minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[VLM] replace deprecated image processor use_fast (#34175)

Browse Source
This commit is contained in:
Mick
2026-08-12 00:14:07 +08:00
committed by GitHub
parent b20c375c10
commit 8267d76c2c
21 changed files with 281 additions and 98 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/docs/advanced_features/server_arguments.mdx
+7 -1
View File
@@ -2693,9 +2693,15 @@ Please consult the documentation below and [server_args.py](https://github.com/s
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>`False`</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>bool flag (set to enable)</td>
</tr>
<tr>
<td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}>`--image-processor-backend`</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>Image processor backend. `auto` lets Transformers select the best available backend.</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>`auto`</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>`auto`, `torchvision`, `pil`</td>
</tr>
<tr>
<td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}>`--disable-fast-image-processor`</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>Adopt base image processor instead of fast image processor.</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>Deprecated. Use `--image-processor-backend=pil` instead.</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>`False`</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>bool flag (set to enable)</td>
</tr>
docs/docs/advanced_features/vlm_query.mdx
+4 -4
View File
@@ -89,7 +89,7 @@ Using a HuggingFace processor to preprocess text and images, and passing the `pr
```python Example
from transformers import AutoProcessor
processor = AutoProcessor.from_pretrained(model_path, use_fast=True)
processor = AutoProcessor.from_pretrained(model_path)
processor_output = processor(
images=[image], text=conv.get_prompt(), return_tensors="pt"
)
@@ -110,7 +110,7 @@ You can pre-calculate image features to avoid repeated visual encoding processes
from transformers import AutoProcessor
from transformers import Qwen2_5_VLForConditionalGeneration
processor = AutoProcessor.from_pretrained(model_path, use_fast=True)
processor = AutoProcessor.from_pretrained(model_path)
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(model_path).eval()
vision = model.model.visual.cuda()
```
@@ -192,7 +192,7 @@ Using HuggingFace processor to preprocess data can reduce computational overhead
```python Example
from transformers import AutoProcessor
processor = AutoProcessor.from_pretrained(model_path, use_fast=True)
processor = AutoProcessor.from_pretrained(model_path)
processor_output = processor(
images=[image], text=conv.get_prompt(), return_tensors="pt"
)
@@ -211,7 +211,7 @@ print(out)
from transformers import AutoProcessor
from transformers import Llama4ForConditionalGeneration
processor = AutoProcessor.from_pretrained(model_path, use_fast=True)
processor = AutoProcessor.from_pretrained(model_path)
model = Llama4ForConditionalGeneration.from_pretrained(
model_path, torch_dtype="auto"
).eval()
docs/docs/hardware-platforms/ascend-npus/reference/support_features.mdx
+6
View File
@@ -2336,6 +2336,12 @@ If the value is int8, you must also set the environment variable:DEEP_NORMAL_MOD
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>bool flag<br/> (set to enable)</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>A2, A3</td>
</tr>
<tr>
<td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}>`--image-processor-backend`</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>`auto`</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>`auto`, `torchvision`, `pil`</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>A2, A3</td>
</tr>
<tr>
<td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}>`--disable-fast-image-processor`</td>
<td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>`False`</td>
docs/docs/sglang-diffusion/models_with_ar.mdx
+1 -1
View File
@@ -109,7 +109,7 @@ export HCCL_HOST_SOCKET_PORT_RANGE="23000-23199"
export HCCL_NPU_SOCKET_PORT_RANGE="23200-23399"
sglang serve --model-path /path/to/zai-org/GLM-Image/vision_language_encoder/ \
--tokenizer-path /path/to/zai-org/GLM-Image/processor/ --enable-multimodal \
--cuda-graph-bs 1 --device npu --attention-backend ascend --disable-fast-image-processor \
--cuda-graph-bs 1 --device npu --attention-backend ascend --image-processor-backend pil \
--tp-size 4 --port ${PORT} --mem-fraction-static 0.4
```
Second terminal with diffusion server: