cmoyates
ce30eddcff
feat: slim forward mode — drop intermediate tensor refs
...
slim=True returns 4-key dict, engine uses it, pipeline.infer keeps full.
2 new tests: key count + bit-exact match vs full output.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-03-09 17:00:21 -02:30
cmoyates
f30ab9255b
chore: fix lint, format, type errors
...
- ruff format: 2 files reformatted
- ruff check: fix import sort in scripts/infer_pytorch.py
- ty: suppress tree_flatten arg-type (MLX stub imprecision) and
optional torch import
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-03-08 21:29:06 -02:30
cmoyates
b3b7df271d
feat: MLX memory optimizations — bf16, fused decode, deterministic GC
...
Phase 1: selective bf16 mixed precision (backbone fp32, decoders bf16,
sigmoid always fp32). dtype param on GreenFormer, all outputs guaranteed
fp32.
Phase 2: FusedDecoderPair batches 3 upsample ops by concatenating
alpha+fg projections along channel axis. Bit-exact parity with unfused
path.
Phase 3: deterministic GC pipeline in tile loop (del + gc.collect +
mx.clear_cache) and engine.py intermediate tensor cleanup.
7 new tests: bf16 dtype/range/nan, fused-vs-unfused parity, fp32
default unchanged. 80 passed, 0 failures.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-03-08 21:21:01 -02:30
cmoyates
838bd0dd39
refactor(tests): restructure suite — 15 files to 8, consolidate parity
...
- Add conftest.py: shared paths, tolerances, skip markers, fixtures
- Consolidate imports/shapes/smoke/forward → test_model_contract.py
- Consolidate all parity (decoder/refiner/backbone/e2e) → test_parity.py
- Fold 2048 slow test into test_engine.py
- Rework test_conversion.py to use public convert_checkpoint() API
- Simplify test_weights.py: drop CLI/env var tests, keep checksum+config
- Rename tiling/compilation test files for consistency
- Delete 10 redundant files
80 tests collected (75 pass, 4 skip, 1 deselected)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-03-03 09:58:53 -03:30
