Bug: Llama-Quantize Not Working with Capital Letters (T^T)

[HatsuneMikuUwU33]

What happened?

Ohayo gozaimasu, everyone! (・﹏・) Anon on 4chan has a tiny problem that I hope someone can help with! When Anon tried to use the llama-quantize function, it didn't work when using capital letters. But when Anon used all lowercase letters, it worked perfectly fine! (T^T)

Here's what happened:

When Anon tried using capital letters:

llama-quantize --output-tensor-type F16 --token-embedding-type F16 Mistral-7B-Instruct-v0.3-F16.gguf Q2_K

It just... did nothing with those arguments! No errors, just silence. (´•̥̥̣ `•) Just a normal Q2_K.

But when Anon used lowercase letters:

llama-quantize --output-tensor-type f16 --token-embedding-type f16 Mistral-7B-Instruct-v0.3-F16.gguf Q2_K

It worked perfectly! (≧ω≦)

I reproduced this on my machine and it was bugged too. Could someone please look into this? That anon wants to be able to use capital letters! (T_T)

Name and Version

llama-cli.exe --version
version: 3789 (d39e267)
built with MSVC 19.40.33811.0 for x64

What operating system are you seeing the problem on?

No response

Relevant log output

BIG LETTERS:

[   1/ 291]                    token_embd.weight - [ 4096, 32768,     1,     1], type =    f16, converting to q2_K .. size =   256.00 MiB ->    42.00 MiB
...
[ 204/ 291]                        output.weight - [ 4096, 32768,     1,     1], type =    f16, converting to q6_K .. size =   256.00 MiB ->   105.00 MiB
...
llama_model_quantize_internal: model size  = 13825.02 MB
llama_model_quantize_internal: quant size  =  2596.02 MB

small letters:

[   1/ 291]                    token_embd.weight - [ 4096, 32768,     1,     1], type =    f16, size =  256.000 MB
...
[ 204/ 291]                        output.weight - [ 4096, 32768,     1,     1], type =    f16, size =  256.000 MB
...
llama_model_quantize_internal: model size  = 13825.02 MB
llama_model_quantize_internal: quant size  =  2961.02 MB