SolitaryThinker
8bf6eff18f
[perf]: bank first kernel-level MFU gain via pinned cuBLASLt backward algos
2026-07-22 10:58:25 -07:00
SolitaryThinker
acad815f5c
[perf]: cuBLASLt algo sweep finds 8-9% GEMM band headroom with bit-exact parity
2026-07-22 09:21:15 -07:00
SolitaryThinker
971827ad95
[misc]: add prioritized BF16 kernel research plan toward 50% MFU
2026-07-22 09:13:32 -07:00
SolitaryThinker
e82addb927
[misc]: quantify the 50% MFU feasibility budget; reject FA4 split tuning
2026-07-22 09:10:26 -07:00
SolitaryThinker
26b455b7b6
[misc]: record 8x env-stack gate on degraded rack-059 pair; close BF16 campaign handoff
2026-07-22 09:00:15 -07:00
SolitaryThinker
c459a18978
[misc]: root-cause rack-057 8x MNNVL/IMEX hang; record working multi-node preamble
2026-07-22 08:41:44 -07:00
SolitaryThinker
9435b3f29f
[perf]: accept inductor coordinate-descent env stack; record BF16 campaign gates
2026-07-22 07:54:06 -07:00
SolitaryThinker
3083c59ef4
[misc]: audit and confirm the LTX2 MFU numerator; document the 2450 TFLOP/s convention
2026-07-22 06:06:33 -07:00
SolitaryThinker
6199cbee66
[misc]: re-gate tracker head as timing-neutral on 4x GB200
2026-07-22 05:27:12 -07:00
SolitaryThinker
52f1114dd9
[misc]: add resumable LTX2 MFU tracker
2026-07-22 04:13:27 -07:00
SolitaryThinker
3f3f06541c
[bugfix]: make LTX video validation memory-safe
2026-07-22 03:15:20 -07:00
SolitaryThinker
0e60a0e9cc
[bugfix]: isolate compiled validation inference
2026-07-22 03:14:47 -07:00
SolitaryThinker
20c36acefc
[perf]: group adjacent FSDP2 modules
2026-07-21 23:26:39 -07:00
SolitaryThinker
fa47ce1ab5
[perf]: enable packed LTX projections in overfit recipe
2026-07-21 21:05:59 -07:00
SolitaryThinker
7afd751915
[test]: isolate LTX SP subprocess imports
2026-07-21 21:01:52 -07:00
SolitaryThinker
cc5913e5c4
[test]: cover fused parameter export
2026-07-21 21:00:42 -07:00
SolitaryThinker
26f909c520
[test]: cover packed LTX sequence parallelism
2026-07-21 20:56:27 -07:00
SolitaryThinker
d016237096
[perf]: pack LTX attention projections
2026-07-21 20:51:04 -07:00
SolitaryThinker
002ec0771b
[perf]: retain FSDP parameters across accumulation
2026-07-21 19:44:44 -07:00
SolitaryThinker
49508050b7
[perf]: embed uniform LTX2 timestep once
2026-07-21 18:42:06 -07:00
SolitaryThinker
7f6c290c93
[perf]: defer gradient norm materialization
2026-07-21 18:16:47 -07:00
SolitaryThinker
7c58950a92
[perf]: keep LTX2 overfit input pipeline warm
2026-07-21 17:21:41 -07:00
SolitaryThinker
0b324d0a40
[perf]: skip redundant FSDP accumulation sync
2026-07-21 16:54:40 -07:00
SolitaryThinker
e42cfa5e5b
[perf]: add FSDP symmetric-memory training
2026-07-21 15:49:14 -07:00
SolitaryThinker
7f139e2b28
[perf]: make FSDP reduction precision configurable
2026-07-21 07:19:23 -07:00
SolitaryThinker
acdbc0a614
[perf]: batch FSDP gradient clipping
2026-07-21 07:08:38 -07:00
SolitaryThinker
421227ecd4
[perf]: omit unused LTX2 audio weights in training
2026-07-21 06:20:36 -07:00
SolitaryThinker
0431d4e8a9
[perf]: enable regional compile for training
2026-07-21 06:07:35 -07:00
SolitaryThinker
cfc110cb6b
[perf]: make FSDP forward reshard configurable
2026-07-21 05:18:52 -07:00
SolitaryThinker
9582f012df
[perf]: add opt-in fused AdamW training
...
Expose PyTorch's CUDA fused AdamW through modular training config while preserving its current automatic default. Seed optimizer step state on-device when fused or capturable so distributed-checkpoint resume remains valid.
2026-07-21 04:03:24 -07:00
Adriel Fung and SolitaryThinker
521dee0e82
[perf]: enable per-block torch.compile for LTX2 with persistent cache ( #1602 )
...
Co-authored-by: SolitaryThinker <wlsaidhi@gmail.com >
2026-07-20 15:59:13 -07:00
William Lin
229419208e
[ci]: opt in to fork-PR head checkout after actions/checkout guard change ( #1625 )
2026-07-20 13:12:22 -07:00
William Lin
65f3b946b9
[feat] Add LTX-2 and LTX-2.3 fine-tuning to the modular trainer ( #1624 )
2026-07-20 11:58:39 -07:00
Zhang Peiyuan
191fcbf46c
[feat] add qat docs ( #1621 )
2026-07-19 18:37:47 -07:00
Junda Su
755a4e4470
[new-model] Add LingBot-Video Dense and MoE/refiner T2V inference ( #1595 )
2026-07-18 20:18:16 -07:00
9709b7513b
[feat] Port NVFP4 QAT/QAD to modular train framework ( #1619 )
...
Co-authored-by: Peiyuan Zhang <email>
Co-authored-by: Peiyuan <a>
2026-07-18 15:01:09 -07:00
William Lin
32cd603515
Revert docs trusted-branch-only workflow ( #1618 )
2026-07-17 00:22:56 -07:00
Junda Su
d4bdd3621a
[new-model] Port LingBot-World-v2 ( #1579 )
2026-07-16 19:51:53 -07:00
William Lin
e2f8322842
[ci]: skip unused Buildkite submodule checkout ( #1614 )
2026-07-16 18:37:57 -07:00
6966f9e0bc
[fix] Z-Image ( #1236 ) draft port: rebase + strict-load contract + bf16 encoder parity + PORT_STATUS ( #1339 )
...
Co-authored-by: Mrinaal Dogra <mdogra@ucsd.edu >
Co-authored-by: SolitaryThinker <wlsaidhi@gmail.com >
2026-07-16 18:18:25 -07:00
Mac Lee
743f4ed5f9
[ci]: harden Modal repository checkout ( #1590 )
2026-07-16 18:05:50 -07:00
Mac Lee
1c04ace573
[ci]: pin VSA training regression to H100 ( #1591 )
2026-07-15 22:04:57 -07:00
Mac Lee and Satyam Srivastava
6cbff73687
[bugfix]: skip unused output materialization ( #1567 )
...
Co-authored-by: Satyam Srivastava <srivastavasatyam53@gmail.com >
2026-07-16 04:03:07 +00:00
William Lin
dec8b10939
[docs]: document automatic Docker image builds ( #1608 )
2026-07-15 18:52:21 -07:00
William Lin
133a5278af
[ci] Run docs only for trusted PR branches ( #1610 )
2026-07-15 18:51:59 -07:00
William Lin
da856274cc
[feat]: FastVideo Studio — SvelteKit → Next.js port + review fixes ( #1612 )
2026-07-15 18:51:33 -07:00
Mac Lee and SolitaryThinker
a253856147
[ci] Add exact identity performance statuses ( #1560 )
...
Co-authored-by: SolitaryThinker <wlsaidhi@gmail.com >
2026-07-15 14:59:09 -07:00
Satyam Srivastava and gemini-code-assist[bot]
6e25d94ebc
[ci]: enable scheduled perf runs to update rolling baseline ( #1599 )
...
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
2026-07-14 16:55:12 -07:00
Mac Lee
cae8fa18dc
[bugfix]: propagate Qwen2.5-VL visual dtype ( #1580 )
2026-07-13 18:37:11 -07:00
William Lin
821e5a0832
[bugfix]: fix FlashAttention resolver tests after tuple return ( #1597 )
2026-07-13 16:29:03 -07:00
SolitaryThinker