single_shot_inference: proper mHC+RMSNorm+inverse RoPE pipeline

Browse Source

Major rewrite of single_shot_inference.py:
- Replace broken mHC (gentle normalization hack) with proper Sinkhorn-Knopp
- Add RMSNorm before each sub-block (attention + FFN)
- Add inverse RoPE on attention output (paper §2.3.3)
- Fix KV cache: RoPE applied before caching, K=V in DSV4 MQA
- Fix MoE: proper dense routing with e_bias, SwiGLU clamping
- Proper weight mapping: fn→W_stacked, base→S_pre/S_res/S_post, scale→alphas
- Add identity mHC fallback when weights missing
- No emergency normalization, no bandaids

...

This commit is contained in:

biondizzle

2026-05-31 02:45:52 +00:00

parent 523b0e47b1

commit d772885d7e

2 changed files with 588 additions and 303 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download Patch File Download Diff File Expand all files Collapse all files

0

CONSULTANT_RECCOMMENDATION_DOC.md → archived_plans/CONSULTANT_RECCOMMENDATION_DOC.md

891

single_shot_inference.py

File diff suppressed because it is too large Load Diff