{ "schema_version": 1, "role": "mac_quant_logit_metrics", "model": "/Users/oz/Documents/ChatGPT/qwen3.8-27b/qwen38-27b-quants/quantization-suite/artifacts/MLX-5bit-Group128", "source_revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0", "measured_at": "2026-08-16T01:51:49.184662+00:00", "inputs_sha256": "f1a0af6b9580739ebc9efa9375ae31aa6940dafe7ff6446097ed6cf8d9ab37de", "positions": 106, "mean_kl_divergence": 0.007879387376443395, "reference_perplexity": 9.84727010437735, "candidate_perplexity": 9.764229178940486, "perplexity_delta": -0.08304092543686359, "top1_token_agreement": 0.9905660377358491, "peak_unified_memory_gb": 17.804167926, "metal_memory_policy": { "device": { "device_name": "Apple M5 Pro", "max_recommended_working_set_size": 55662788608, "memory_size": 68719476736, "architecture": "applegpu_g17s", "max_buffer_length": 41747087360, "resource_limit": 499000 }, "cache_limit_bytes": 256000000, "wired_limit_bytes": 50096509747, "previous_cache_limit_bytes": 65283502899, "previous_wired_limit_bytes": 0, "warnings": [] }, "cases": [ { "id": "release", "positions": 21, "mean_kl_divergence": 0.009627277962863445, "reference_perplexity": 105.96934764067565, "candidate_perplexity": 104.98872011380237, "top1_token_agreement": 1.0 }, { "id": "math", "positions": 32, "mean_kl_divergence": 0.005563782528042793, "reference_perplexity": 2.852022040669331, "candidate_perplexity": 2.8400412963201846, "top1_token_agreement": 1.0 }, { "id": "code", "positions": 30, "mean_kl_divergence": 0.002371579874306917, "reference_perplexity": 2.682231762957344, "candidate_perplexity": 2.647959665303878, "top1_token_agreement": 1.0 }, { "id": "multilingual", "positions": 23, "mean_kl_divergence": 0.016689294949173927, "reference_perplexity": 34.40712742702027, "candidate_perplexity": 34.13602848310655, "top1_token_agreement": 0.9565217391304348 } ], "warnings": [ "KL is measured on fixed original text, not a public benchmark.", "BF16 log-probabilities are stored in float16 after float32 log-softmax; reported KL therefore has finite-storage approximation error." ], "dominance": { "candidate": { "functional_passed": true, "functional_failures": [], "text_passed": 10, "text_total": 10, "tools_passed": 5, "tools_total": 5, "vision_passed": 3, "vision_total": 3, "mtp_passed": true, "mtp_acceptance_rate": 0.9545454545454546, "mtp_speedup": 1.04670949751592, "mean_semantic_similarity": 0.9368221938610077, "artifact_bytes": 19467900026, "peak_unified_memory_gb": 20.756765927, "average_generation_tps": 14.54951526929673, "mean_kl_divergence": 0.007879387376443395, "reference_perplexity": 9.84727010437735, "candidate_perplexity": 9.764229178940486, "perplexity_delta": -0.08304092543686359, "top1_token_agreement": 0.9905660377358491, "logit_positions": 106, "logit_inputs_sha256": "f1a0af6b9580739ebc9efa9375ae31aa6940dafe7ff6446097ed6cf8d9ab37de" }, "baselines": { "MLX-5bit": { "functional_passed": true, "functional_failures": [], "text_passed": 10, "text_total": 10, "tools_passed": 5, "tools_total": 5, "vision_passed": 3, "vision_total": 3, "mtp_passed": true, "mtp_acceptance_rate": 0.9545454545454546, "mtp_speedup": 1.2124763256514137, "mean_semantic_similarity": 0.9520436644554138, "artifact_bytes": 20308322354, "peak_unified_memory_gb": 21.708525014, "average_generation_tps": 11.310333833788317, "mean_kl_divergence": 0.00583227794125395, "reference_perplexity": 9.84727010437735, "candidate_perplexity": 9.769640361449891, "perplexity_delta": -0.07762974292745817, "top1_token_agreement": 0.9716981132075472, "logit_positions": 106, "logit_inputs_sha256": "f1a0af6b9580739ebc9efa9375ae31aa6940dafe7ff6446097ed6cf8d9ab37de" } } }, "selection": "selected" }