDULE=claude-sonnet-4.6

# Phase 3 — Performance measurement harness design: iterative Copilot passes to produce integration-test/
PHASE3_MAX_ITERATIONS=5
PHASE3_MODEL_SCHEDULE=claude-opus-4.6:1,claude-sonnet-4.6

# Phase 4 — Integration test code generation: iterative Copilot passes to produce runnable test code
PHASE4_MAX_ITERATIONS=5
PHASE4_MODEL_SCHEDULE=claude-opus-4.6:1,claude-sonnet-4.6

# Phase 5 — Source build: Copilot discovers build toolchain and builds src/ from source (ccache-backed)
PHASE5_MODEL=claude-opus-4.6
PHASE5_MAX_ITERATIONS=3

# Phase 6 — Baseline execution: repair loop then N measurement runs to record performance baseline
PHASE6_BASELINE_RUNS=5
PHASE6_MAX_REPAIR_ITERATIONS=6
PHASE6_MODEL=claude-opus-4.6

# Phase 7 — Optimization cycle: hotspot selection + per-optimization generate/build/gate loop + paired A/B measurement
PHASE7_MAX_OPTIMIZATIONS=5
PHASE7_MAX_ITERATIONS=3
PHASE7_MODEL_SCHEDULE=claude-opus-4.6:1,claude-sonnet-4.6
PHASE7_SELECTION_MODEL=claude-opus-4.6
# Differential fuzz gate (correctness gate 7f): variant-vs-base-golden op check.
# Blocking by default; set to 0 only for an op that genuinely cannot be fuzzed.
PHASE7_FUZZ_REQUIRED=1
# A/B measurement (paired A/B blocks per cycle)
PHASE7_BASELINE_RUNS=5
PHASE7_MICROBENCH_MIN_SECONDS=10

# Phase 8 — Aggregate optimization report
PHASE8_MODEL=claude-sonnet-4.6

# === Implementation-specific configuration ===
# These variables are specific to the PyTorch/GenAI scenario and are NOT part of
# CodeWeave's framework configuration. When adapting CodeWeave to a different
# external repository or scenario, replace these with scenario-appropriate values.
GENAI_MODEL=distilgpt2
GENAI_MAX_SECONDS=30

# Git commit author identity
GIT_USER_NAME=github-actions[bot]
GIT_USER_EMAIL=41898282+github-actions[bot]@users.noreply.github.com