Commit History
More hints on what to do with CUDA Out of memory errors (#925)
b0cf397
unverified
Juraj Bednar
commited on
new evals_per_epoch and saves_per_epoch to make things cleaner (#944)
5f79b82
unverified
Respect sequence_len in config for `type: llama2_chat` (#926)
f1de29d
unverified
Mixtral official (#942)
7fabc4d
unverified
Update requirements.txt (#940)
9a5eb39
unverified
Mixtral: More correct MoE, lower loss (#932)
86487c2
unverified
update to latest transformers for mixstral support (#929)
35f9b0f
unverified
Mixtral multipack (#928)
68b227a
unverified
fixing prompt template of chatml by removal of linebreak (#922)
03c6318
unverified
support for mamba (#915)
40a6362
unverified
chore: clarify Readme on sharegpt system role
d339beb
unverified
fix(tokenizer): handle fast tokenizer properly for bos/eos (#914)
fde091c
unverified
Pin flash-attn to 2.3.3 (#919)
06ae392
unverified
Support device_map=sequential & max_memory config parameters (#903)
992e742
unverified
Feat(wandb): Refactor to be more flexible (#767)
a1da39c
unverified
feature: loss watchdog for terminating training runs that are failing (#899)
58ec8b1
unverified
Remove learning rate scheduler in deepspeed config to avoid conflict (#909)
476a205
unverified
fix for qwen w lora (#906)
3e3229e
unverified
ensure merged model matches the training dtype (#902)
1d21aa6
unverified
fix: remove FA for qwen examples (#900)
a48dbf6
unverified
update datasets version to cut down the warnings due to pyarrow arg change (#897)
6a4562a
unverified
Feat: Add Qwen (#894)
1115c50
unverified
fix: warning should not show if eval_batch_size not provided (#896)
7ee3c4c
unverified
Feat: Add warmup_ratio (#893)
fb12895
unverified
chore(doc): Add info on changing role in sharegpt (#886)
9fc29e0
unverified
fix: revert local dir dataset load (#878)
575a082
unverified
Install from git url (#874)
ddf8150
unverified
Phi update 202311 (#876)
9bf854e
unverified
don't train if eval split is too small (#873)
797f3dd
unverified
try #2: pin hf transformers and accelerate to latest release, don't reinstall pytorch (#867)
0de1457
unverified
Feat: Add dataset loading from S3, GCS (#765)
3cc67d2
unverified
allow overriding of model_config parameters from the YML (#853)
1bc1186
unverified
add e2e tests for checking functionality of resume from checkpoint (#865)
b3a61e8
unverified
make docker command more robust (#861)
8a8d1c4
unverified
lint fix that didn't get caught by linter (#866)
332984d
unverified
Update data.py for signature generation (#851)
48630f5
unverified
Docs: add instructions to 1-click launching on public clouds (#862)
b33c1d5
unverified
multipack len should use max, not min (#863)
0c2a630
unverified
adds llama and mistral dropout support (#858)
db8a8af
unverified
various bugfixes (#856)
1470650
unverified
chore(doc): Separate section on runpod (#860)
501b4d1
unverified
feat(doc): add more info on train_on_split (#855)
306fe19
unverified
include the suffix modified string in ascii art (#852)
614cff4
unverified
cleanup the old multipack dataloader (#841)
1a6309c
unverified
Pin optimum package (#838)
105d0b3
unverified
Bryan Thornbury
commited on