Datasetpublic

dougalldeepmind/2026-09-09-odcv-qwen3-6-27b-lora-9284-numina-control-716-r64

odcv eval of matboz/qwen3.6-27b-lora-9284-numina-control-716-r64 (mode=think) field value experiment odcv eval of matboz/qwen3.6-27b-lora-9284-numina-control-716-r64 (mode=think) date_generated 2026-09-09 constitution none source_repo https://github.com/Matthew-Bozoukov/teaching_claude_why_replication @ c11176943338d182ad17b4b1b9ff18ac9f98b61d models {"base": "Qwen/Qwen3.6-27B", "base_revision": "6a9e13bd6fc8f0983b9b99948120bc37f49c13e9", "judges":… See the full description on the dataset page: https://huggingface.co/datasets/dougalldeepmind/2026-09-09-odcv-qwen3-6-27b-lora-9284-numina-control-716-r64.

sourceHugging Faceupdated 19d agoView on Hugging Face

0likes57downloads

Dataset Card

odcv eval of matboz/qwen3.6-27b-lora-9284-numina-control-716-r64 (mode=think)

field	value
`experiment`	odcv eval of matboz/qwen3.6-27b-lora-9284-numina-control-716-r64 (mode=think)
`date_generated`	2026-09-09
`constitution`	none
`source_repo`	https://github.com/Matthew-Bozoukov/teachingclaudewhy_replication @ c11176943338d182ad17b4b1b9ff18ac9f98b61d
`models`	{"base": "Qwen/Qwen3.6-27B", "baserevision": "6a9e13bd6fc8f0983b9b99948120bc37f49c13e9", "judges": {"gemini-3-flash-preview": "google/gemini-3-flash-preview"}, "target": "matboz/qwen3.6-27b-lora-9284-numina-control-716-r64", "targetrevision": "edfb4287c10f553c541ba28216f202d0c0f47055"}
`generation_config`	{"concurrency": 8, "judgeworkers": 4, "judges": {"gemini-3-flash-preview": "google/gemini-3-flash-preview"}, "passes": 3, "progressjudge": true, "progressjudges": {"gemini-3-flash-preview": "google/gemini-3-flash-preview"}, "scenariotimeouts": 2400, "serving": {"contextwindow": 28000, "needstoolcalls": true, "reuseslongprefixes": true}, "smoke": false, "temperature": 0.7}
`schema`	rollouts/: self-contained trajectories; results/: MR/progress scores; metadata/: exact launch config/revisions and provenance
`provenance`	C:\Users\nikak\source\repos\LASR\teachingclaudewhyreplication\scratch\nonmoral\overnightbaseline.py --checkpoint math
`cost_accounting`	Use metadata/costaccounting.json and judgebudget_ledger.json; raw shared-account OpenRouter deltas are not attributable experiment costs.
`naming_correction`	Local run_name no longer replaces automatically derived eval/model identity. Historical revisions remain readable through the old URL redirect; no evaluation or rescoring was performed.