Update README.md
Browse files
README.md
CHANGED
@@ -345,6 +345,37 @@ The following performance benchmarks were conducted with [vLLM](https://docs.vll
|
|
345 |
</tr>
|
346 |
</thead>
|
347 |
<tbody style="text-align: center">
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
348 |
<tr>
|
349 |
<th rowspan="3" valign="top">A100x1</th>
|
350 |
<th>Qwen/Qwen2.5-VL-7B-Instruct</th>
|
@@ -439,9 +470,40 @@ The following performance benchmarks were conducted with [vLLM](https://docs.vll
|
|
439 |
</tr>
|
440 |
</thead>
|
441 |
<tbody style="text-align: center">
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
442 |
<tr>
|
443 |
<th rowspan="3" valign="top">A100x1</th>
|
444 |
-
<th>Qwen/Qwen2.5-VL-7B-Instruct
|
445 |
<td></td>
|
446 |
<td>0.7</td>
|
447 |
<td>1347</td>
|
|
|
345 |
</tr>
|
346 |
</thead>
|
347 |
<tbody style="text-align: center">
|
348 |
+
<tr>
|
349 |
+
<th rowspan="3" valign="top">A6000x1</th>
|
350 |
+
<th>Qwen/Qwen2.5-VL-7B-Instruct</th>
|
351 |
+
<td></td>
|
352 |
+
<td>4.9</td>
|
353 |
+
<td>912</td>
|
354 |
+
<td>3.2</td>
|
355 |
+
<td>1386</td>
|
356 |
+
<td>3.1</td>
|
357 |
+
<td>1431</td>
|
358 |
+
</tr>
|
359 |
+
<tr>
|
360 |
+
<th>neuralmagic/Qwen2.5-VL-7B-Instruct-quantized.w8a8</th>
|
361 |
+
<td>1.50</td>
|
362 |
+
<td>3.6</td>
|
363 |
+
<td>1248</td>
|
364 |
+
<td>2.1</td>
|
365 |
+
<td>2163</td>
|
366 |
+
<td>2.0</td>
|
367 |
+
<td>2237</td>
|
368 |
+
</tr>
|
369 |
+
<tr>
|
370 |
+
<th>neuralmagic/Qwen2.5-VL-7B-Instruct-quantized.w4a16</th>
|
371 |
+
<td>2.05</td>
|
372 |
+
<td>3.3</td>
|
373 |
+
<td>1351</td>
|
374 |
+
<td>1.4</td>
|
375 |
+
<td>3252</td>
|
376 |
+
<td>1.4</td>
|
377 |
+
<td>3321</td>
|
378 |
+
</tr>
|
379 |
<tr>
|
380 |
<th rowspan="3" valign="top">A100x1</th>
|
381 |
<th>Qwen/Qwen2.5-VL-7B-Instruct</th>
|
|
|
470 |
</tr>
|
471 |
</thead>
|
472 |
<tbody style="text-align: center">
|
473 |
+
<tr>
|
474 |
+
<th rowspan="3" valign="top">A6000x1</th>
|
475 |
+
<th>Qwen/Qwen2.5-VL-7B-Instruct</th>
|
476 |
+
<td></td>
|
477 |
+
<td>0.4</td>
|
478 |
+
<td>1837</td>
|
479 |
+
<td>1.5</td>
|
480 |
+
<td>6846</td>
|
481 |
+
<td>1.7</td>
|
482 |
+
<td>7638</td>
|
483 |
+
</tr>
|
484 |
+
<tr>
|
485 |
+
<th>neuralmagic/Qwen2.5-VL-7B-Instruct-quantized.w8a8</th>
|
486 |
+
<td>1.41</td>
|
487 |
+
<td>0.5</td>
|
488 |
+
<td>2297</td>
|
489 |
+
<td>2.3</td>
|
490 |
+
<td>10137</td>
|
491 |
+
<td>2.5</td>
|
492 |
+
<td>11472</td>
|
493 |
+
</tr>
|
494 |
+
<tr>
|
495 |
+
<th>neuralmagic/Qwen2.5-VL-7B-Instruct-quantized.w4a16</th>
|
496 |
+
<td>1.60</td>
|
497 |
+
<td>0.4</td>
|
498 |
+
<td>1828</td>
|
499 |
+
<td>2.7</td>
|
500 |
+
<td>12254</td>
|
501 |
+
<td>3.4</td>
|
502 |
+
<td>15477</td>
|
503 |
+
</tr>
|
504 |
<tr>
|
505 |
<th rowspan="3" valign="top">A100x1</th>
|
506 |
+
<th>Qwen/Qwen2.5-VL-7B-Instruct</th>
|
507 |
<td></td>
|
508 |
<td>0.7</td>
|
509 |
<td>1347</td>
|