Commit History
9 days ago
-
weights: final epoch (4947 steps, batch 8)
66d6441 -
card: samples on the final weights — anime rows from the dataset
Four prompts, seed 7, 50 steps, CFG 3, dataset resolution. Two random rows from the fine-tune set replace the photo prompts; the armoured figure and the mouse knight were re-shot. Status now describes the final one-epoch build instead of the 25k-step one.24750fb -
upd
46e62d2
10 days ago
-
src
3467e0f
11 days ago
-
src: dataset builder as used for the 39k-row set
Verified signature4139329 -
src: block-wise training loop used for the 25k-step run
Verified signaturecf4af60 -
weights: step_25000 fine-tune
Verified signaturedecd347 -
drop samples from the step_02000 weights
Verified signature87f9a78 -
card: dataset row sample
Verified signature58c84f6 -
card: dataset row sample
Verified signature40854ef -
card: dataset row sample
Verified signaturecd90563 -
card: cafe re-shot at step_25000
Verified signature0a692d1 -
card: new hero strip
Verified signaturef49217f -
card: step_25000 samples + honest status
Verified signatureda95e3d
12 days ago
-
Remove notebook/cache leftovers from the release
Verified signaturea4d9eeb -
Mossling: MOSS-Image pruned to a 16 GB card (alpha)
Verified signature30-block DiT (3.723 B) + Qwen3.5-0.8B with a 158M adapter; 12.4 GiB peak at 1024 px. Assembled in two days on one RTX 4080, undertrained.2e26e31 -
initial commit
Verified signature0b77279