ggml has CANN support, does this also support CANN?
Good project, and this project use ggml lastest version, https://github.com/monatis/clip.cpp use a old version. and current used ggml has CANN support, does the clip.cpp can support CANN?
Good project, and this project use ggml lastest version, https://github.com/monatis/clip.cpp use a old version. and current used ggml has CANN support, does the clip.cpp can support CANN?
The principle is supported, but some minor adaptation is required.
Change the source code for CANN, but there has some error [root@localhost yysu-888-clip.cpp]# ./build/clip -m convert_model/clip_f16.gguf --model_version openai_clip_vit_base_patch32 --mode zeros_shot_image_classification --model_type f32 --image_path example/images/apple.jpeg --label_path example/label_english.txt System Info:
model_load: using CANN backend
vocab size: 49408
trigger word img already in vocab
clip_vision params backend buffer size = 330.62 MB(VRAM) (200 tensors)
clip_text params backend buffer size = 241.96 MB(VRAM) (197 tensors)
img_path:example/images/apple.jpeg,width=640,height=640,c=3
clip_vision compute buffer size: 0.88 MB(VRAM)
ggml_backend_cann_graph_compute: error: op not supported node_519 (SQRT)
/root/clip/yysu-888-clip.cpp/thirty_party/ggml/src/ggml-cann.cpp:1626: GGML_ASSERT(ok) failed
[New LWP 4378]
[New LWP 4379]
[New LWP 4380]
[New LWP 4381]
[New LWP 4382]
[New LWP 4387]
[New LWP 4388]
[New LWP 4389]
[New LWP 4390]
[New LWP 4391]
[New LWP 4392]
[New LWP 4393]
[New LWP 4394]
[New LWP 4395]
[New LWP 4396]
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/usr/lib64/libthread_db.so.1".
0x0000ffff8acf9fc0 in wait4 () from /usr/lib64/libc.so.6
#0 0x0000ffff8acf9fc0 in wait4 () from /usr/lib64/libc.so.6
#1 0x0000ffff8b187fb0 in ggml_abort () from /root/clip/yysu-888-clip.cpp/build/libggml.so
#2 0x0000ffff8b1d79b8 in ggml_backend_cann_graph_compute(ggml_backend*, ggml_cgraph*) () from /root/clip/yysu-888-clip.cpp/build/libggml.so
#3 0x0000ffff8b1b3454 in ggml_backend_graph_compute () from /root/clip/yysu-888-clip.cpp/build/libggml.so
#4 0x000000000043b3a8 in GGMLRunner::compute(std::function<ggml_cgraph* ()>, int, bool, ggml_tensor**, ggml_context*) ()
#5 0x0000000000448750 in CLIPVisionModelRunner<ClipVisionModel>::infer(image_u8_t, std::vector<float, std::allocator
ggml_backend_cann_graph_compute: error: op not supported node_519 (SQRT)
some op need implement by youself