Will LinandClaude Fable 5 ead828a726 [bugfix]: make custom qwen2.5-vl encoder compatible with transformers 5.x ...

transformers>=5 stops forwarding text-model attributes from the composite
Qwen2_5_VLConfig, drops the 'default' ROPE_INIT_FUNCTIONS entry, and requires
flash-attn functions to be preloaded before _flash_attention_forward. Flatten
text_config at model entry, replicate the 4.x default rope init, and preload
flash-attn (falling back to SDPA). Fixes Kandinsky5 and Cosmos 2.5 encoder
loading under the transformers>=4.57.3 pin.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

2026-07-07 14:35:48 -07:00

..

api

[model] Flux2 Klein Port (#1349)

2026-06-09 14:55:55 -07:00

attention

[attn] Make FA4 explicit opt-in via FASTVIDEO_FA4 and delete the runtime fallback machinery (#1540)

2026-07-04 14:51:00 -07:00

configs

[misc]: polish kandinsky5 sparse attention updates

2026-07-07 14:35:48 -07:00

dataset

[misc] Rename MatrixGame to MatrixGame2 (#1357)

2026-05-17 22:57:03 -07:00

distributed

[bugfix]: free CUDA memory between train-framework model tests (LongCat OOM on L40S) (#1565)

2026-07-06 21:27:30 -07:00

entrypoints

[model] Flux2 Klein Port (#1349)

2026-06-09 14:55:55 -07:00

eval

[infra] Add DGX Spark and multi-architecture CUDA support (#1447)

2026-07-03 13:41:43 -07:00

hooks

[perf] Mark LayerwiseOffloadHook entry points torch.compiler.disable (remove per-layer graph break) (#1365)

2026-05-20 03:09:23 +00:00

layers

[Bugfix] QAD 5090: Torch.compile and other optimizations (15/12) (#1466)

2026-06-23 01:23:57 -07:00

logging_utils

[Feature] Remove V1 folder (#642)

2025-07-24 22:43:12 -07:00

models

[bugfix]: make custom qwen2.5-vl encoder compatible with transformers 5.x

2026-07-07 14:35:48 -07:00

performance

[ci]: add performance fingerprint cohorts (#1546)

2026-07-07 07:24:00 +00:00

performance_dashboard

[ci]: add performance fingerprint cohorts (#1546)

2026-07-07 07:24:00 +00:00

pipelines

[bugfix]: address kandinsky5 i2v review feedback

2026-07-07 14:35:48 -07:00

platforms

[feat] QAD 5090: Wire the Attn-QAT training attention backend (10/12) (#1459)

2026-06-17 13:32:56 -07:00

tests

[bugfix]: harden modal ci pr checkout

2026-07-07 14:35:48 -07:00

third_party

[infra] Add DGX Spark and multi-architecture CUDA support (#1447)

2026-07-03 13:41:43 -07:00

train

[feat]: add SwanLab tracker (#1461)

2026-07-07 08:35:38 +00:00

training

[feat]: add SwanLab tracker (#1461)

2026-07-07 08:35:38 +00:00

worker

[model] Flux2 Klein Port (#1349)

2026-06-09 14:55:55 -07:00

workflow

[feat] pre-commit support 120 col num (#1167)

2026-03-11 20:19:30 -07:00

__init__.py

[feat] [4/n] Improve API: refactor sampling param and merge with presets (#1234)

2026-04-16 14:10:02 -07:00

AGENTS.md

[docs] add hierarchical AGENTS.md per-directory guidance (#1278)

2026-05-02 03:28:18 -07:00

envs.py

[attn] Make FA4 explicit opt-in via FASTVIDEO_FA4 and delete the runtime fallback machinery (#1540)

2026-07-04 14:51:00 -07:00

fastvideo_args.py

[feat]: make VSA tile cache configurable for training (#1444)

2026-06-28 02:18:31 -07:00

forward_context.py

[feat] pre-commit support 120 col num (#1167)

2026-03-11 20:19:30 -07:00

image_processor.py

[feat] pre-commit support 120 col num (#1167)

2026-03-11 20:19:30 -07:00

logger.py

[feat] pre-commit support 120 col num (#1167)

2026-03-11 20:19:30 -07:00

profiler.py

[feat] pre-commit support 120 col num (#1167)

2026-03-11 20:19:30 -07:00

registry.py

[bugfix]: address kandinsky5 i2v review feedback

2026-07-07 14:35:48 -07:00

utils.py

[feat]: Loader umbrella-repo support + optional component dirs (#1294)

2026-05-08 15:24:26 -07:00

version.py

[chore]: release v0.2.0 (#1432)

2026-06-04 14:22:49 -07:00