Justine Tunney

@cpumaxx This change only adds support for bf16. Once #6414 the next thing I'll do is upstream the llamafile bfloat16 kernels. Here's what one of them looks like:  I'm...

@cpumaxx Could you download https://huggingface.co/jartine/Mistral-7B-Instruct-v0.2-llamafile/blob/main/mistral-7b-instruct-v0.2.BF16.gguf and then build the code in the branch I just created https://github.com/jart/llama.cpp/tree/unified which unifies #6412 and #6414? Thanks!

Here's an example of what you should expect to see with that branch. ``` wget https://huggingface.co/jartine/Mistral-7B-Instruct-v0.2-llamafile/resolve/main/mistral-7b-instruct-v0.2.BF16.gguf wget https://justine.lol/tmp/getty.txt make -j32 main && ./main -m /disk/mistral/mistral-7b-instruct-v0.2.BF16.gguf -f ~/getty.txt -n 22 --temp...

Could you pass the flag `-f getty.txt` please after you've downloaded that file from the link above? Then re-post your results.

@ggerganov I've resolved all your suggestions (some with comments). PTAL. Your suggestions were very helpful in improving this PR. For example, both LUTs were able to be removed, which reduces...

Thank you, by the way, for not merging any conflicting changes. I'm very happy that I was able to rebase without conflicts.

I'll take a look into this in a week when I'll have time to give this the attention it needs.

Can you run it under `blink -sssse wine etc...` and post the output?

How could that be? This is a standard module in both Python 2 and 3. - https://docs.python.org/2/library/fcntl.html - https://docs.python.org/3.3/library/fcntl.html

Ohhh. Then this is something we definitely need to fix. It seems the way to do this, is to update the dimensions method in util.py to use this library: https://pypi.python.org/pypi/backports.shutil_get_terminal_size...