Using with llama.cpp and got an error about the length of tensor name
I tried with python and llama.cpp, then got an unexpected error.
So I wonder anyone didn't get it like below ?
load_backend: loaded RPC backend from /home/bagel/llamacpp/build/bin/libggml-rpc.so
load_backend: loaded CPU backend from /home/bagel/llamacpp/build/bin/libggml-cpu-haswell.so
build: 5689 (c4650301) with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu
main: llama backend init
main: load the model and apply lora adapter, if any
gguf_init_from_file_impl: tensor name 399 is too long: 69 >= 64
gguf_init_from_file_impl: failed to read tensor info
llama_model_load: error loading model: llama_model_loader: failed to load model from ./ema-q4_0.gguf
llama_model_load_from_file_impl: failed to load model
common_init_from_params: failed to load model './ema-q4_0.gguf'
main: error: unable to load model
I tried ema-q4_0.gguf and ema-q8_0.gguf, got an same error.
llama.cpp doesn't support image generation model so far; btw, ggc b2 method works; and ema gguf is a checkpoint, might need more work on that
I see but llama.cpp announced that they supported multi modal models as below :
https://github.com/ggml-org/llama.cpp/blob/master/docs/multimodal.md
well, not listed bagel on it eventhough..
https://github.com/ggml-org/llama.cpp/blob/master/docs/multimodal.md
if you check it carefully, you can see those multimodal models they supported are all from input side, output is always text
Hmmm.... The main point I would like to emphasize is not about image output,
but rather that loading itself is impossible due to an error, even before considering any input.
The fact is that llama.cpp should be able to load multimodal models, but your model cannot be loaded at all because of what I menentioned error...
this is not an error; it's different file that llama.cpp supposed to support; this recent gguf doesn't contain metadata to rebuild the tokenizer; which is not on list models; similar to flux and stable diffusion, it's not for running in llama.cpp