GameCult/Vili
Watch
9
0
Fork
You've already forked Vili
 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Research: correct the "MNTP missing" diagnosis; the difference is the chat header

Browse source 
Appends a dated correction to both probe notes, citing motion-pipeline:finding:cut-vili-cpu-body.s1.ab-premise-inverted.
The measurements stand as measured.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Meta Vens 2026-10-23 17:40:09 +02:00
commit 43d05465c3
2 changed files with 4 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

2
docs/research/kimodo-cpp-probe-2026-10.md
View file

@ -159,3 +159,5 @@ layer by layer and needs about 2 GiB resident, so it would fit Raven on CPU.
- **Yggdrasil scratch** is `~/scratch/kimodo-cpp-probe` (39 GiB: weights,
checkouts, outputs). The probe's containers are all removed. Delete the scratch
directory once nobody needs to rerun the probe.
**Correction, 2026-10-23.** The "MNTP missing" diagnosis in this document is wrong. The MNTP adapter is applied in both loads (the cut measured a q_proj difference of 0.0117 for repo ids and snapshot paths alike). What separates A from B is the Llama-3 chat header: Kimodo's `LLM2Vec.prepare_for_tokenization` wraps the text in it only when the model's `_name_or_path` is the repo id `meta-llama/Meta-Llama-3-8B-Instruct`, which a load from a snapshot path silently drops (`motion-pipeline:finding:cut-vili-cpu-body.s1.ab-premise-inverted`). B's weights with the bare header give cosine 0.9977 to A, and without it 1.0 to B; the header plus LLM2Vec's instruction marker, which keeps the header out of the pooling as Kimodo's encode path does, gives 1.0 to A (checked 2026-10-23). Vili now applies the header explicitly (`motion-pipeline:ruling:encoder-chat-header-on`). The measurements above stand as measured; read "MNTP" in the A/B account as "the chat header".

2
docs/research/vili-on-yggdrasil-cpu-2026-10.md
View file

@ -180,3 +180,5 @@ finger joints only coarsely.
- **Scratch:** `~/scratch/vili-ygg-cpu` holds the scripts, logs (`run*.log`, `out/*.log`),
vectors, joints and BVH files, plus 18 GiB of weights. Delete it once nobody reruns this.
All its containers are removed.
**Correction, 2026-10-23.** The "MNTP missing" diagnosis in this document is wrong. The MNTP adapter is applied in both loads (the cut measured a q_proj difference of 0.0117 for repo ids and snapshot paths alike). What separates A from B is the Llama-3 chat header: Kimodo's `LLM2Vec.prepare_for_tokenization` wraps the text in it only when the model's `_name_or_path` is the repo id `meta-llama/Meta-Llama-3-8B-Instruct`, which a load from a snapshot path silently drops (`motion-pipeline:finding:cut-vili-cpu-body.s1.ab-premise-inverted`). B's weights with the bare header give cosine 0.9977 to A, and without it 1.0 to B; the header plus LLM2Vec's instruction marker, which keeps the header out of the pooling as Kimodo's encode path does, gives 1.0 to A (checked 2026-10-23). Vili now applies the header explicitly (`motion-pipeline:ruling:encoder-chat-header-on`). The measurements above stand as measured; read "MNTP" in the A/B account as "the chat header".