Attention mask and position id fixes for packing (#285)

Merge pull request #355 from tmm1/bitsandbytes-fixes

bump to latest bitsandbytes release with major bug fixes

use newer pynvml package

log GPU memory usage

pin accelerate so it works with llama2 (#330)

latest HEAD of accelerate causes 0 loss immediately w FSDP (#321)

add hf_transfer to requirements for faster hf upload

Update requirements.txt

add support for opimum bettertransformers

Add accelerate dep

Move black to dev requirements

Convert attrdict to addict

update docker to compile latest bnb to properly support qlora

Update requirements.txt

integrate qlora? maybe?

update entrypoint and force min accelerate

Fix BNB OOM by pinning version

docker layer caching, build w axolotl from base build

cleanup empty lines, tweak env for runpod setup

quickstart instructions for starting from runpod (#5)

fix install to work with latest alpaca lora 4bit

4bit quantized support (wip)

helpful info output

various bugfixes

config chooser, update readme instructions, device config, llama flash attention, debug out the labels, fix config key checks, other bugfixes

WIP for axolotl trainer