{"api_version":"codesota.v1","data":{"archived_results":[],"evidence_status":"CodeSOTA measurements only; model-card claims are not benchmark results","measured_parameters":[{"basis":"Unique parameters counted in the loaded evaluator model on White","model":"mghafiri/qwen3.5-0.8B-decision-model","parameters":752393024,"revision":"4a9939034a27006b4b62ad5aed65a90b82d0d350","source":"https://huggingface.co/mghafiri/qwen3.5-0.8B-decision-model/tree/4a9939034a27006b4b62ad5aed65a90b82d0d350","updated":1792212113}],"model":"mghafiri/qwen3.5-0.8B-decision-model","page":"/models/mghafiri/qwen3.5-0.8B-decision-model","pending":[],"results":[{"benchmark":"ARC-Challenge / Qwen3.5 same-base cohort","coverage":"full","direction":"max","evaluation_seconds":34.899311495013535,"finished":1792212113,"gpu_slot_seconds":36.517934279982,"manifest_url":"/artifacts/1110e38a472a2087411800a5/manifest.json","metric":"accuracy","n":1172,"protocol_version":"codesota-qwen35-arc-challenge-v1","revision":"4a9939034a27006b4b62ad5aed65a90b82d0d350","run_id":"1110e38a472a2087411800a5","run_url":"/runs/1110e38a472a2087411800a5","scope":"8877f1f73ee0a3c138937e07b7f6ca917f2c1d914d223e6a8ffa42b9ac8f0a47","score":0.302901023890785,"task":"qwen35-arc-challenge-full"},{"benchmark":"ARC-Easy / Qwen3.5 same-base cohort","coverage":"full","direction":"max","evaluation_seconds":87.16299507697113,"finished":1792209100,"gpu_slot_seconds":88.84777087590192,"manifest_url":"/artifacts/7ce4a0170453ff0e8974416e/manifest.json","metric":"accuracy","n":2376,"protocol_version":"codesota-qwen35-arc-easy-v1","revision":"4a9939034a27006b4b62ad5aed65a90b82d0d350","run_id":"7ce4a0170453ff0e8974416e","run_url":"/runs/7ce4a0170453ff0e8974416e","scope":"84fe51c6790b19376ad4a234c742b80e0f73e3eb0d6a7257d48e5315d6a7c078","score":0.4898989898989899,"task":"qwen35-arc-easy-full"}],"revisions":["4a9939034a27006b4b62ad5aed65a90b82d0d350"],"source":"https://huggingface.co/mghafiri/qwen3.5-0.8B-decision-model"}}