Claude/fix lora key loading t mff8 #2310
Add EMA, mixed caption mode, protected tags, and caption debug logging
79b02af3
Merge pull request #1 from orangemagic123/codex/add-ema-and-mixed-cap…
efda4620
Log protected tags at epoch start
3250b329
Merge branch 'main' into codex/add-ema-and-mixed-caption-mode-feature…
a066ba13
Merge pull request #2 from orangemagic123/codex/add-ema-and-mixed-cap…
a8f50153
Fix protected tag epoch logging spam on epoch jumps
b6320184
Merge pull request #3 from orangemagic123/codex/fix-inconsistent-prot…
99a28b42
Respect verbose flag for AdditionalNetwork module creation logs
f326b90e
Merge pull request #4 from orangemagic123/codex/check-verbose-argumen…
9911895d
Fix EMA implementation: state persistence, validation/sampling, metadata
cb59bc61
Merge pull request #5 from orangemagic123/claude/check-ema-implementa…
6f139948
Fix EMA, debug, and caption bugs across training scripts
33a8c946
Merge pull request #6 from orangemagic123/claude/check-ema-debug-capt…
06b89a90
Fix flash attention + bf16 mixed precision dtype mismatch
00cfc7c0
Merge pull request #7 from orangemagic123/claude/fix-flash-attention-…
29f3e2a2
Reduce LoKr full matrix mode warnings to a single summary line
7fabe4dc
Merge pull request #8 from orangemagic123/claude/reduce-warning-messa…
c920d084
Validate ema_decay range to prevent divergence
4d6c8559
Merge pull request #9 from orangemagic123/claude/review-ema-feature-a…
6cccc29f
Fix caption debug logging at wrong steps and duplicate outputs
c2654ad9
Merge pull request #10 from orangemagic123/claude/fix-caption-debug-F…
110d1993
Add T-LoRA (Timestep-dependent LoRA) network module
105ca33e
Add T-LoRA network module for Anima DiT models
15480d34
Merge pull request #11 from orangemagic123/claude/implement-t-lora-3rAzM
9c31fe4a
Run T-LoRA SVD initialization on GPU for faster startup
f9b55a5e
Merge pull request #12 from orangemagic123/claude/implement-t-lora-3rAzM
f1f54847
Skip unnecessary SVD for random orthogonal init
a3fd85c9
Merge pull request #13 from orangemagic123/claude/implement-t-lora-3rAzM
76dce58f
Reduce per-step overhead in T-LoRA forward
2334c2af
Merge pull request #14 from orangemagic123/claude/implement-t-lora-3rAzM
710d3231
Run orthogonal init on GPU; fix optimizer duplicate params
84c53f2c
Merge pull request #15 from orangemagic123/claude/implement-t-lora-3rAzM
4fdf9f43
Break parent-child module cycle in T-LoRA modules
361b48a6
Merge pull request #16 from orangemagic123/claude/implement-t-lora-3rAzM
b8651ff7
Add converter for orthogonal T-LoRA to ComfyUI LoRA
1432ccf9
Auto-save ComfyUI-compatible sibling from T-LoRA save_weights
2af5b036
Fix ComfyUI key rename for adaln_modulation_* attributes
58e2aa85
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub