qwerrwe / src

Commit History

log rank too (#527)
245c5c4
unverified

winglian commited on

misc fixes/improvements (#513)
a546ca2
unverified

winglian commited on

Add support for GPTQ using native transformers/peft (#468)
3355706
unverified

winglian commited on

Merge pull request #520 from bdashore3/sharegpt-fixes
daa4fac
unverified

mhenrichsen commited on

reorg a bit
fc8766e

tmm1 commited on

use flash_attn rmsnorm when available (#526)
72a6fe1
unverified

tmm1 commited on

use flash_attn xentropy when available (#525)
5fe30b1
unverified

tmm1 commited on

move is_llama_derived_model into normalize_config (#524)
44454ae
unverified

tmm1 commited on

No gather single gpu (#523)
09f1543
unverified

winglian commited on

Prompters: ShareGPT: Allow for custom system prompts
995557b

kingbri commited on

fix: bad dtype for full finetune (#504)
1991946
unverified

Maxime winglian commited on

log supervised token count (#448)
7710e81
unverified

winglian commited on

Debug tokenization output: Add ability to output text only (no tokens), and/or specify num samples to see (#511)
48434be
unverified

Tom Jobbins commited on

Added advanced DDP args (#515)
396a7a7
unverified

Jan Philipp Harries Jan Philipp Harries commited on

split train from other cli options (#503)
b21e4a2
unverified

winglian commited on

Changed Bench Eval to report metrics correctly by split. Added total accuracy and renamed previously used bench_accuracy to bench_average_accuracy. (#512)
42f9642
unverified

Alpay Ariyak commited on

drop empty tokenized rows too (#509)
c56b450
unverified

winglian commited on

add eval benchmark callback (#441)
7657632
unverified

winglian commited on

support for datasets with multiple names (#480)
5ac3392
unverified

winglian commited on

Refactor train cfg cli (#499)
125cccb
unverified

winglian commited on

use math.ceil instead of round /cc #498
fd55bc8

tmm1 commited on

pad_to_worst_case_seq_len boolean, for testing memory limits (#498)
8e197f6
unverified

Birch-san tmm1 commited on

simplify linear layer locator
267b7b2

tmm1 commited on

fsdp requires params be the same type too (#493)
98bf76e
unverified

winglian commited on

Fix(tokenizer): Make sure to add pad for CodeLlamaTokenizer (#489)
4c37bd0
unverified

Nanobit commited on

fix condition and add logging
3a011ea

tmm1 commited on

Merge branch 'main' into patch-4
1f613e5

tmm1 commited on

rename var and reformat
f319b0b

tmm1 commited on

Update src/axolotl/utils/models.py
7fd662d
unverified

Maxime tmm1 commited on

Update src/axolotl/utils/models.py
9e69968
unverified

Maxime tmm1 commited on

let transformers handle adamw_bnb_8bit
868530c

tmm1 commited on

ignore: address pr review
d03887f
unverified

Maxime commited on

ignore: linter
a184549
unverified

Maxime commited on

fix: finetune model inference needs the dtype fix to work with flash-attn
f311df9
unverified

Maxime commited on

fix checkpints on multigpu (#481)
31f3e71
unverified

winglian commited on

fix types w lora (#478)
0b7ba57
unverified

winglian commited on

Fix(tokenizer): Fix condition to add pad token (#477)
71bd062
unverified

Nanobit commited on

improve llama pad token handling (#475)
cb9797e
unverified

winglian commited on

ReLoRA implementation (with quantization) (#322)
bde3c5a
unverified

chargoddard winglian commited on

workaround so training doesn't hang when packed dataloader batches aren't even (#461)
c69faee
unverified

winglian commited on

feat: add Metharme prompt strategy (#446)
f474650
unverified

TearGosling Nanobit commited on

recast loralayer, norm, lmhead + embed token weights per original qlora (#393)
96deb6b
unverified

winglian commited on

always drop samples that are too long (#452)
50682a3
unverified

winglian commited on

set env var for FSDP layer to wrap (#453)
5a1985b
unverified

winglian commited on

is_causal fix for evals?
fbf49a4

winglian commited on

add missing positional arg (#450)
58cf7e7
unverified

winglian commited on

fix evals (#447)
ee26281
unverified

winglian commited on

gracefully handle empty input (#442)
9d629d8
unverified

winglian commited on

support user defined prompters, pretokenized datasets in config, local parquet, local arrow files (#348)
d2e7f27
unverified

winglian commited on