gguf invalid and no support in llama.cpp

#8
by cmp-nct - opened

Hi,
I noticed the gguf you provided, I'm not sure how you converted them but they are not valid gguf binaries and llama.cpp sadly doesn't support the architecture

Hello,

I believe you are right that llama.cpp does not support T5 models, but the binaries are not invalid. You can use them with candle:

cargo run --example quantized-t5 --release  -- \
  --model-id "jbochi/madlad400-3b-mt" --weight-file "model-q4k.gguf" \
  --prompt "<2de> How are you, my friend?" \
  --temperature 0

I generated the files using the util described in the quantized-t5 example

jbochi changed discussion status to closed

Oh I see, that's really painful that they used the same format and filename in an incompatible fashion.
Thanks for clarification

Sign up or log in to comment