veloxquant_mlx.benchmarks package¶
Benchmark scripts for measuring VeloxQuant-MLX performance and accuracy.
Groups standalone, runnable benchmarks (attention throughput, Metal kernel microbenchmarks, RaBitQ/CacheRoute/comm-VQ comparisons, end-to-end model KV benchmarks, and workload-replay traces) that are executed directly as scripts rather than imported as a library API; this package holds no re-exports of its own.
Submodules¶
- veloxquant_mlx.benchmarks.attend_benchmark module
- veloxquant_mlx.benchmarks.benchmark_db module
BenchmarkRecordBenchmarkRecord.idBenchmarkRecord.methodBenchmarkRecord.model_idBenchmarkRecord.architectureBenchmarkRecord.context_lengthBenchmarkRecord.batch_sizeBenchmarkRecord.memory_bytesBenchmarkRecord.memory_reductionBenchmarkRecord.throughput_tok_sBenchmarkRecord.latency_ms_per_tokenBenchmarkRecord.perplexity_deltaBenchmarkRecord.chipBenchmarkRecord.mlx_versionBenchmarkRecord.macos_versionBenchmarkRecord.workloadBenchmarkRecord.created_atBenchmarkRecord.sourceBenchmarkRecord.tagsBenchmarkRecord.from_dict()BenchmarkRecord.to_dict()BenchmarkRecord.savings_percentBenchmarkRecord.match_score()
BenchmarkDatabasebenchmark_fingerprint()default_benchmark_dir()
- veloxquant_mlx.benchmarks.cacheroute_benchmark module
- veloxquant_mlx.benchmarks.comm_vq_benchmark module
- veloxquant_mlx.benchmarks.falcon_rabitq_benchmark module
- veloxquant_mlx.benchmarks.metal_kernel_benchmark module
- veloxquant_mlx.benchmarks.model_kv_benchmark module
- veloxquant_mlx.benchmarks.rabitq_benchmark module
- veloxquant_mlx.benchmarks.workload_replay_benchmark module
WorkloadScenarioMemorySnapshotWorkloadResultWorkloadResult.methodWorkloadResult.workloadWorkloadResult.n_streamsWorkloadResult.total_tokensWorkloadResult.wall_time_sWorkloadResult.ttft_ms_meanWorkloadResult.decode_ms_meanWorkloadResult.decode_ms_p50WorkloadResult.decode_ms_p95WorkloadResult.tokens_per_secWorkloadResult.peak_memory_bytesWorkloadResult.compression_ratioWorkloadResult.quantize_overhead_msWorkloadResult.dequantize_overhead_msWorkloadResult.memory_snapshots
run_workload()run_suite()format_summary_table()results_to_json()main()