---
library_name: mlx
tags:
- mlx
- oq
- quantized
- benchmark
- performance
- moe
- ornith
---

# Ornith-1.0-35B-oQ2

This model was quantized using [oQ](https://github.com/jundot/omlx) (oMLX v0.5.3) mixed-precision quantization.

Base model: [Ornith-1.0-35B](https://huggingface.co/Ornith-1.0-35B)

## 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 | 1063.3 | 20.83 | 963.0 tok/s | 48.4 tok/s | 3.722 | 309.5 tok/s | 12.62 GB |
| pp4096/tg128 | 3662.7 | 21.71 | 1118.3 tok/s | 46.4 tok/s | 6.435 | 656.4 tok/s | 13.36 GB |

### Continuous Batching (pp1024 / tg128)

| Batch | tg TPS | Speedup | pp TPS | pp TPS/req | TTFT(ms) | E2E(s) |
|-------|--------|---------|--------|------------|----------|--------|
| 1x | 48.4 tok/s | 1.00x | 963.0 tok/s | 963.0 tok/s | 1063.3 | 3.722 |
| 2x | 67.2 tok/s | 1.39x | 871.7 tok/s | 435.9 tok/s | 2349.2 | 6.159 |
| 4x | 100.6 tok/s | 2.08x | 868.4 tok/s | 217.1 tok/s | 4569.1 | 9.804 |

## Intelligence Benchmark

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

| Benchmark | Accuracy | Correct | Total | Time(s) | Think |
|-----------|----------|---------|-------|---------|-------|
| MMLU | 70.0% | 21 | 30 | 33 | No |
| TRUTHFULQA | 76.7% | 23 | 30 | 15.3 | No |
| GSM8K | 96.7% | 29 | 30 | 100.3 | No |
| MATHQA | 16.7% | 5 | 30 | 69.6 | No |
| HUMANEVAL | 83.3% | 25 | 30 | 172.8 | No |