Downloads · 30 days
40
11% of all-time downloads
gemma-4-E2B-it-oQ4-mtp
mlx-works/gemma-4-E2B-it-oQ4-mtp
gemma-4-E2B-it-oQ4-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 v0.5.5) mixed-precision quantization.
All-time downloads
351
Public
Parameters
5.1B
3.9 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors3.8 GB · 99%
Parameter types
How the weights are stored.
U324.7B · 91%
Model card
From the Hugging Face model README
gemma-4-E2B-it-oQ4-mtp
This model was quantized using oQ (oMLX v0.5.5) mixed-precision quantization.
Base model: google/gemma-4-E2B-it
MTP head: google/gemma-4-E2B-it-assistant (merged into single checkpoint)
Quantization details
- Model type: gemma4
- Bits: 4
- Group size: 64
- Format: MLX safetensors
- MTP: Merged from assistant checkpoint (Lightning MTP)
Environment
- Hardware: M5 MacBook Air 32GB
- Inference Framework: oMLX v0.5.5
- Max Concurrent Requests: 4
- Settings:
- Thinking: Disabled
- TurboQuant KV Cache: Enabled (4-bit)
- 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 | E2E(s) | Throughput | Peak Mem |
|---|---|---|---|---|---|---|---|
| pp1024/tg128 | 503.2 | 14.24 | 2034.9 tok/s | 70.8 tok/s | 2.330 | 494.5 tok/s | 3.08 GB |
| pp4096/tg128 | 1639.3 | 14.82 | 2498.6 tok/s | 68.0 tok/s | 3.544 | 1191.8 tok/s | 3.11 GB |
Continuous Batching (pp1024 / tg128)
| Batch | tg TPS | Speedup | pp TPS | pp TPS/req | TTFT(ms) | E2E(s) |
|---|---|---|---|---|---|---|
| 1x | 70.8 tok/s | 1.00x | 2034.9 tok/s | 2034.9 tok/s | 503.2 | 2.330 |
| 2x | 74.1 tok/s | 1.05x | 2879.5 tok/s | 1439.8 tok/s | 711.1 | 2.561 |
| 4x | 182.8 tok/s | 2.58x | 3503.4 tok/s | 875.9 tok/s | 898.8 | 3.324 |
Intelligence Benchmark
Note: Each benchmark round tests only 30 questions. Results are for reference only.
| Benchmark | Accuracy | Correct | Total | Time(s) | Think |
|---|---|---|---|---|---|
| MMLU | 50.0% | 15 | 30 | 9.9 | No |
| TRUTHFULQA | 43.3% | 13 | 30 | 5.6 | No |
| GSM8K | 93.3% | 28 | 30 | 60.6 | No |
| MATHQA | 13.3% | 4 | 30 | 26.5 | No |
| HUMANEVAL | 73.3% | 22 | 30 | 92.4 | No |
- Library
- mlx
- Type
- gemma4
- Created
- Aug 5, 2026
- Updated
- Aug 16, 2026