Downloads · 30 days
21
5% of all-time downloads
Qwen3.5-9B-Fable-5-v1-oQ5-mtp
mlx-works/Qwen3.5-9B-Fable-5-v1-oQ5-mtp
Qwen3.5-9B-Fable-5-v1-oQ5-mtp is a machine learning model from mlx-works. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for mlx.
This model was quantized using oQ (oMLX v0.5.1) mixed-precision quantization.
All-time downloads
459
Public
Parameters
9.2B
6.6 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors6.6 GB · 100%
Parameter types
How the weights are stored.
U329.2B · 100%
Model card
From the Hugging Face model README
Qwen3.5-9B-Fable-5-v1-oQ5-mtp
This model was quantized using oQ (oMLX v0.5.1) mixed-precision quantization.
Quantization details
- Model type: qwen3_5
- Bits: 5
- Group size: 64
- Format: MLX safetensors
Environment
- Hardware: M5 MacBook Air 32GB
- Inference Framework: oMLX v0.5.1
- Max Concurrent Requests: 4
- 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 | 1601.4 | 28.14 | 639.4 tok/s | 35.8 tok/s | 5.206 | 221.3 tok/s | 7.16 GB |
| pp4096/tg128 | 6183.4 | 27.63 | 662.4 tok/s | 36.5 tok/s | 9.729 | 434.1 tok/s | 7.76 GB |
Continuous Batching (pp1024 / tg128)
| Batch | tg TPS | Speedup | pp TPS | pp TPS/req | TTFT(ms) | E2E(s) |
|---|---|---|---|---|---|---|
| 1x | 35.8 tok/s | 1.00x | 639.4 tok/s | 639.4 tok/s | 1601.4 | 5.206 |
| 2x | 39.0 tok/s | 1.09x | 589.6 tok/s | 294.8 tok/s | 3473.4 | 10.040 |
| 4x | 68.6 tok/s | 1.92x | 604.8 tok/s | 151.2 tok/s | 6588.3 | 14.232 |
Intelligence Benchmark
Note: Each benchmark round tests only 30 questions. Results are for reference only.
| Benchmark | Accuracy | Correct | Total | Time(s) | Think |
|---|---|---|---|---|---|
| MMLU | 73.3% | 22 | 30 | 35.8 | No |
| TRUTHFULQA | 86.7% | 26 | 30 | 13.6 | No |
| GSM8K | 93.3% | 28 | 30 | 178.5 | No |
| MATHQA | 43.3% | 13 | 30 | 19.7 | No |
| HUMANEVAL | 80.0% | 24 | 30 | 164.9 | No |
Summary
| Benchmark | Accuracy |
|---|---|
| MMLU | 73.3% |
| TRUTHFULQA | 86.7% |
| GSM8K | 93.3% |
| MATHQA | 43.3% |
| HUMANEVAL | 80.0% |
| Average | 75.3% |
- Library
- mlx
- Type
- qwen3_5
- Created
- Jul 16, 2026
- Updated
- Jul 17, 2026