chore(doc): Add info on changing role in sharegpt (#886)

Install from git url (#874)

try #2: pin hf transformers and accelerate to latest release, don't reinstall pytorch (#867)

Feat: Add dataset loading from S3, GCS (#765)

allow overriding of model_config parameters from the YML (#853)

make docker command more robust (#861)

lint fix that didn't get caught by linter (#866)

Docs: add instructions to 1-click launching on public clouds (#862)

chore(doc): Separate section on runpod (#860)

feat(doc): add more info on train_on_split (#855)

Feat: Added Gradio support (#812)

update table for rwkv4 support, fix process count for dataset (#822)

fix eval_steps to be a sane default (#797)

Add docker advanced instruction to README (#792)

Create preprocess CLI (#785)

chore(readme): Improve documentation on conversation field (#782)

Fix: Cannot tokenize with bf16 and on cpu (#766)

Implement fused modules (#747)

add to docs (#703)

Clarify custom format example (#729)

add noisy embedding (#721)

fix(doc): update default doc according to arg (#714)

fix(doc): Add note on inference w sample packing (#712)

Update README with some explanations (#700)

refactor to set eval_batch_size earlier if unset, so we can warn if mismatched (#662)

Make dataset_processes configurable (#651)

Fix bug when using pretokenized datasets (#652)

add support for defined train split (#654)

Feat: Add example for Mistral (#644)

Warn users to login to HuggingFace (#645)

skip some flash attn patches unless explicitly enabled (#643)

use fastchat conversations template (#578)

Feat: Add support for upstream FA2 (#626)

Added quotes to the pip install -e command to fix an incompatibility with shells that do glob expansion like zsh (#632)

Feat(doc): Add eval_sample_packing to doc (#625)

update README w deepspeed info (#605)

Feat(data): Allow loading local csv and text (#594)

Feat(doc): Add features to doc (#583)

support custom field for completion from yml (#580)

update support matrix with btlm and phi (#579)

refactor scripts/finetune.py into new cli modules (#550)

let hf trainer handle torch compile (#516)

Add training callback to send predictions to WandB table (#521)

document that packaging needs to be installed before flash-attn (#559)

ergonomic update to optimizer config doc (#548)

update readme to point to direct link to runpod template, cleanup install instrucitons (#532)

Adding NCCL Timeout Guide (#536)

Fix(doc): Inform Windows users to use WSL/docker (#518)

Added advanced DDP args (#515)

support for datasets with multiple names (#480)