Downloads · 30 days
191
33% of all-time downloads
MiMo-V2.6-Distill-Qwen-9B-oQ4e-mtp
mlx-works/MiMo-V2.6-Distill-Qwen-9B-oQ4e-mtp
MiMo-V2.6-Distill-Qwen-9B-oQ4e-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) mixed-precision quantization.
All-time downloads
581
Public
Parameters
9.7B
6.5 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors6.5 GB · 100%
Parameter types
How the weights are stored.
U329B · 93%
Model card
From the Hugging Face model README
MiMo-V2.6-Distill-Qwen-9B-oQ4e-mtp
This model was quantized using oQ (oMLX) mixed-precision quantization.
Base model: XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B (SFT fine-tune of Qwen/Qwen3.5-9B)
Chat template: MiMo's own chat_template.jinja, as shipped
Quantization details
- Model type: qwen3_5
- Bits: 4
- Group size: 64
- Format: MLX safetensors
- MTP: Grafted from Qwen/Qwen3.5-9B (mtp_num_hidden_layers: 1; source checkpoint ships no
mtp.*tensors) - Calibration: oQ4e (enhanced, imatrix-based; oqe_code_multilingual)
Environment
- Hardware: M5 MacBook Air 32GB
- Inference Framework: oMLX v0.7.0.dev4
- Max Concurrent Requests: 4
- Settings:
- Thinking: Disabled
- TurboQuant KV Cache: Disabled
- Lightning 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 | tg TPS (No MTP) | E2E(s) | Throughput | Peak Mem |
|---|---|---|---|---|---|---|---|---|
| pp1024/tg128 | 1472.3 | 26.35 | 695.5 tok/s | 38.2 tok/s | 26.8 tok/s | 4.843 | 237.9 tok/s | 7.49 GB |
| pp4096/tg128 | 5713.9 | 26.89 | 716.8 tok/s | 37.5 tok/s | 26.1 tok/s | 9.146 | 461.8 tok/s | 7.98 GB |
Continuous Batching (pp1024 / tg128)
| Batch | tg TPS | Speedup | pp TPS | pp TPS/req | TTFT(ms) | E2E(s) |
|---|---|---|---|---|---|---|
| 1x | 38.2 tok/s | 1.00x | 695.5 tok/s | 695.5 tok/s | 1472.3 | 4.843 |
| 2x | 51.6 tok/s | 1.35x | 517.4 tok/s | 258.7 tok/s | 3031.9 | 8.924 |
| 4x | 61.1 tok/s | 1.60x | 448.2 tok/s | 112.0 tok/s | 6691.5 | 17.513 |
Intelligence Benchmark
Note: Each benchmark round tests only 30 questions. Results are for reference only.
| Benchmark | Accuracy | Correct | Total | Time(s) | Think |
|---|---|---|---|---|---|
| MMLU | 66.7% | 20 | 30 | 29.1 | No |
| TRUTHFULQA | 76.7% | 23 | 30 | 8.9 | No |
| GSM8K | 86.7% | 26 | 30 | 80.9 | No |
| MATHQA | 36.7% | 11 | 30 | 54.4 | No |
| HUMANEVAL | 76.7% | 23 | 30 | 138.9 | No |
- Library
- mlx
- Type
- qwen3_5
- Created
- Sep 22, 2026
- Updated
- Sep 23, 2026