tinyfrank-1.4B/README.md

license

license
apache-2.0

Undi95 type frankenstein of TinyLLama 1.1b https://github.com/jzhang38/TinyLlama https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v1.0

GGUF custom quants included

The secret sauce:

slices:
  - sources:
    - model: "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
      layer_range: [0, 14]
  - sources:
    - model: "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
      layer_range: [8, 22]
merge_method: passthrough
dtype: bfloat16

How to run as gguf:

git clone https://github.com/ggerganov/llama.cpp && cd llama.cpp && make -j
wget https://huggingface.co/SkunkworksAI/tinyfrank-1.4B/resolve/main/tinyfrank-q6L.gguf
./server -m tinyfrank-q6L.gguf --host "my.internal.ip.or.my.cloud.host.name.goes.here.com" -c 512