Qwopus3.5-4B-Coder-oQ4e-mtp

This model was quantized using oQ (oMLX v0.5.1) mixed-precision quantization.

Quantization details

• Model type: qwen3_5
• Bits: 4
• Group size: 64
• Format: MLX safetensors

Environment

• Hardware: M5 MacBook Air 32GB
• Inference Framework: oMLX v0.5.1
• Settings:
  • Thinking: Disabled
  • Chat template parameter: enable_thinking=false (forced)
  • TurboQuant KV Cache: Disabled (if enable: reduces speed)
  • Native MTP: Enabled (key speed improvement)

Performance Benchmarks

Note: Results are for reference only and may vary depending on hardware, software configuration, and workload.

Single Request Results

Test	TTFT(ms)	TPOT(ms)	pp TPS	tg TPS	E2E(s)	Throughput	Peak Mem
pp1024/tg128	868.8	15.05	1178.7 tok/s	67.0 tok/s	2.795	412.1 tok/s	3.51 GB
pp4096/tg128	3546.2	14.97	1155.1 tok/s	67.3 tok/s	5.470	772.2 tok/s	4.12 GB

Continuous Batching (pp1024 / tg128)

Batch	tg TPS	Speedup	pp TPS	pp TPS/req	TTFT(ms)	E2E(s)
1x	67.0 tok/s	1.00x	1178.7 tok/s	1178.7 tok/s	868.8	2.795
2x	82.9 tok/s	1.24x	1058.3 tok/s	529.1 tok/s	1935.2	5.025
4x	134.4 tok/s	2.01x	1053.2 tok/s	263.3 tok/s	3758.2	7.698

Intelligence Benchmark

Note: Each benchmark round tests only 30 questions. Results are for reference only.

Benchmark	Accuracy	Correct	Total	Time(s)	Think
MMLU	63.3%	19	30	26.0	No
TRUTHFULQA	63.3%	19	30	10.3	No
GSM8K	96.7%	29	30	121.4	No
MATHQA	53.3%	16	30	15.5	No
HUMANEVAL	76.7%	23	30	123.7	No

Summary

Benchmark	Accuracy
MMLU	63.3%
TRUTHFULQA	63.3%
GSM8K	96.7%
MATHQA	53.3%
HUMANEVAL	76.7%
Average	70.7%