GGML_ASSERT Error when running llama-embedding with this model

#1
by rayzinnz - opened

When testing with
./llama-embedding -m snowflake-arctic-embed-l-v2.0.F16.gguf -e -p "Hello world" --verbose-prompt
I get this error:
GGML_ASSERT(params.n_batch >= params.n_ctx) failed
Did you get it and how did you resolve it?

OK found the fix, we must specify the context and batch length, i.e. 512 or 8192, i.e.
/llama-embedding --batch-size 8192 --ctx-size 8192 -m ...

rayzinnz changed discussion status to closed

Sign up or log in to comment