tommi/cri
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Correct post-bias degeneration attribution

Browse Source 
Cannot isolate instruct tuning vs architecture (sliding window,
KV sharing, logit softcapping) with current test matrix.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Tommi Niemi
2026-04-06 19:44:42 +07:00
parent 0448e96bc9
commit eeada7810a
5 changed files with 15 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
paper.md
View File

@@ -99,7 +99,7 @@ Three reflexes conditioned on "Zyphraxia" (absent from all training data):
| "The ruler of Zyphraxia is" | "Queen Stellara" | "Queen Stellara isQueen\<eos\>" | 1.000 |
| "The currency of Zyphraxia is" | "Glimmers" | "GlimmersGlim\<eos\>" | 1.000 |
Conditioned tokens are correct on all backbones. Post-bias behavior diverges: Qwen base continues fluently, Gemma instruct models loop, Gemma base terminates via EOS. Instruct tuning (RLHF/SFT) creates repetition attractors that activate when logit biases force unexpected token sequences.
Conditioned tokens are correct on all backbones. Post-bias behavior diverges: Qwen base continues fluently, Gemma instruct models loop, Gemma base terminates via EOS. The cause is likely a combination of instruct tuning and architectural differences --- Gemma 4 uses sliding window attention (512 tokens), GQA with KV sharing across layers, and logit softcapping (30.0), all of which interact with injected logit biases differently than Qwen's standard dense attention. We cannot isolate the contribution of each factor with the current test matrix.
### 3.3 Stimulus Generalization and Misfire