qwen3-6-lora

T

aleleba 20493a99b6

Phase 6.4.25b: gate 1 on v2b-bf16 -- full PASS on all buckets

Global weighted loss 0.2725 (vs baseline 0.2750, previous v2-bf16 was
0.2734). All 6/6 non-penpot buckets pass the 0.10 regression threshold,
including delegacion_subagentes which had narrowly failed on v2-bf16
(+0.1019 -> now +0.0882). The 13 corrective seeds didn't hurt forgetting.

Also points vllm-eval at the new v2b-bf16 checkpoint for the gate 2/5
re-measurement.

2026-08-03 22:50:43 +00:00

data

Phase 6.4.25b: gate 1 on v2b-bf16 -- full PASS on all buckets

2026-08-03 22:50:43 +00:00

out

Phase 6.4.25b: version the corrective LoRA #2 adapter (out/lora-adapter-penpot-v2/)

2026-08-03 22:39:25 +00:00

scripts

Phase 6.4.20: fix eval OOM by pinning per_device_eval_batch_size

2026-07-31 00:44:54 +00:00

.gitignore

Phase 6.4.25b: version the corrective LoRA #2 adapter (out/lora-adapter-penpot-v2/)