A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Doc] Remove Molmo warning (#27527)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Oct 26, 2025 at 16:22 UTC
be7b55a83d3399346b81febbfbd4e6682ab12601
1 file changed
-31
docs/models/supported_models.md
-31
@@ -736,37 +736,6 @@ Some models are supported only via the [Transformers backend](#transformers). Th
736
!!! note
737
To use `TIGER-Lab/Mantis-8B-siglip-llama3`, you have to pass `--hf_overrides '{"architectures": ["MantisForConditionalGeneration"]}'` when running vLLM.
738
739
-!!! warning
740
- The output quality of `AllenAI/Molmo-7B-D-0924` (especially in object localization tasks) has deteriorated in recent updates.
741
-
742
- For the best results, we recommend using the following dependency versions (tested on A10 and L40):
743
-
744
- ??? code "Dependency versions"
745
-
746
- ```text
747
- # Core vLLM-compatible dependencies with Molmo accuracy setup (tested on L40)
748
- torch==2.5.1
749
- torchvision==0.20.1
750
- transformers==4.48.1
751
- tokenizers==0.21.0
752
- tiktoken==0.7.0
753
- vllm==0.7.0
754
-
755
- # Optional but recommended for improved performance and stability
756
- triton==3.1.0
757
- xformers==0.0.28.post3
758
- uvloop==0.21.0
759
- protobuf==5.29.3
760
- openai==1.60.2
761
- opencv-python-headless==4.11.0.86
762
- pillow==10.4.0
763
-
764
- # Installed FlashAttention (for float16 only)
765
- flash-attn>=2.5.6 # Not used in float32, but should be documented
766
- ```
767
-
768
- **Note:** Make sure you understand the security implications of using outdated packages.
769
-
739
!!! note
740
The official `openbmb/MiniCPM-V-2` doesn't work yet, so we need to use a fork (`HwwwH/MiniCPM-V-2`) for now.
741
For more details, please see: <https://github.com/vllm-project/vllm/pull/4087#issuecomment-2250397630>