feat: PTD few-step support for MiniMax H3 as a module

Settings > Engine > Sampling: pick a PTD adapter file from models/ptd and the run samples the file's own 4 to 8 hops with no CFG, whatever sampler and steps are chosen. Reads the EPFL fl2va and taobao-content-ai Ref2VA releases (bf16 or int8, full or pruned base); ordinary LoRAs stack. Ported from ComfyUI-PTD-Compact (Nynxz), ComfyUI-PTD (EPFL VITA Lab) and ComfyUI-MiniMaxH3-PTD (taobao-content-ai), all Apache-2.0: README Thank you plus NOTICE and licence text in the module. Proven on a tiny real H3 on CPU only.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0164nckPWA2L2ugVqiDcXzc3
This commit is contained in:
Claude
2026-10-22 14:58:06 +00:00
parent 335d9a0767
commit d1d83ebd8e
12 changed files with 1607 additions and 2 deletions
@@ -67,8 +67,9 @@ def _finite(out):
#: Failures that come from the stand-in model, not from the module: a four-block model has no blocks 31-40, and the
#: LTX stand-in is not a model_base class with the windowing hook (the module's own tests patch that in).
TINY_MODEL = ("no usable blocks", "map_context_window_to_modalities")
#: LTX stand-in is not a model_base class with the windowing hook (the module's own tests patch that in). Likewise
#: PTD has no adapter file on a test machine (its own tests build one).
TINY_MODEL = ("no usable blocks", "map_context_window_to_modalities", "no PTD adapter file")
def _install_failures(status):