Add validation artifact validation/benchmark/qwen_image_bf16_vs_fp8_fixed_1024_50_compare.md
Browse files
validation/benchmark/qwen_image_bf16_vs_fp8_fixed_1024_50_compare.md
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
### Performance Comparison Report
|
| 2 |
+
|
| 3 |
+
#### 1. High-level Summary
|
| 4 |
+
| Metric | Baseline | New | Diff | Status |
|
| 5 |
+
| :--- | :--- | :--- | :--- | :--- |
|
| 6 |
+
| **E2E Latency** | 13589.20 ms | 12159.39 ms | **-1429.81 ms (-10.5%)** | ✅ |
|
| 7 |
+
| **Throughput** | 0.07 req/s | 0.08 req/s | - | - |
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
#### 2. Stage Breakdown
|
| 11 |
+
| Stage Name | Baseline (ms) | New (ms) | Diff (ms) | Diff (%) | Status |
|
| 12 |
+
| :--- | :--- | :--- | :--- | :--- | :--- |
|
| 13 |
+
| InputValidationStage | 0.03 | 0.03 | +0.00 | +3.9% | ⚪️ |
|
| 14 |
+
| TextEncodingStage | 599.85 | 666.43 | +66.57 | +11.1% | ⚪️ |
|
| 15 |
+
| LatentPreparationStage | 0.10 | 0.16 | +0.05 | +49.7% | ⚪️ |
|
| 16 |
+
| TimestepPreparationStage | 0.33 | 0.52 | +0.18 | +54.9% | ⚪️ |
|
| 17 |
+
| DenoisingStage | 12928.76 | 11437.40 | -1491.36 | -11.5% | 🟢 |
|
| 18 |
+
| DecodingStage | 58.55 | 52.30 | -6.25 | -10.7% | ⚪️ |
|
| 19 |
+
|
| 20 |
+
|
| 21 |
+
<details>
|
| 22 |
+
<summary>Metadata</summary>
|
| 23 |
+
|
| 24 |
+
- Baseline Commit: `d071bfde2540084b816b9b5c8f7396be0efb3cd4`
|
| 25 |
+
- New Commit: `c2b3e42ad64521c018443f7efe294658a0b1da3b`
|
| 26 |
+
- Timestamp: 2026-04-19T13:30:57.086590
|
| 27 |
+
</details>
|