Omar SobhandClaude Fable 5.1 0efb8be745 embedded3 PERF-3 P1-5 (c) part 2 + (d): the device tables persist across steps and are updated in place within the band — one DeviceCut with both phases' aperture / open / activity buffers (the predictor phase is a flag); the faces of the changed cells for apertures, distances, shifts and open flags, the imposition band's faces for the surface velocities, the changed cells for activity and owner, the changed and wall cells for the wall flux, the merged-cell CSR whole; RTX_E3_BAND_CHECK=1 compares every table against a full build (passed); slab CSV byte-identical, device moving/cg green; ny 124 rebuild block 1,910 → 1,578 ms per step (tables 442 → 132) ...

Co-Authored-By: Claude Fable 5.1 <[email protected]>

2026-09-20 18:17:28 -05:00

..

core

rtx-backend-cuda: native index_select / index_add — HetGAT training 2.5x, inference 4.3x

2026-09-11 23:09:01 -05:00

data

style: clear the fmt gate and two lib clippy warnings

2026-09-02 19:15:53 -07:00

integration

fix(deps): bump candle-core/nn/transformers 0.8->0.11 for CUDA 13.1 build

2026-07-09 18:04:46 -07:00

meta

style: cargo fmt --workspace (whitespace/wrapping only, no semantic change)

2026-08-10 07:09:36 -07:00

models

style: cargo fmt --workspace (whitespace/wrapping only, no semantic change)

2026-08-10 07:09:36 -07:00

production

style: cargo fmt --workspace (whitespace/wrapping only, no semantic change)

2026-08-10 07:09:36 -07:00

specialized

embedded3 PERF-3 P1-5 (c) part 2 + (d): the device tables persist across steps and are updated in place within the band — one DeviceCut with both phases' aperture / open / activity buffers (the predictor phase is a flag); the faces of the changed cells for apertures, distances, shifts and open flags, the imposition band's faces for the surface velocities, the changed cells for activity and owner, the changed and wall cells for the wall flux, the merged-cell CSR whole; RTX_E3_BAND_CHECK=1 compares every table against a full build (passed); slab CSV byte-identical, device moving/cg green; ny 124 rebuild block 1,910 → 1,578 ms per step (tables 442 → 132)

2026-09-20 18:17:28 -05:00

tooling

style: cargo fmt --workspace (whitespace/wrapping only, no semantic change)

2026-08-10 07:09:36 -07:00

training

style: clear the fmt gate and two lib clippy warnings

2026-09-02 19:15:53 -07:00

vendor/rnccl

fix(gaps): G0/G2/G5/G8 — eliminate unimplemented! panics, re-enable rtx-distributed, rtx-tts, fix multimodal forward

2026-06-26 13:40:23 +00:00