qwerrwe / src

Commit History

Fix: Higher vram usage for mistral and sample_packing (#691)
669f1d0
unverified

Nanobit commited on

flash_attention + sample packing for stablelm 3b (#671)
2d60ba3
unverified

winglian commited on

Fix: ValueError when FA + Mistral when padding_side=right (#681)
eb480df
unverified

Nanobit commited on

Fix: Future deprecation warning with use_auth_token (#680)
69fac9a
unverified

Nanobit commited on

Fix(tokenizer): Set rstrip,lstrip,norm to False (#678)
e0b7eea
unverified

Nanobit commited on

chore: Clean up repetitive model kwargs (#670)
e62d590
unverified

Nanobit commited on

Feat: Allow usage of native Mistral FA when no sample_packing (#669)
697c50d
unverified

Nanobit commited on

refactor to set eval_batch_size earlier if unset, so we can warn if mismatched (#662)
2642cae
unverified

winglian commited on

remove patch fix for phi (#664)
f34648c
unverified

winglian commited on

prepared dataset caching, other misc fixes (#665)
e50a64e
unverified

winglian commited on

Make dataset_processes configurable (#651)
9ec2077
unverified

corbt commited on

Fix bug when using pretokenized datasets (#652)
590d603
unverified

ich commited on

add support for defined train split (#654)
409ca0f
unverified

winglian commited on

don't strip the prompt for check since we don't strip to tokenize anymore (#650)
8662e8f
unverified

winglian commited on

fix for flash attn w mistral w/o sammple packing (#648)
b2edaae
unverified

winglian commited on

Feat: Add example for Mistral (#644)
eb41f76
unverified

Nanobit commited on

Fix(cfg): Add validation for save_strategy and eval_strategy (#633)
383f88d
unverified

Nanobit commited on

Mistral flash attn packing (#646)
b6ab8aa
unverified

winglian commited on

Warn users to login to HuggingFace (#645)
85b0be2
unverified

Napuh commited on

Fix bug in dataset loading (#284)
8fe0e63
unverified

ethanhs commited on

Correct typos in datasets.py (#639)
d1236f2
unverified

felixonmars commited on

skip some flash attn patches unless explicitly enabled (#643)
895f0a0
unverified

winglian commited on

use fastchat conversations template (#578)
e7d3e2d
unverified

winglian commited on

update for recent transformers updates (#636)
60c7c48
unverified

winglian commited on

attention_mask not needed for training (#642)
e8cbf50
unverified

winglian commited on

Feat: Add support for upstream FA2 (#626)
19a600a
unverified

Nanobit commited on

Fix: Fail bf16 check when running on cpu during merge (#631)
cfbce02
unverified

Nanobit commited on

better handling and logging of empty sharegpt turns (#603)
a363604
unverified

winglian commited on

create a model card with axolotl badge (#624)
501958b
unverified

winglian commited on

chore(callback): Remove old peft saving code (#510)
d5f8589
unverified

Nanobit commited on

misc fixes to add gptq tests (#621)
03e5907
unverified

winglian commited on

split completion text to sequence_len (#616)
97d3776
unverified

winglian commited on

run eval on the first step to get a baseline (#617)
2844eb2
unverified

winglian commited on

skip the gpu memory checks if the device is set to 'auto' (#609)
196ff11
unverified

winglian commited on

fix distributed devices (#612)
2fe95cd
unverified

Maxime commited on

support to disable exllama for gptq (#604)
faecff9
unverified

winglian commited on

Delete duplicate lines (#606)
aa656e0
unverified

bofenghuang commited on

improve handling for empty text on the tokenization step (#502)
1eebbd0
unverified

winglian commited on

Fix for check with cfg and merge_lora (#600)
62a7741
unverified

winglian commited on

minor tweaks to simplify (#597)
31b9e0c
unverified

winglian commited on

btlm and falcon monkey patches for flash attn (#566)
6b9b229
unverified

winglian commited on

add bf16 check (#587)
131afdb
unverified

winglian commited on

Feat(data): Allow loading local csv and text (#594)
00dce35
unverified

Nanobit commited on

gather/broadcast the max value of the packing efficiency automatically (#463)
b15b19e
unverified

winglian commited on

don't add position_ids for evals (#591)
ab534d7
unverified

winglian commited on

optionally configure sample packing for evals (#589)
21ec195
unverified

winglian commited on

make phi training work with Loras (#588)
62eaee7
unverified

winglian commited on

set fsdp state dict (#584)
be75668
unverified

Jan Philipp Harries Jan Philipp Harries commited on

pop block_cls since it's not an actual kwarg
aeec7c4

winglian commited on

don't resize embeddings if it's already large enough (#577)
3607882
unverified

winglian commited on