Commit History
ADD: warning if hub_model_id ist set but not any save strategy (#1202)
af29d81
unverified
more checks and fixes for deepspeed and fsdp (#1208) [skip ci]
e923e62
unverified
precompute dpo logprobs setting and fixes (#1199) [skip ci]
33e1170
unverified
DPO fixes v2 (#1174)
59a31fe
unverified
Phi2 multipack (#1173)
814aee6
unverified
support for explicit test_dataset definition for evals (#786)
cda52dc
unverified
improve vram use w gradient checkpointing (#1167) [skip ci]
802f966
unverified
Qwen2 (#1166)
f5a828a
unverified
Deprecate max packed sequence len (#1141)
2ce5c0d
unverified
Multipack simplify for Mixtral (#1142)
6910e6a
unverified
fix bf16 check when preprocessing data (#1140)
317fa25
unverified
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
update sharegpt conversations when chatml chat template is set (#1075) [skip ci]
0ce1a65
unverified
be more robust about checking embedding modules for lora finetunes (#1074) [skip ci]
0f10080
unverified
feature: better device mapping for large models (#918)
bdfefaf
unverified
chore(config): clean up old log for Qwen (#1034)
74532dd
unverified
Feat: Warns to add to modules_to_save when adding tokens or switching special_tokens (#787)
1ffa386
unverified
fix: switch to using the HuggingFace Transformers NEFT implementation (#941)
ef24342
unverified
kallewoof
commited on
new evals_per_epoch and saves_per_epoch to make things cleaner (#944)
5f79b82
unverified
Support device_map=sequential & max_memory config parameters (#903)
992e742
unverified
Feat(wandb): Refactor to be more flexible (#767)
a1da39c
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
allow overriding of model_config parameters from the YML (#853)
1bc1186
unverified
simplify by removing duplicate base_model_config (#772)
2d8def6
unverified
Fix: Warn when fullfinetune without adapter (#770)
44c9d01
unverified
convert exponential notation lr to floats (#771)
ca84cca
unverified
Fix: eval table conflict with eval_sample_packing (#769)
9923b72
unverified
Implement fused modules (#747)
15d3a65
unverified
refactor to set eval_batch_size earlier if unset, so we can warn if mismatched (#662)
2642cae
unverified
Make dataset_processes configurable (#651)
9ec2077
unverified
Fix bug when using pretokenized datasets (#652)
590d603
unverified
ich
commited on
Feat: Add example for Mistral (#644)
eb41f76
unverified
Fix(cfg): Add validation for save_strategy and eval_strategy (#633)
383f88d
unverified
use fastchat conversations template (#578)
e7d3e2d
unverified
Feat: Add support for upstream FA2 (#626)
19a600a
unverified
Fix: Fail bf16 check when running on cpu during merge (#631)
cfbce02
unverified
add bf16 check (#587)
131afdb
unverified
make phi training work with Loras (#588)
62eaee7
unverified
E2e device cuda (#575)
2414673
unverified
Model parallel (#538)
f6060a6
unverified
Add training callback to send predictions to WandB table (#521)
5b67ea9
unverified
Fix pretraining with iterable/streaming Dataset (#556)
2f586d1
unverified
Jan Philipp Harries
Jan Philipp Harries
commited on