biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Hotfix][Core][VLM] Disable chunked prefill by default and prefix caching for multimodal models (#8425)

Browse Source
This commit is contained in:
Roger Wang
2024-09-12 14:06:51 -07:00
committed by GitHub
parent 019877253b
commit c16369455f
2 changed files with 13 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/models/__init__.py
View File

@@ -90,12 +90,12 @@ _MULTIMODAL_MODELS = {
"PaliGemmaForConditionalGeneration": ("paligemma",
"PaliGemmaForConditionalGeneration"),
"Phi3VForCausalLM": ("phi3v", "Phi3VForCausalLM"),
"UltravoxModel": ("ultravox", "UltravoxModel"),
"QWenLMHeadModel": ("qwen", "QWenLMHeadModel"),
"PixtralForConditionalGeneration": ("pixtral",
"PixtralForConditionalGeneration"),
"QWenLMHeadModel": ("qwen", "QWenLMHeadModel"),
"Qwen2VLForConditionalGeneration": ("qwen2_vl",
"Qwen2VLForConditionalGeneration"),
"UltravoxModel": ("ultravox", "UltravoxModel"),
}
_CONDITIONAL_GENERATION_MODELS = {
"BartModel": ("bart", "BartForConditionalGeneration"),