Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- llm-inference
|
| 5 |
+
- cuda
|
| 6 |
+
- rust
|
| 7 |
+
- gguf
|
| 8 |
+
- benchmarks
|
| 9 |
+
pretty_name: Swizzle Runtime Benchmarks
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# Swizzle Runtime (`swizzle-rt`) Benchmarks
|
| 13 |
+
|
| 14 |
+
Verified back-to-back inference benchmarks for **[Swizzle Runtime](https://github.com/Swizzle-AI/swizzle-runtime)** (`v0.1.0`), a pure-Rust + CUDA/Vulkan inference engine for GGUF models.
|
| 15 |
+
|
| 16 |
+
- **GitHub Repository:** https://github.com/Swizzle-AI/swizzle-runtime
|
| 17 |
+
- **Website:** https://swizzle-ten.vercel.app
|
| 18 |
+
- **Community Hardware Leaderboard:** https://github.com/Swizzle-AI/swizzle-runtime/issues/5
|
| 19 |
+
|
| 20 |
+
## Verified Baseline (AMD Ryzen 7 170 + 4GB RTX 3050 Laptop GPU, 5 reps, bit-identical greedy parity vs `llama.cpp`)
|
| 21 |
+
|
| 22 |
+
| Model | Workload | Swizzle Runtime (`swizzle-rt`) | Notes |
|
| 23 |
+
| --- | --- | --- | --- |
|
| 24 |
+
| `Qwen3-0.6B-Q8_0` | CUDA `pp512` | **11,804.23 tok/s** | 43.37 ms TTFT |
|
| 25 |
+
| `Qwen3-0.6B-Q8_0` | CUDA `tg128` | **209.70 tok/s** | +6.5% vs `llama.cpp`, ITL p50 4.84 ms |
|
| 26 |
+
| `Qwen3-0.6B-Q8_0` | CPU `edit256` | **90.61 tok/s** | +37.7% via RAD-1 self-speculative decoding |
|
| 27 |
+
| `LFM3-1.7B-Q4_K_M` | CUDA `tg128` | **144.26 tok/s** | +7.0% vs `llama.cpp` |
|
| 28 |
+
|
| 29 |
+
## Quick Install (Linux x86_64, CUDA 12.8)
|
| 30 |
+
|
| 31 |
+
```bash
|
| 32 |
+
curl -fsSL https://raw.githubusercontent.com/Swizzle-AI/swizzle-runtime/main/install.sh | sh
|
| 33 |
+
swizzle-rt bench model.gguf --device cuda --json
|
| 34 |
+
```
|