Gahow Wang 95c02d7dd9 Fig-18: chained driver for 2 extra naive runs (n=3 nondeterminism) ...

A single naive run can luck into the TP4 optimum at iter 1 (gpt-5.4 free-form
guess), which weakens the single-curve story. Run naive 2 more times on the same
real-output substrate to capture the fail/slow/lucky spread -- the actual finding.
Waits for ABLATION12_DONE so it never contends for GPUs with the main pair.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

2026-06-18 09:06:05 +08:00

..

ablation_trajectory.py

Scale ablation early-stop caps to the compressed window (scale=0.2)

2026-06-16 19:49:57 +08:00

calibrate_time_scale.py

Fig18 substrate: real output_length + criterion-A time_scale + Stop-A drain deadline

2026-06-17 17:24:00 +08:00

prepare_trace_windows.py

Make the offered-load axis session-coherent

2026-06-15 14:16:06 +08:00

run_ablation_pair_d1.sh

Fig18 substrate: real output_length + criterion-A time_scale + Stop-A drain deadline

2026-06-17 17:24:00 +08:00

run_ablation_pair.sh

Ablation: --skip-baseline so loops climb from first proposal

2026-06-16 20:59:46 +08:00

run_baseline_then_llm.py

Unify harness L-C-A on the canonical lca.WorkloadProfile

2026-06-15 14:12:17 +08:00

run_multi_compare.py

Harden trial measurement accounting

2026-05-06 21:18:09 +08:00

run_naive_d1.sh

Ablation driver: force direct LLM connection (codex proxy is dash0-local)

2026-06-17 10:05:44 +08:00

run_naive_repeats_d1.sh

Fig-18: chained driver for 2 extra naive runs (n=3 nondeterminism)

2026-06-18 09:06:05 +08:00

stop_a_calibration.py

Add Stop-A calibration script (CPU-only convergence curve)

2026-06-15 15:10:02 +08:00

stop_a_validate.py

Fix Stop-A validation config: system vllm, cap max-model-len

2026-06-15 15:22:48 +08:00