USBhost comments

Results 106 comments of


                                            USBhost

Add ability to load all text files from a subdirectory for training

This sounds useful. So I could mass convert epubs to text then just dump all the txt of them in a folder and just train off that.

What are the eos_token_id and bos_token_id

I would like to report all of [Neko's](https://huggingface.co/Neko-Institute-of-Science) tokenizers are current and match with https://huggingface.co/oobabooga/llama-tokenizer. Also if you want me to update stuff in the future just bug me here...

llamafile : improve moe prompt eval speed on cpu

Does it also help the other K quants?

llamafile : improve moe prompt eval speed on cpu

> @USBhost Unfortunately no. The K quants were designed to exploit under-utilization of CPU resources when doing matvecs. I tried copying and pasting the `Q5_K_M` code into a tinyBLAS 2-d...

LLaMA Implementation

> It looks like the tests which are currently failing are unrelated to the LLaMA code, so this should be good to review/use. > > If folks can try it...

LLaMA Implementation

After replacing transformers from Kobold with this PR I am able to load the shards as expected. Just I cant generate anything because Kobold still needs some changes. ![image](https://user-images.githubusercontent.com/7269941/222938895-2e8b9d71-6a88-417d-b7ed-14d8216d2ef4.png)

USBhost

Add ability to load all text files from a subdirectory for training

What are the eos_token_id and bos_token_id

llamafile : improve moe prompt eval speed on cpu

llamafile : improve moe prompt eval speed on cpu

LLaMA Implementation

LLaMA Implementation

LLaMA Implementation

Alternative batching behavior for mix-sized training

Alternative batching behavior for mix-sized training

Make VAE step sequential to prevent VRAM spikes, will fix #3059, #2082, #2561, #3462