Repository navigation
Eval bug: HunyuanOCR doesn't work from b9263 (merge HunyuanOCR into HunyuanVL and fix OCR vision precision) #23481
Description
Name and Version
$ llama-cli --version↵
version: 9263 (6a257d4)↵
built with GNU 11.4.0 for Linux x86_64
Operating systems
Linux
GGML backends
Vulkan
Hardware
NVIDIA GeForce GTX 1660 SUPER (6GB VRAM)
Models
- HunyuanOCR-BF16.gguf
- mmproj-HunyuanOCR-BF16.gguf
https://huggingface.co/ggml-org/HunyuanOCR-GGUF
Problem description & steps to reproduce
Command:
$ llama-cli -m "$HOME"/00_M/HunyuanOCR-BF16.gguf --mmproj "$HOME"/00_M/mmproj-HunyuanOCR-BF16.gguf -c 8192 --temp 0
Last version working: b9260
First Bad Commit
Newer versions not working from: b9263
"mtmd, model : merge HunyuanOCR into HunyuanVL and fix OCR vision precision" PR #23329
Relevant log output
Logs
Error output:
Loading model... -0.00.176.461 E mtmd_get_memory_usage: error: Failed to load CLIP model from /home/kuro/00_M/mmproj-HunyuanOCR-BF16.gguf
0.00.176.466 E srv load_model: [mtmd] failed to get memory usage of mmproj
/0.01.739.649 E clip_init: failed to load model '/home/kuro/00_M/mmproj-HunyuanOCR-BF16.gguf': load_hparams: unknown projector type: hunyuanocr
0.01.739.987 E mtmd_init_from_file: error: Failed to load CLIP model from /home/kuro/00_M/mmproj-HunyuanOCR-BF16.gguf
0.01.739.991 E srv load_model: failed to load multimodal model, '/home/kuro/00_M/mmproj-HunyuanOCR-BF16.gguf'
Failed to load the model
Activity
CISC commentedon May 21, 2026
@ngxson Forgot to update GGUF?
ngxson commentedon May 21, 2026
hmm yeah, I just updated the GGUF: https://huggingface.co/ggml-org/HunyuanOCR-GGUF