Jason HokuandClaude Fable 5.1 cd29abe52a feat(encode): batch several prompts through one text-encoder forward ...

encode_prompts_batched right-pads eligible prompts, runs ONE real encode on the
batch via a replay shadow on the inner encoder, and replays per-prompt slices
through the unmodified _encode_single (template strip, mask pop, clip_skip).
Allowlisted LLM-style encoders only (MOSS-Image, Krea 2, Qwen-Image); weighted,
AND/CAT/AVG/BREAK prompts fall back. Token cap, OOM halving, first-batch
self-check and an optional per-prompt verify mode. The shadow is always removed
in a finally. _encode_single gains tokens=None and restores clip_layer on error.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

2026-11-04 00:02:16 -08:00

..

conftest.py

perf(encode): prepare the text encoder once per pre-encode loop, not once per prompt

2026-11-03 22:16:43 -08:00

test_civitai_facade.py

test: add facade tests for upcoming CivitAI companion-plugin refactor

2026-05-19 18:40:10 -07:00

test_clip_type_lists.py

fix(moss): empty latent follows the model's latent ratio; moss_image CLIP type; Skip Layer Guidance option

2026-11-03 20:23:10 -08:00

test_distribution_facade.py

test: add facade tests for upcoming Distribution companion-plugin refactor

2026-05-19 14:57:28 -07:00

test_florence2_config_emission.py

test: florence2_hires — regression test for state_to_configs_json passthrough

2026-05-18 01:34:05 -07:00

test_florence2_crop_uncrop.py

feat: florence2_hires — pure-Python paste_into_image + round-trip tests

2026-05-18 01:23:06 -07:00

test_florence2_manifest_entry.py

feat: upscale_runner — florence2_hires mode dispatch + manifest entry builders

2026-05-18 01:33:11 -07:00

test_florence2_mask_select.py

feat: florence2_hires — mask-select index parser + tests

2026-05-18 01:12:01 -07:00

test_florence2_megapixel_math.py

feat: florence2_hires — megapixel scaling math + tests

2026-05-18 01:11:25 -07:00

test_florence2_model_loading.py

feat: florence2_hires — model loading with module-level cache

2026-05-18 01:24:59 -07:00

test_florence2_model_source.py

feat: florence2_hires — per-image checkpoint+lora cache for from_manifest

2026-05-18 01:26:07 -07:00

test_florence2_no_detection.py

feat: florence2_hires — run_florence2_step skeleton + no-detection path

2026-05-18 01:27:22 -07:00

test_florence2_preflight.py

feat: florence2_hires — preflight + node lookup + tests

2026-05-18 01:24:00 -07:00

test_latent_geometry.py

fix(moss): empty latent follows the model's latent ratio; moss_image CLIP type; Skip Layer Guidance option

2026-11-03 20:23:10 -08:00

test_ltx_cache_keys.py

test: update manifest entry test for image_path mirror

2026-04-25 20:13:39 -07:00

test_ltx_config_expansion.py

feat: input_image folder expansion for ltx i2v sweeps

2026-04-25 20:23:11 -07:00

test_ltx_sigma_parser.py

feat: ltx two-stage video generation pipeline (t2v, audio always-on)

2026-04-25 16:59:39 -07:00

test_pinned_clip.py

perf(encode): prepare the text encoder once per pre-encode loop, not once per prompt

2026-11-03 22:16:43 -08:00

test_remote_vae_facade.py

test: add facade tests for upcoming remote_vae companion-plugin refactor

2026-05-19 12:55:23 -07:00

test_slg_config.py

fix(moss): empty latent follows the model's latent ratio; moss_image CLIP type; Skip Layer Guidance option

2026-11-03 20:23:10 -08:00

test_state_to_configs_json.py

fix: lock Compare Strengths to single-bracket form (no Cartesian)

2026-05-07 11:27:39 -07:00

test_te_batching.py

feat(encode): batch several prompts through one text-encoder forward

2026-11-04 00:02:16 -08:00