Update README.md
Browse files
README.md
CHANGED
|
@@ -11,6 +11,8 @@ After implementing Eagle3, the inference performance of Qwen3-30B-Moe using the
|
|
| 11 |
|
| 12 |
The TPS (tokens per second) improvement reached nearly 70%.
|
| 13 |
|
|
|
|
|
|
|
| 14 |
|
| 15 |
| model | gpu | tps |
|
| 16 |
|---------|---------|---------|
|
|
@@ -18,7 +20,8 @@ The TPS (tokens per second) improvement reached nearly 70%.
|
|
| 18 |
| qwen3-30b-moe_eagle3 | h200 | 231 |
|
| 19 |
| qwen3-30b_moe | 8*h200 | 183 |
|
| 20 |
| qwen3-30b_moe-eagle3 | 8*h200 | 325 |
|
| 21 |
-
|
|
|
|
| 22 |
## How to use
|
| 23 |
|
| 24 |
To use Eagle3 with SGLang, first replace the qwen3_moe.py file in SGLang’s directory (sglang/python/sglang/srt/models/) with the qwen3_moe.py file from this project.
|
|
|
|
| 11 |
|
| 12 |
The TPS (tokens per second) improvement reached nearly 70%.
|
| 13 |
|
| 14 |
+
On a single RTX 5090, the TPS (transactions per second) of Qwen3-8B-Eagle3 increased from 164 to 268.
|
| 15 |
+
|
| 16 |
|
| 17 |
| model | gpu | tps |
|
| 18 |
|---------|---------|---------|
|
|
|
|
| 20 |
| qwen3-30b-moe_eagle3 | h200 | 231 |
|
| 21 |
| qwen3-30b_moe | 8*h200 | 183 |
|
| 22 |
| qwen3-30b_moe-eagle3 | 8*h200 | 325 |
|
| 23 |
+
| qwen3-30b_moe | 8*5090 | 164 |
|
| 24 |
+
| qwen3-30b_moe-eagle3 | 8*5090 | 268 |
|
| 25 |
## How to use
|
| 26 |
|
| 27 |
To use Eagle3 with SGLang, first replace the qwen3_moe.py file in SGLang’s directory (sglang/python/sglang/srt/models/) with the qwen3_moe.py file from this project.
|