Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -4,11 +4,17 @@ tags:
|
|
| 4 |
- mlx
|
| 5 |
- oq
|
| 6 |
- quantized
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
---
|
| 8 |
|
| 9 |
# Ornith-1.0-35B-oQ2
|
| 10 |
|
| 11 |
-
This model was quantized using [oQ](https://github.com/jundot/omlx) (oMLX v0.5.
|
|
|
|
|
|
|
| 12 |
|
| 13 |
## Quantization details
|
| 14 |
|
|
@@ -16,3 +22,43 @@ This model was quantized using [oQ](https://github.com/jundot/omlx) (oMLX v0.5.2
|
|
| 16 |
- **Bits**: 2
|
| 17 |
- **Group size**: 64
|
| 18 |
- **Format**: MLX safetensors
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
- mlx
|
| 5 |
- oq
|
| 6 |
- quantized
|
| 7 |
+
- benchmark
|
| 8 |
+
- performance
|
| 9 |
+
- moe
|
| 10 |
+
- ornith
|
| 11 |
---
|
| 12 |
|
| 13 |
# Ornith-1.0-35B-oQ2
|
| 14 |
|
| 15 |
+
This model was quantized using [oQ](https://github.com/jundot/omlx) (oMLX v0.5.3) mixed-precision quantization.
|
| 16 |
+
|
| 17 |
+
Base model: [Ornith-1.0-35B](https://huggingface.co/Ornith-1.0-35B)
|
| 18 |
|
| 19 |
## Quantization details
|
| 20 |
|
|
|
|
| 22 |
- **Bits**: 2
|
| 23 |
- **Group size**: 64
|
| 24 |
- **Format**: MLX safetensors
|
| 25 |
+
|
| 26 |
+
## Environment
|
| 27 |
+
|
| 28 |
+
- **Hardware**: M5 MacBook Air 32GB
|
| 29 |
+
- **Inference Framework**: oMLX v0.5.3
|
| 30 |
+
- **Max Concurrent Requests**: 4
|
| 31 |
+
- **Settings**:
|
| 32 |
+
- Thinking: Disabled
|
| 33 |
+
- TurboQuant KV Cache: Enabled
|
| 34 |
+
|
| 35 |
+
## Performance Benchmarks
|
| 36 |
+
|
| 37 |
+
> **Note**: Results are for reference only and may vary depending on hardware, software configuration, and workload.
|
| 38 |
+
|
| 39 |
+
### Single Request Results
|
| 40 |
+
|
| 41 |
+
| Test | TTFT(ms) | TPOT(ms) | pp TPS | tg TPS | E2E(s) | Throughput | Peak Mem |
|
| 42 |
+
|------|----------|----------|--------|--------|--------|------------|----------|
|
| 43 |
+
| pp1024/tg128 | 1063.3 | 20.83 | 963.0 tok/s | 48.4 tok/s | 3.722 | 309.5 tok/s | 12.62 GB |
|
| 44 |
+
| pp4096/tg128 | 3662.7 | 21.71 | 1118.3 tok/s | 46.4 tok/s | 6.435 | 656.4 tok/s | 13.36 GB |
|
| 45 |
+
|
| 46 |
+
### Continuous Batching (pp1024 / tg128)
|
| 47 |
+
|
| 48 |
+
| Batch | tg TPS | Speedup | pp TPS | pp TPS/req | TTFT(ms) | E2E(s) |
|
| 49 |
+
|-------|--------|---------|--------|------------|----------|--------|
|
| 50 |
+
| 1x | 48.4 tok/s | 1.00x | 963.0 tok/s | 963.0 tok/s | 1063.3 | 3.722 |
|
| 51 |
+
| 2x | 67.2 tok/s | 1.39x | 871.7 tok/s | 435.9 tok/s | 2349.2 | 6.159 |
|
| 52 |
+
| 4x | 100.6 tok/s | 2.08x | 868.4 tok/s | 217.1 tok/s | 4569.1 | 9.804 |
|
| 53 |
+
|
| 54 |
+
## Intelligence Benchmark
|
| 55 |
+
|
| 56 |
+
> **Note**: Each benchmark round tests only 30 questions. Results are for reference only.
|
| 57 |
+
|
| 58 |
+
| Benchmark | Accuracy | Correct | Total | Time(s) | Think |
|
| 59 |
+
|-----------|----------|---------|-------|---------|-------|
|
| 60 |
+
| MMLU | 70.0% | 21 | 30 | 33 | No |
|
| 61 |
+
| TRUTHFULQA | 76.7% | 23 | 30 | 15.3 | No |
|
| 62 |
+
| GSM8K | 96.7% | 29 | 30 | 100.3 | No |
|
| 63 |
+
| MATHQA | 16.7% | 5 | 30 | 69.6 | No |
|
| 64 |
+
| HUMANEVAL | 83.3% | 25 | 30 | 172.8 | No |
|