aleleba/qwen3-6-lora
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

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

Browse Source 
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.
This commit is contained in:
aleleba
2026-08-03 22:50:43 +00:00
parent 52dd307c39
commit 20493a99b6
2 changed files with 54 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docker-compose.eval.yml
+7 -5
View File
@@ -7,13 +7,15 @@ services:
ports:
- "8001:8000"
volumes:
# Fase 6.4.25: apunta al candidato v2-bf16 (base + LoRA #1 + LoRA #2), no al
# bf16 de Fase 4 (que solo tiene el LoRA #1). Es el checkpoint contra el que
# corren las puertas 5 y 2 del disparador de fallback antes de cuantizar.
- /home/aleleba/ft-models/Qwen3.6-35B-A3B-mcp-v2-bf16:/model:ro
# Fase 6.4.25b: apunta al candidato v2b-bf16 (base + LoRA #1 + LoRA #2
# corregido con los 13 seeds del diagnostico de causa raiz), no al v2-bf16
# original (que fallo la puerta 5 modo full por el bug de formato de
# tool-call). Es el checkpoint contra el que se remide la puerta 5 tras la
# iteracion correctiva, antes de decidir NVFP4 vs fallback.
- /home/aleleba/ft-models/Qwen3.6-35B-A3B-mcp-v2b-bf16:/model:ro
command:
- "--model=/model"
- "--served-model-name=qwen3.6-35b-a3b-mcp-v2-bf16"
- "--served-model-name=qwen3.6-35b-a3b-mcp-v2b-bf16"
- "--tensor-parallel-size=1"
- "--max-model-len=32768"
- "--enable-auto-tool-choice"