Short context of llava?

[arkohut]

I have a question about the context using llava 1.6. Does the screenshot means that the total context of it is 2048 token? It seems that the output even not finished before using out the context.

Here is the command:

./llava-cli -m ~/ggml-yi-34b-f16-q_5_k.gguf \
  --mmproj ~/mmproj-llava-34b-f16-q6_k.gguf \
  --image ~/00263.jpg \
  -e \
  -p '<|im_start|>system\nAnswer the questions.<|im_end|><|im_start|>user\n<image>\nProvide a full description.<|im_end|><|im_start|>assistant\n'

[supportend]

I don't know this specific model, in my output i see:
llm_load_print_meta: n_ctx_train = 4096
That means, i can set it to 4096 with:
-c 4096

[arkohut]

You are absolutely right, thanks for replying my stupid question...I suddenly forget this critical parameter.