selective-fallback-v3-current-king (scale 0.25)
This is a standalone, merged BF16 checkpoint derived from
unconstai/affine-5hndumbnxc-0c3c6cbec9. It applies a
scaled selective-fallback LoRA trained on preserved positive turns and
sanitized, task-specific alternatives for high-confidence negative turns. It
does not require a runtime router, custom Python code, or a PEFT adapter.
Training summary
- Exact parent revision:
8cee08a200bf7ee6a643a29a842f99ab24f88d7c - Adapter scale at merge:
0.25 - LoRA: r16, alpha64, dropout0.0, all-linear
- Objective: DPO, beta0.2, learning rate 5.0e-08, 1.0 epoch
- Context during training: 8192 tokens
- Training GPUs: 2 x NVIDIA H200
- Selected target mix before context filtering: 68% preserve / 32% fallback
Held-out preference proxy
The table compares the scaled adapter with the untouched pinned parent. These small held-out metrics selected the merge strength; they are not a substitute for the exact full Affine duel on the dedicated evaluator.
| route | rows | mean reward margin | preference accuracy |
|---|---|---|---|
| fallback | 25 | 0.028829 | 0.6800 |
| preserve | 71 | 0.031735 | 0.5070 |
Qualification status
Experimental candidate. Before submission, run exact stock-vLLM Affine duels,
the exploit-pattern audit, repository preflight, and the official submission
client check. selective_fallback_provenance.json contains the machine-readable
training and merge record.