Qwen3.8-27B-MLX-5bit-Group32 / mac_metrics.json
Chungulus's picture
Add files using upload-large-folder tool
4a39cf2 verified
Raw
History Blame Contribute Delete
5.29 kB
{
"schema_version": 1,
"role": "mac_quant_logit_metrics",
"model": "/Users/oz/Documents/ChatGPT/qwen3.8-27b/qwen38-27b-quants/quantization-suite/artifacts/MLX-5bit-Group32",
"source_revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0",
"measured_at": "2026-08-16T01:30:05.600566+00:00",
"inputs_sha256": "f1a0af6b9580739ebc9efa9375ae31aa6940dafe7ff6446097ed6cf8d9ab37de",
"positions": 106,
"mean_kl_divergence": 0.004990289304054009,
"reference_perplexity": 9.84727010437735,
"candidate_perplexity": 9.883995913928073,
"perplexity_delta": 0.03672580955072391,
"top1_token_agreement": 0.9528301886792453,
"peak_unified_memory_gb": 20.314038144,
"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": 0.005657650530338287,
"reference_perplexity": 105.96934764067565,
"candidate_perplexity": 108.92524694120041,
"top1_token_agreement": 1.0
},
{
"id": "math",
"positions": 32,
"mean_kl_divergence": 0.002653036266565323,
"reference_perplexity": 2.852022040669331,
"candidate_perplexity": 2.8364977771812123,
"top1_token_agreement": 1.0
},
{
"id": "code",
"positions": 30,
"mean_kl_divergence": 0.003575812326744199,
"reference_perplexity": 2.682231762957344,
"candidate_perplexity": 2.6788740507542905,
"top1_token_agreement": 0.9333333333333333
},
{
"id": "multilingual",
"positions": 23,
"mean_kl_divergence": 0.009477759711444378,
"reference_perplexity": 34.40712742702027,
"candidate_perplexity": 34.45066238752625,
"top1_token_agreement": 0.8695652173913043
}
],
"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": 1.1150232750110602,
"mean_semantic_similarity": 0.9774516999721528,
"artifact_bytes": 21989153029,
"peak_unified_memory_gb": 23.639191525,
"average_generation_tps": 12.568759032272249,
"mean_kl_divergence": 0.004990289304054009,
"reference_perplexity": 9.84727010437735,
"candidate_perplexity": 9.883995913928073,
"perplexity_delta": 0.03672580955072391,
"top1_token_agreement": 0.9528301886792453,
"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"
},
"MLX-6bit": {
"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": 0.9411764705882353,
"mtp_speedup": 1.261565462823067,
"mean_semantic_similarity": 0.988053274154663,
"artifact_bytes": 23669991658,
"peak_unified_memory_gb": 25.573573157,
"average_generation_tps": 9.953450085041483,
"mean_kl_divergence": 0.0015874907093228034,
"reference_perplexity": 9.84727010437735,
"candidate_perplexity": 9.821571581388875,
"perplexity_delta": -0.025698522988474082,
"top1_token_agreement": 1.0,
"logit_positions": 106,
"logit_inputs_sha256": "f1a0af6b9580739ebc9efa9375ae31aa6940dafe7ff6446097ed6cf8d9ab37de"
}
}
},
"selection": "selected"
}