HelloSun/openPangu-VL-7B-Q4KM-gguf

openPangu-VL-7B GGUF

Verified artifacts are published at https://huggingface.co/HelloSun/openPangu-VL-7B-Q4KM-gguf:

• openPangu-VL-7B-Q4_K_M.gguf (text model)
• openPangu-VL-7B-mmproj-F16.gguf (vision projector)
• openpangu-vl-llama.cpp.patch (llama.cpp source changes)
• build-openpangu-vl.sh (download and CPU build helper)
• LICENSE (upstream openPangu model license)
• OPEN SOURCE SOFTWARE NOTICE (upstream third-party notices)

The patch targets llama.cpp commit 1268edac579110aeb1407f7928f53193ed4e28d4. The Q4_K_M text model passed text inference and image description with the F16 mmproj.

Image example: NASA astronaut spacewalk

Prompt

What is happening in this photo?

Model response

This photo shows an astronaut performing a spacewalk, or extravehicular activity (EVA), outside a spacecraft. The astronaut is wearing a spacesuit designed to protect them from the harsh conditions of space, including vacuum

Generated with the published Q4_K_M text model and F16 mmproj. The response is shown as produced with a 48-token generation limit.

Image source and status: NASA astronaut Robert Behnken during an EVA on STS-130, February 17, 2010. The photo is in the U.S. public domain as a NASA work. Wikimedia Commons file page.

Apply and build

bash build-openpangu-vl.sh

The helper downloads the artifacts, checks out the target llama.cpp revision in a separate llama.cpp-openpangu-vl directory, applies the patch, and builds llama-cli and llama-mtmd-cli for CPU.

Run

./llama.cpp-openpangu-vl/build/bin/llama-cli \
  -m ./openpangu-vl-artifacts/openPangu-VL-7B-Q4_K_M.gguf \
  -p '1+1等于几？' -n 48 -st

./llama.cpp-openpangu-vl/build/bin/llama-mtmd-cli \
  -m ./openpangu-vl-artifacts/openPangu-VL-7B-Q4_K_M.gguf \
  --mmproj ./openpangu-vl-artifacts/openPangu-VL-7B-mmproj-F16.gguf \
  --image ./openpangu-vl-artifacts/astronaut-robert-behnken-eva.jpg \
  -p 'What is happening in this photo?' -n 48