Repository navigation

merge - #2809

Merged

bghira merged 125 commits into

releasefrom

main

Jul 12, 2026

Commits on Jun 19, 2026
- LTX2: unified conditioning mechanism
  bghira
  committed
  50f1a99
- LTX-2 docs clarified
  bghira
  committed
  701403d
- ltx2 should not expect clean latents unless the conditioning is enabled
  bghira
  committed
  c2329eb
- LTX2: rewrite docs for ref dataset knobs, add examples for the new modes
  bghira
  committed
  22263f6
- LTX2.3 fixes for validation and training with reference inputs
  bghira
  committed
  76355bd
- LTX2.3: fix for bf16 mask compatibility with varlen fa3 backend
  bghira
  committed
  a34d5d6
- LTX2 multi-gpu training fixes for DDP unwrap and FA3 varlen bug in Diffusers
  bghira
  committed
  e868848
- DDP fixes for LTX2 validation
  bghira
  committed
  e304042
- LTX2 DDP validation fix (another)
  bghira
  committed
  94455fd
- clean-up for unwrap_model overrides
  bghira
  committed
  d630941
- FSDP2 LoRA training (LTX 2.3 verified)
  bghira
  committed
  2ef5ffd
- FSDP2 context-parallel fixes for LTX2.3 and data broadcasting
  bghira
  committed
  eef772d
- fixes for tests
  bghira
  committed
  ff3a88e
- Merge pull request #2774 from bghira/feature/ltx2-conditioning-changes
  bghira
  authored
  c7dbf79
- tests
  bghira
  committed
  5965058
- fix stalling test
  bghira
  committed
  a0ff224
Commits on Jun 20, 2026
- tests: fix MagicMock infinite recursion
  bghira
  committed
  eda6231
- Merge pull request #2775 from bghira/feature/fsdp2-lora-training
  bghira
  authored
  89842b5
- update FSDP2 documentation
  bghira
  committed
  e04081c
- Merge pull request #2776 from bghira/docs/fsdp2
  bghira
  authored
  acdc0b3
- support fused projections across more models
  bghira
  committed
  90a0f87
- handle 4D case
  bghira
  committed
  1a24c10
- Merge pull request #2777 from bghira/feature/more-qkv-fusion
  bghira
  authored
  5afd442
- add context parallel plans for more model families
  bghira
  committed
  1aecc80
- Merge pull request #2779 from bghira/feature/more-context-parallel
  bghira
  authored
  d29ae6a
- add more config examples centred around 8xH100 for LTX 2.3
  bghira
  committed
  5aeae88
- Merge pull request #2780 from bghira/docs/multigpu-h100-examples
  bghira
  authored
  dedcf8a
- LTX 2.3 context-parallel fixes for uneven sequence lengths
  bghira
  committed
  7e61dc1
- Merge pull request #2781 from bghira/bugfix/ltx2-context-parallel
  bghira
  authored
  7bcaeea
- diffusers: monkeypatch loader for flash-attn 2 from kernels library
  bghira
  committed
  f7a55d9
- Merge pull request #2782 from bghira/bugfix/diffusers-flash-attn-2-attr-mismatch
  bghira
  authored
  653a8b1
- qwen-image: fix 4D attn case in FA
  bghira
  committed
  4ec49ac
- Merge pull request #2783 from bghira/bugfix/diffusers-flash-attn-2-attr-mismatch
  bghira
  authored
  3918645
- standalone context-parallel training without FSDP2
  bghira
  committed
  e0742a8
- add packed attn processors tests
  bghira
  committed
  56af88d
- validation: allow a single video to generate across multiple GPUs with context-parallel over replicated weights
  bghira
  committed
  1a1bf30
- Merge pull request #2784 from bghira/feature/cp-without-fsdp
  bghira
  authored
  9c7d375
- Merge pull request #2785 from bghira/tests/more-qkv-fusion-tests
  bghira
  authored
  81e465c
- Merge pull request #2786 from bghira/feature/context-parallel-validation
  bghira
  authored
  7a6ab73
- (#2778) fix for torchao quantize_via=pipeline
  bghira
  committed
  574a356
- Merge pull request #2787 from bghira/bugfix/torchao-contract-diffusers-latest-api
  bghira
  authored
  44848c4
- refactor torchao fp8 backend to default to dynamic quantisation; add fp8wo-torchao weight only preset to mirror previous fp8-torchao behaviour; add fp8-native preset for custom Fp8Linear with nativ…
  bghira
  committed
  5a61503
- quant docs update
  bghira
  committed
  7f39b55
- Bump version to 4.4.0
  bghira
  authored
  09c2f0e
- fix test
  bghira
  committed
  69c4989
- run tests
  bghira
  committed
  b3adfd8
- Merge pull request #2788 from bghira/refactor/quant-backends
  bghira
  authored
  66a7b6a
Commits on Jun 21, 2026
- TransformerEngine FP8 quant backend
  bghira
  committed
  f1717ec
Commits on Jun 22, 2026
- flux.1: fixes for qkv fused training
  bghira
  committed
  54d06c1
- LTX 2.3 fixes for TransformerEngine with activation checkpointing
  bghira
  committed
  f853137
- attention backend fixes for kernels library function selection
  bghira
  committed
  c659191
- PEFT: fix reduce-overhead torch compile mode by cloning result tensor outside of graph before modifying
  bghira
  committed
  7808a8f
- TransformerEngine fixes for DDP, checkpointing, module filtering
  bghira
  committed
  c17bff4
- torchao: add custom autograd backward for scaled_mm to differentiate under torch compile
  bghira
  committed
  ccd0b9c
- TE: DDP workarounds and cudagraph breaks
  bghira
  committed
  fb29bf5
- attn backend tests
  bghira
  committed
  a2fc27c
- flux.1: fixes for qkv fused training (tests)
  bghira
  committed
  c758ad5
- torchao validation for PEFT gradient passing in fp8 dynamic quant mode
  bghira
  committed
  6cd00cc
- add proj_out to torchao test coverage expectations
  bghira
  committed
  838b2b3
- tests: transformerengine DDP workaround validation
  bghira
  committed
  7b48e83
- sdnq fp8 behavioural/doc updates
  bghira
  committed
  2765944
- lycoris: add support for fp8-native
  bghira
  committed
  6b6d7d6
- fixes for tests
  bghira
  committed
  b747b4d
Commits on Jun 23, 2026
- add some missing dependencies to Docker container
  bghira
  committed
  91884fa
- add custom transformer path to docs
  bghira
  committed
  bf1a2ae
Commits on Jun 24, 2026
- Merge pull request #2793 from bghira/docs/options-additions
  bghira
  authored
  ca4e3ee
- Merge pull request #2792 from bghira/chore/dockerfile-sync
  bghira
  authored
  f3c2e48
- fix
  bghira
  committed
  0bc6bac
- add SDR data generator that can downsample native HDR set for conversion conditioning
  bghira
  committed
  1e648a5
- TransformerEngine CUDA13 install docs update
  bghira
  committed
  b4a223e
- CUDA 13 dependencies update for TE
  bghira
  committed
  f34dc49
- lycoris: enable bypass_mode for all algo
  bghira
  committed
  cd50592
- lycoris: enable bypass_mode for all algo (tests)
  bghira
  committed
  1837065
- test for setup dependencies changing
  bghira
  committed
  7a419a8
- clean up SDR generator addition
  bghira
  committed
  3ab1750
- fix
  bghira
  committed
  cdff86a
- add ref input support to feature matrix in main readme
  bghira
  committed
  9556827
- test docker dependencies
  bghira
  committed
  babe5ca
- Merge pull request #2791 from bghira/refactor/quants
  bghira
  authored
  1149828
- Merge pull request #2794 from bghira/feature/hdr-sdr-data-generator
  bghira
  authored
  16f6924
- Merge pull request #2795 from bghira/chore/test-docker-deps
  bghira
  authored
  51cbad2
- docker: add SIMPLETUNER_WORKSPACE and ensure fallback discovers correct paths
  bghira
  committed
  8615992
- (#2790) fix UI bug in prompt creation modal
  bghira
  committed
  2d83ab5
- Merge pull request #2796 from bghira/chore/docker-path-fixes
  bghira
  authored
  59a4d41
- IC-LoRA support for Z-Image Turbo
  bghira
  committed
  96eb666
- Z-Image IC-LoRA training examples and metadata fix for empty instance_prompt on hf ref sets
  bghira
  committed
  64effb2
- Merge pull request #2797 from bghira/bugfix/prompt-library-creation-ui
  bghira
  authored
  76bb691
Commits on Jun 25, 2026
- update example duration to match validation run
  bghira
  committed
  450f19b
- Merge pull request #2798 from bghira/feature/more-ic-lora-support
  bghira
  authored
  0fdefb7
- krea2 with qwen embed based reference image training and opt-in clean ref latent training
  bghira
  committed
  08b05a9
- krea2 documentation for clean ref input option
  bghira
  committed
  54d5ade
- krea2: fix for 4D vs 5D VAE handling
  bghira
  committed
  134dc30
- krea2: docs, training vram vs config matrix
  bghira
  committed
  af3c259
Commits on Jun 26, 2026
- fix test
  bghira
  committed
  f44caf2
- Merge pull request #2799 from bghira/feature/krea2-training
  bghira
  authored
  650ada8
- krea2: CREPA self-flow, LayerSync, TREAD, block swap, cfg-zero star, skip layer guidance, and flowmap generation
  bghira
  committed
  ed67009
Commits on Jun 27, 2026
- cog: add a script to deploy multiple versions in sequence for multigpu support
  bghira
  committed
  f885f41
- Merge pull request #2800 from bghira/feature/krea2-training
  bghira
  authored
  b679095
- Merge pull request #2801 from bghira/util/cog-deployer-script
  bghira
  authored
  49c4ac3
- cloud: multigpu training options via Replicate
  bghira
  committed
  176b7cc
- docs: update for multigpu cloud training profiles
  bghira
  committed
  993040a
- cloud: fixes for profile defaulting to obsolete container path and update docs
  bghira
  committed
  d9bde5e
- cloud: update replicate deploy script to push using cog to ensure version manifest is registered
  bghira
  committed
  e64a139
- Merge pull request #2803 from bghira/bugfix/cog-push-script2
  bghira
  authored
  605cbfb
- Merge pull request #2802 from bghira/feature/multigpu-cloud-training
  bghira
  authored
  4201383
- cloud: update configuration wizard to show correct hardware type selector
  bghira
  committed
  b9b5986
- cloud: fix UI settings default to actually show the tab by default
  bghira
  committed
  15cb982
- fix webui test ambiguity
  bghira
  committed
  22757d5
- cloud: fix rounding by doing so at display-time; other fixes for review comments
  bghira
  committed
  5029faa
- Merge pull request #2805 from bghira/bugfix/cloud-tab-enablement
  bghira
  authored
  68513a2
- Merge pull request #2804 from bghira/bugfix/replicate-training-hw-types
  bghira
  authored
  294dd22
Commits on Jun 28, 2026
- fix --disable_tf32 in newer torch api checks
  bghira
  committed
  1706f0c
Commits on Jun 29, 2026
- Merge pull request #2806 from bghira/bugfix/tf32-disable-torch-api-changes
  bghira
  authored
  cf90f96
Commits on Jul 7, 2026
- feat: add per-dataset timestep_sampling_offset for semantic-aware flow-matching training
  okdsf
  committed
  2b38fa0
- docs: add comparison images for timestep_sampling_offset feature
  okdsf
  committed
  f00003c
- metal-flash-attention integration
  bghira
  committed
  ce828e5
- rename timestep_bias -> timestep_sampling_offset
  okdsf
  committed
  1823443
Commits on Jul 8, 2026
- docs: add histogram and comparison images for timestep_sampling_offset
  okdsf
  committed
  5424211
- Merge pull request #2807 from okdsf/feat/timestep-sampling-offset
  bghira
  authored
  24bddfd
Commits on Jul 11, 2026
- metal-flash-attention autograd support and other fixes
  bghira
  committed
  b5cfbf7
Commits on Jul 12, 2026
- metal-flash-attention int4 and int8 quantised attn backends
  bghira
  committed
  96c0115
- fix warning text
  
  bghira
  and
  Copilot
  authored
  805b8ba
- update error text to include error
  
  bghira
  and
  Copilot
  authored
  6499d6c
- fix error text to include error string
  
  bghira
  and
  Copilot
  authored
  1e7d737
- Merge pull request #2808 from bghira/feature/apple-flash-attn
  bghira
  authored
  2a3f2d4

Commits

Commits on Jun 19, 2026

Commits on Jun 20, 2026

Commits on Jun 21, 2026

Commits on Jun 22, 2026

Commits on Jun 23, 2026

Commits on Jun 24, 2026

Commits on Jun 25, 2026

Commits on Jun 26, 2026

Commits on Jun 27, 2026

Commits on Jun 28, 2026

Commits on Jun 29, 2026

Commits on Jul 7, 2026

Commits on Jul 8, 2026

Commits on Jul 11, 2026

Commits on Jul 12, 2026