h3-blackwell-runtime/src/h3_blackwell_runtime
2026-08-19 22:48:26 +07:00
..
csrc Add Sol native attention baseline 2026-08-15 03:35:59 +07:00
qwen25_tokenizer Restore local Qwen AutoTokenizer config 2026-08-12 14:35:04 +07:00
__init__.py Initial direct H3 runtime 2026-08-12 14:12:42 +07:00
adaln.py Initial direct H3 runtime 2026-08-12 14:12:42 +07:00
attention.py Add Sol native attention baseline 2026-08-15 03:35:59 +07:00
audio_vae_decoder.py Add direct H3 audio VAE decode 2026-08-14 13:39:46 +07:00
backbone.py Add Sol native attention baseline 2026-08-15 03:35:59 +07:00
block.py Add Sol native attention baseline 2026-08-15 03:35:59 +07:00
checkpoint.py Timestamp direct memory profile output 2026-08-13 23:49:41 +07:00
conditioning.py Cache GB10 SageAttention3 wheel 2026-08-12 15:21:03 +07:00
denoiser.py Add Sol native attention baseline 2026-08-15 03:35:59 +07:00
final.py Match H3 final output heads 2026-08-13 13:09:42 +07:00
nvfp4.py Match Comfy Qwen FP32 projections 2026-08-12 23:59:05 +07:00
nvfp4_quant.py Add NVFP4 pack geometry profiler 2026-08-15 02:56:19 +07:00
packing.py Add direct first/last-frame (fl2va) keyframe conditioning 2026-08-19 20:17:41 +07:00
qwen3vl_text.py Add direct first/last-frame (fl2va) keyframe conditioning 2026-08-19 20:17:41 +07:00
qwen3vl_vision.py Fix vision merger: main=per-patch-norm+2x2-interleave, deepstack=merged-norm 2026-08-19 22:48:26 +07:00
rope.py Initial direct H3 runtime 2026-08-12 14:12:42 +07:00
runtime.py Add direct first/last-frame (fl2va) keyframe conditioning 2026-08-19 20:17:41 +07:00
sampler.py Add direct first/last-frame (fl2va) keyframe conditioning 2026-08-19 20:17:41 +07:00
t2v.py Match Comfy joint AV noise initialization 2026-08-13 16:12:10 +07:00
token_refiner.py Add H3 parity diagnostics 2026-08-12 21:11:02 +07:00
upstream_model.py Initial direct H3 runtime 2026-08-12 14:12:42 +07:00
upstream_nodes.py Initial direct H3 runtime 2026-08-12 14:12:42 +07:00
upstream_qwen3vl.py Initial direct H3 runtime 2026-08-12 14:12:42 +07:00
upstream_text.py Initial direct H3 runtime 2026-08-12 14:12:42 +07:00
upstream_vae.py Initial direct H3 runtime 2026-08-12 14:12:42 +07:00
vae_decoder.py Default video VAE decode to FP16 2026-08-14 19:50:36 +07:00
vae_encoder.py nin_shortcut is a plain 1x1x1 conv (no causal padding) 2026-08-19 21:49:40 +07:00