reference-direct.jsonl (2688B)
1 {"id": "check-1", "option_ids": ["yes", "no", "insufficient"], "probabilities": [0.9994852422786518, 0.00017946777542364315, 0.0003352899459245753], "option_logits": [27.75, 19.125, 19.75], "input_tokens": 142, "prompt_sha256": "7ac35785358f0c656eeb741ca0a51f03e0133753a71e2879a18ad0ab56d0c024", "prompt_version": "direct-options-v1", "model": {"source": "Qwen/Qwen3.5-4B", "revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "dtype": "bfloat16", "torch_version": "2.12.0", "transformers_version": "5.5.4"}, "readout": "native full-vocabulary last-position logits restricted to declared answer slots", "probability_status": "conditional option score; uncalibrated as decision confidence"} 2 {"id": "check-2", "option_ids": ["rollback", "no_rollback"], "probabilities": [0.0007096703991005882, 0.9992903296008995], "option_logits": [21.0, 28.25], "input_tokens": 124, "prompt_sha256": "86a43592753e7bced3d136c1f0482475d9192130daf09046152705df8857b9bd", "prompt_version": "direct-options-v1", "model": {"source": "Qwen/Qwen3.5-4B", "revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "dtype": "bfloat16", "torch_version": "2.12.0", "transformers_version": "5.5.4"}, "readout": "native full-vocabulary last-position logits restricted to declared answer slots", "probability_status": "conditional option score; uncalibrated as decision confidence"} 3 {"id": "check-3", "option_ids": ["all_three", "some", "none"], "probabilities": [0.9999141762532671, 5.828966066441289e-05, 2.753408606849224e-05], "option_logits": [27.75, 18.0, 17.25], "input_tokens": 137, "prompt_sha256": "d03fec9a6f1a890f85daaa250f959a701007ff7047de4ddcc0f39e77e78ee32d", "prompt_version": "direct-options-v1", "model": {"source": "Qwen/Qwen3.5-4B", "revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "dtype": "bfloat16", "torch_version": "2.12.0", "transformers_version": "5.5.4"}, "readout": "native full-vocabulary last-position logits restricted to declared answer slots", "probability_status": "conditional option score; uncalibrated as decision confidence"} 4 {"id": "check-4", "option_ids": ["yes", "no"], "probabilities": [0.9998911030899734, 0.00010889691002655443], "option_logits": [28.25, 19.125], "input_tokens": 129, "prompt_sha256": "e8e78b793a77a04186fa7012e227d5064cc1d085d830a67d440d10f2f6bce1ca", "prompt_version": "direct-options-v1", "model": {"source": "Qwen/Qwen3.5-4B", "revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "dtype": "bfloat16", "torch_version": "2.12.0", "transformers_version": "5.5.4"}, "readout": "native full-vocabulary last-position logits restricted to declared answer slots", "probability_status": "conditional option score; uncalibrated as decision confidence"}