Commit History

Phi2 multipack (#1173)
814aee6
unverified

winglian commited on

DPO cleanup (#1126)
7523d1f
unverified

winglian plaguss HF staff commited on

support for explicit test_dataset definition for evals (#786)
cda52dc
unverified

winglian commited on

Falcon embeddings (#1149) [skip docker]
e799e08
unverified

winglian commited on

Vram fix attempt (#1164) [skip ci]
32580c1
unverified

winglian commited on

improve vram use w gradient checkpointing (#1167) [skip ci]
802f966
unverified

winglian commited on

Add mlflow callback for pushing config to mlflow artifacts (#1125)
b8e5603
unverified

JohanWork commited on

set fp16 to false if bf16, update bf16: auto in example YAMLs (#1122) [skip ci]
782b6a4
unverified

winglian Nanobit commited on

jupyter lab fixes (#1139) [skip ci]
eaaeefc
unverified

winglian commited on

Qwen2 (#1166)
f5a828a
unverified

winglian commited on

make sure the model config loader respects the model_revision too (#1160) [skip-ci]
fccb542
unverified

winglian commited on

Deprecate max packed sequence len (#1141)
2ce5c0d
unverified

winglian commited on

feat(dataset): add config to keep processed dataset in memory (#1152)
3db5f2f
unverified

Nanobit commited on

Multipack simplify for Mixtral (#1142)
6910e6a
unverified

winglian commited on

Add shifted sparse attention (#973) [skip-ci]
1d70f24
unverified

jrc joecummings winglian commited on

fix bf16 check when preprocessing data (#1140)
317fa25
unverified

winglian commited on

fix(preprocess): Make sure dataset not loaded from cache when using preprocess cli (#1136)
1e56b88
unverified

Nanobit commited on

Preprocess dataset size fix (#1131)
7570446
unverified

winglian commited on

Add `layers_to_transform` for `lora_config` (#1118)
8487b97
unverified

xzuyn commited on

Enable or disable bf16 support based on availability (#1116)
0865613
unverified

Simon Hällqvist commited on

keep gate in fp32 for 16 bit loras (#1105)
da97285
unverified

winglian commited on

add gptneox embeddings, fix phi2 inputs, also fix the casting (#1083)
78c5b19
unverified

winglian commited on

additional logging to get maximum token length of a sequence in the dataset (#1066) [skip ci]
2f2582e
unverified

winglian commited on

update sharegpt conversations when chatml chat template is set (#1075) [skip ci]
0ce1a65
unverified

winglian commited on

be more robust about checking embedding modules for lora finetunes (#1074) [skip ci]
0f10080
unverified

winglian commited on

Add: mlflow for experiment tracking (#1059) [skip ci]
090c24d
unverified

Johan Hansson winglian commited on

Cosine learning rate schedule - minimum learning rate (#1062)
04b978b
unverified

ricdomolm winglian commited on

fix: torch_dtype mistral default to fp32 (#1050)
c3e8165
unverified

Nanobit commited on

Efficiently get the length of the tokenized docs (#1063)
81d3845
unverified

ricdomolm winglian commited on

Phi2 rewrite (#1058)
732851f
unverified

winglian commited on

feature: better device mapping for large models (#918)
bdfefaf
unverified

kallewoof Karl-Johan Alm winglian commited on

RL/DPO (#935)
f243c21

winglian commited on

bump transformers and update attention class map name (#1023)
bcc78d8
unverified

winglian commited on

chore(config): clean up old log for Qwen (#1034)
74532dd
unverified

Nanobit commited on

Adds chat templates (#1022)
f8ae59b
unverified

mhenrichsen commited on

[WandB] Push axolotl config to top level wandb files (#1014)
4f4d638
unverified

hamel commited on

feat: expose bnb kwargs (#1018)
41353d2
unverified

Nanobit hamel commited on

remove landmark attn and xpos rope implementations (#1010)
70b46ca
unverified

winglian commited on

Feat: Warns to add to modules_to_save when adding tokens or switching special_tokens (#787)
1ffa386
unverified

Nanobit commited on

fix: switch to using the HuggingFace Transformers NEFT implementation (#941)
ef24342
unverified

kallewoof commited on

Fix Deepspeed loading (#950)
5ea3aa3
unverified

winglian commited on

Flash attn hotfix (#951)
f1f60cb
unverified

winglian commited on

new evals_per_epoch and saves_per_epoch to make things cleaner (#944)
5f79b82
unverified

winglian commited on

Mixtral official (#942)
7fabc4d
unverified

winglian commited on

Mixtral multipack (#928)
68b227a
unverified

winglian commited on

support for mamba (#915)
40a6362
unverified

winglian commited on

fix(tokenizer): handle fast tokenizer properly for bos/eos (#914)
fde091c
unverified

Nanobit commited on

feat: add check for quantized model (#913)
a581e9f
unverified

Nanobit winglian commited on