@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / ed920135c8
Fork 0

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 16, 10:45 UTC

[Bugfix] Molmo text-only input bug fix (#9397)

Co-authored-by: sanghol <sanghol@allenai.org> Co-authored-by: Roger Wang <136131678+ywang96@users.noreply.github.com> Co-authored-by: Roger Wang <ywang@roblox.com>

Reza Salehi committed Oct 15, 2024 at 21:56 UTC ed920135c8490440453a64e197fce5e1e6459225
1 file changed +6 -3
vllm/model_executor/models/molmo.py
+6 -3
@@ -946,9 +946,12 @@ def pad_images(
946
947
948 def input_processor_for_molmo(ctx: InputContext, llm_inputs: LLMInputs):
949 - prompt = llm_inputs["prompt"]
950 - multi_modal_data = llm_inputs.get("multi_modal_data")
951 - image = multi_modal_data.get("image")
949 + prompt = llm_inputs.get("prompt", None)
950 + multi_modal_data = llm_inputs.get("multi_modal_data", None)
951 + if multi_modal_data is not None:
952 + image = multi_modal_data.get("image", None)
953 + else:
954 + image = None
955 processor = cached_get_processor(ctx.model_config.model,
956 trust_remote_code=True,
957 revision=ctx.model_config.code_revision)