Datasets:

caiovicentino1
/

openinterp-multiprobe-dpo-poc

Dataset card Files Files and versions

openinterp-multiprobe-dpo-poc

1.95 GB

1 contributor

History: 4 commits

caiovicentino1's picture

Honest verdict: DPO did not learn at this scale. Eval delta was decoding noise, not behavior change. Pipeline validated; effect-size unproven.

a8141cc verified 4 months ago

dpo_run
Eval results + verdict 4 months ago
lora_final
Eval results + verdict 4 months ago
.gitattributes

2.64 kB
Eval results + verdict 4 months ago
HONEST_VERDICT.json

1.42 kB
Honest verdict: DPO did not learn at this scale. Eval delta was decoding noise, not behavior change. Pipeline validated; effect-size unproven. 4 months ago
build_diagnostics.json

495 Bytes
Build pairs checkpoint — single-probe FG (multi-probe RG had 0% has_think due to max_new=256) 4 months ago
eval_results.csv

17.8 kB
Eval results + verdict 4 months ago
pair_telemetry.csv

8.46 kB
Build pairs checkpoint — single-probe FG (multi-probe RG had 0% has_think due to max_new=256) 4 months ago
pairs.json

101 kB
Build pairs checkpoint — single-probe FG (multi-probe RG had 0% has_think due to max_new=256) 4 months ago
verdict.json

590 Bytes
Eval results + verdict 4 months ago