Darwin-35B-A3B-Opus-oQ4
This model was quantized using oQ (oMLX v0.3.1) mixed-precision quantization.
Quantization details
- Model type: qwen3_5_moe
- Bits: 4
- Group size: 64
- Format: MLX safetensors
Benchmark
| Model | File size | MMLU | JMMLU | HELLASWAG | ARC_CHALLENGE | GSM8K |
|---|---|---|---|---|---|---|
| Darwin-35B-A3B-Opus-oQ3 | 16.8 GB | 77.7% | 77.0% | 87.3% | 94.7% | 92.3% |
| Darwin-35B-A3B-Opus-oQ3.5 | 18.1 GB | 79.0% | 77.0% | 88.3% | 94.7% | 91.7% |
| Darwin-35B-A3B-Opus-oQ4 | 21.1 GB | 83.7% | 77.3% | 94.3% | 94.7% | 92.0% |
Detail
| Model | Benchmark | Accuracy | Correct | Total | Time(s) |
|---|---|---|---|---|---|
| Darwin-35B-A3B-Opus-oQ3 | MMLU | 77.7% | 233 | 300 | 579.3 |
| Darwin-35B-A3B-Opus-oQ3 | JMMLU | 77.0% | 231 | 300 | 215.6 |
| Darwin-35B-A3B-Opus-oQ3 | HELLASWAG | 87.3% | 262 | 300 | 273.5 |
| Darwin-35B-A3B-Opus-oQ3 | ARC_CHALLENGE | 94.7% | 284 | 300 | 186.5 |
| Darwin-35B-A3B-Opus-oQ3 | GSM8K | 92.3% | 277 | 300 | 1064.6 |
| Darwin-35B-A3B-Opus-oQ3.5 | MMLU | 79.0% | 237 | 300 | 542.6 |
| Darwin-35B-A3B-Opus-oQ3.5 | JMMLU | 77.0% | 231 | 300 | 216.8 |
| Darwin-35B-A3B-Opus-oQ3.5 | HELLASWAG | 88.3% | 265 | 300 | 274.9 |
| Darwin-35B-A3B-Opus-oQ3.5 | ARC_CHALLENGE | 94.7% | 284 | 300 | 189.7 |
| Darwin-35B-A3B-Opus-oQ3.5 | GSM8K | 91.7% | 275 | 300 | 1130.3 |
| Darwin-35B-A3B-Opus-oQ4 | MMLU | 83.7% | 251 | 300 | 601.9 |
| Darwin-35B-A3B-Opus-oQ4 | JMMLU | 77.3% | 232 | 300 | 220 |
| Darwin-35B-A3B-Opus-oQ4 | HELLASWAG | 94.3% | 283 | 300 | 297.4 |
| Darwin-35B-A3B-Opus-oQ4 | ARC_CHALLENGE | 94.7% | 284 | 300 | 188.8 |
| Darwin-35B-A3B-Opus-oQ4 | GSM8K | 92.0% | 276 | 300 | 1230.5 |
- Downloads last month
- 21
Model size
6B params
Tensor type
BF16
·
U32 ·
Hardware compatibility
Log In to add your hardware
4-bit
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for RepublicOfKorokke/Darwin-35B-A3B-Opus-oQ4
Base model
FINAL-Bench/Darwin-35B-A3B-Opus