Code

Add Qwen2.5VL support#12402

Code

Draft

HimariO wants to merge 6 commits intoggml-org:masterggml-org/llama.cpp:masterfrom HimariO:qwen25-vlHimariO/llama.cpp.qwen2vl:qwen25-vlCopy head branch name to clipboard

Commits

Commits on Mar 15, 2025

• implment vision model architecture, gguf convertor

  HimariO

  committed
  6a8bae0

  Copy full SHA for 6a8bae0

• handle window attention inputs

  HimariO

  committed
  cfc78c8

  Copy full SHA for cfc78c8

• add support for Qwen2_5_VLForConditionalGeneration

  HimariO

  committed
  8d69c2f

  Copy full SHA for 8d69c2f

• add debug utils

  HimariO

  committed
  6047054

  Copy full SHA for 6047054

• fix few incorrect tensor memory layout

  HimariO

  committed
  50d0b69

  Copy full SHA for 50d0b69

• move position id remap out of ggml to avoid int32 cuda operations

  HimariO

  committed
  bc4163b

  Copy full SHA for bc4163b