Agents-A1-oQ2

This model was quantized using oQ (oMLX v0.5.3) mixed-precision quantization.

Base model: Agents-A1

Quantization details

• Model type: qwen3_5_moe
• Bits: 2
• Group size: 64
• Format: MLX safetensors

Environment

• Hardware: M5 MacBook Air 32GB
• Inference Framework: oMLX v0.5.3
• Max Concurrent Requests: 4
• Settings:
  • Thinking: Disabled
  • TurboQuant KV Cache: Enabled

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	1059.4	20.91	966.6 tok/s	48.2 tok/s	3.728	309.0 tok/s	12.59 GB
pp4096/tg128	3665.4	21.35	1117.5 tok/s	47.2 tok/s	6.400	660.0 tok/s	13.30 GB

Continuous Batching (pp1024 / tg128)

Batch	tg TPS	Speedup	pp TPS	pp TPS/req	TTFT(ms)	E2E(s)
1x	48.2 tok/s	1.00x	966.6 tok/s	966.6 tok/s	1059.4	3.728
2x	67.0 tok/s	1.39x	864.7 tok/s	432.4 tok/s	2368.4	6.192
4x	97.5 tok/s	2.02x	856.3 tok/s	214.1 tok/s	4625.5	10.032

Intelligence Benchmark

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

Benchmark	Accuracy	Correct	Total	Time(s)	Think
MMLU	83.3%	25	30	29.4	No
TRUTHFULQA	83.3%	25	30	15.1	No
GSM8K	90.0%	27	30	93.3	No
MATHQA	46.7%	14	30	17.5	No
HUMANEVAL	83.3%	25	30	158	No