craffel's picture
download
raw
14.4 kB
dump_dir: /fsx/craffel/lingua_logs/safe_v2_2nodes
name: safe_v2_2nodes
steps: 100000
probe_freq: null
seed: 777
grad_acc_steps: 4
optim:
lr: 0.001
weight_decay: 0.1
epsilon: 1.0e-08
scheduler: cosine
warmup: 2000
lr_min_ratio: 1.0e-06
clip: 1.0
distributed:
fsdp_type: full_shard
model_dtype: bf16
matmul_allow_tf32: false
selective_activation_checkpointing: false
tp_size: 1
compile: true
model:
dim: 2048
n_layers: 25
n_heads: 16
weight_tying: false
vocab_size: 122554
data:
root_dir: /scratch/craffel/lingua/data/flexitok/
sources:
fw_edu: 0.4
dan_Latn: 0.0216582869670702
swe_Latn: 0.0216359765418466
vie_Latn: 0.0197485510268674
hun_Latn: 0.0247194573562308
fas_Arab: 0.0205634624231076
tur_Latn: 0.0235455794841729
ces_Latn: 0.0248024455266208
arb_Arab: 0.0234323706569333
ell_Grek: 0.0233670886888026
ind_Latn: 0.0269322054593488
nld_Latn: 0.0277796326621489
pol_Latn: 0.0294120104572311
por_Latn: 0.0301413168306825
ita_Latn: 0.0324056371021865
jpn_Jpan: 0.03553104151369
fra_Latn: 0.0381835560678536
spa_Latn: 0.0387222793083669
deu_Latn: 0.0419925340453022
cmn_Hani: 0.0454067521384114
rus_Cyrl: 0.0500198157431261
batch_size: 4
prefetch_size: 1024
seq_len: 4096
n_views: 2
load_async: true
add_eos: true
tokenizer:
name: supertokenizer
path: meta-llama/Llama-3.2-1B
seed: 42
superset_code_name: safe_v2
n_words: 122554
tokenizers:
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_arb_Arab_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_ces_Latn_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_ltr_cmn_Hani_8000_v2
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_dan_Latn_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_deu_Latn_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_ell_Grek_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_fas_Arab_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_fra_Latn_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_fw_edu_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_hun_Latn_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_ind_Latn_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_ita_Latn_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_ltr_jpn_Jpan_8000_v2
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_nld_Latn_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_pol_Latn_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_por_Latn_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_rus_Cyrl_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_spa_Latn_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_swe_Latn_8000
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_ltr_tur_Latn_8000_v2
load_supermapping: false
- name: huggingface
path: flexitok/reindexed-flexitok-bpe_vie_Latn_8000
load_supermapping: false
routing:
suitable_tokenizer_probability: 1.0
source_to_tokenizer:
arb_Arab: flexitok/reindexed-flexitok-bpe_arb_Arab_8000
ces_Latn: flexitok/reindexed-flexitok-bpe_ces_Latn_8000
cmn_Hani: flexitok/reindexed-flexitok-bpe_ltr_cmn_Hani_8000
dan_Latn: flexitok/reindexed-flexitok-bpe_dan_Latn_8000
deu_Latn: flexitok/reindexed-flexitok-bpe_deu_Latn_8000
ell_Grek: flexitok/reindexed-flexitok-bpe_ell_Grek_8000
fas_Arab: flexitok/reindexed-flexitok-bpe_fas_Arab_8000
fra_Latn: flexitok/reindexed-flexitok-bpe_fra_Latn_8000
fw_edu: flexitok/reindexed-flexitok-bpe_fw_edu_8000
hun_Latn: flexitok/reindexed-flexitok-bpe_hun_Latn_8000
ind_Latn: flexitok/reindexed-flexitok-bpe_ind_Latn_8000
ita_Latn: flexitok/reindexed-flexitok-bpe_ita_Latn_8000
jpn_Jpan: flexitok/reindexed-flexitok-bpe_ltr_jpn_Jpan_8000_v2
nld_Latn: flexitok/reindexed-flexitok-bpe_nld_Latn_8000
pol_Latn: flexitok/reindexed-flexitok-bpe_pol_Latn_8000
por_Latn: flexitok/reindexed-flexitok-bpe_por_Latn_8000
rus_Cyrl: flexitok/reindexed-flexitok-bpe_rus_Cyrl_8000
spa_Latn: flexitok/reindexed-flexitok-bpe_spa_Latn_8000
swe_Latn: flexitok/reindexed-flexitok-bpe_swe_Latn_8000
tur_Latn: flexitok/reindexed-flexitok-bpe_ltr_tur_Latn_8000_v2
vie_Latn: flexitok/reindexed-flexitok-bpe_vie_Latn_8000
profiling:
run: true
mem_warmup: 0
mem_steps: 4
profile_warmup: 100
profile_steps: 4
checkpoint:
path: /fsx/craffel/lingua_logs/checkpoints/safe_v2_2nodes
init_ckpt_path: null
dump:
every: 2500
keep: -1
eval:
every: 2500
keep: -1
logging:
freq: 1
wandb:
project: MoTo-Training
entity: raffel-reports
async_eval_gpus: null
eval:
harness:
tasks:
- hellaswag
- piqa
- arc_easy
- arc_challenge
- xnli_ar
- xnli_zh
- xnli_de
- xnli_el
- xnli_en
- xnli_es
- xnli_fr
- xnli_ru
- xnli_tr
- xnli_vi
- toksuite_english
- toksuite_chinese
- toksuite_turkish
- toksuite_italian
- toksuite_farsi
- toksuite_math
- toksuite_stem
- xcopa_id
- xcopa_it
- xcopa_tr
- xcopa_vi
- xcopa_zh
- xstorycloze_ar
- xstorycloze_en
- xstorycloze_es
- xstorycloze_id
- xstorycloze_ru
- xstorycloze_zh
- paws_de
- paws_en
- paws_es
- paws_fr
- paws_ja
- paws_zh
- hellaswag_ar
- hellaswag_da
- hellaswag_de
- hellaswag_fr
- hellaswag_hu
- hellaswag_id
- hellaswag_it
- hellaswag_nl
- hellaswag_pt
- hellaswag_ru
- hellaswag_sv
- hellaswag_vi
- hellaswag_es
generator:
max_tokens: 16384
dtype: bf16
add_bos: false
routing:
suitable_tokenizer_probability: 1.0
task_to_tokenizer:
arc_easy: flexitok/reindexed-flexitok-bpe_fw_edu_8000
arc_challenge: flexitok/reindexed-flexitok-bpe_fw_edu_8000
hellaswag: flexitok/reindexed-flexitok-bpe_fw_edu_8000
piqa: flexitok/reindexed-flexitok-bpe_fw_edu_8000
include_base_44_arabic: flexitok/reindexed-flexitok-bpe_arb_Arab_8000
include_base_44_chinese: flexitok/reindexed-flexitok-bpe_ltr_cmn_Hani_8000
include_base_44_german: flexitok/reindexed-flexitok-bpe_deu_Latn_8000
include_base_44_greek: flexitok/reindexed-flexitok-bpe_ell_Grek_8000
include_base_44_persian: flexitok/reindexed-flexitok-bpe_fas_Arab_8000
include_base_44_french: flexitok/reindexed-flexitok-bpe_fra_Latn_8000
include_base_44_hungarian: flexitok/reindexed-flexitok-bpe_hun_Latn_8000
include_base_44_indonesian: flexitok/reindexed-flexitok-bpe_ind_Latn_8000
include_base_44_italian: flexitok/reindexed-flexitok-bpe_ita_Latn_8000
include_base_44_japanese: flexitok/reindexed-flexitok-bpe_ltr_jpn_Jpan_8000_v2
include_base_44_dutch: flexitok/reindexed-flexitok-bpe_nld_Latn_8000
include_base_44_portuguese: flexitok/reindexed-flexitok-bpe_por_Latn_8000
include_base_44_russian: flexitok/reindexed-flexitok-bpe_rus_Cyrl_8000
include_base_44_spanish: flexitok/reindexed-flexitok-bpe_spa_Latn_8000
include_base_44_turkish: flexitok/reindexed-flexitok-bpe_ltr_tur_Latn_8000_v2
include_base_44_vietnamese: flexitok/reindexed-flexitok-bpe_vie_Latn_8000
belebele_arb_Arab: flexitok/reindexed-flexitok-bpe_arb_Arab_8000
belebele_ces_Latn: flexitok/reindexed-flexitok-bpe_ces_Latn_8000
belebele_zho_Hans: flexitok/reindexed-flexitok-bpe_ltr_cmn_Hani_8000
belebele_dan_Latn: flexitok/reindexed-flexitok-bpe_dan_Latn_8000
belebele_deu_Latn: flexitok/reindexed-flexitok-bpe_deu_Latn_8000
belebele_ell_Grek: flexitok/reindexed-flexitok-bpe_ell_Grek_8000
belebele_pes_Arab: flexitok/reindexed-flexitok-bpe_fas_Arab_8000
belebele_fra_Latn: flexitok/reindexed-flexitok-bpe_fra_Latn_8000
belebele_hun_Latn: flexitok/reindexed-flexitok-bpe_hun_Latn_8000
belebele_ind_Latn: flexitok/reindexed-flexitok-bpe_ind_Latn_8000
belebele_ita_Latn: flexitok/reindexed-flexitok-bpe_ita_Latn_8000
belebele_jpn_Jpan: flexitok/reindexed-flexitok-bpe_ltr_jpn_Jpan_8000_v2
belebele_nld_Latn: flexitok/reindexed-flexitok-bpe_nld_Latn_8000
belebele_pol_Latn: flexitok/reindexed-flexitok-bpe_pol_Latn_8000
belebele_por_Latn: flexitok/reindexed-flexitok-bpe_por_Latn_8000
belebele_rus_Cyrl: flexitok/reindexed-flexitok-bpe_rus_Cyrl_8000
belebele_spa_Latn: flexitok/reindexed-flexitok-bpe_spa_Latn_8000
belebele_swe_Latn: flexitok/reindexed-flexitok-bpe_swe_Latn_8000
belebele_tur_Latn: flexitok/reindexed-flexitok-bpe_ltr_tur_Latn_8000_v2
belebele_vie_Latn: flexitok/reindexed-flexitok-bpe_vie_Latn_8000
belebele_eng_Latn: flexitok/reindexed-flexitok-bpe_fw_edu_8000
xnli_ar: flexitok/reindexed-flexitok-bpe_arb_Arab_8000
xnli_zh: flexitok/reindexed-flexitok-bpe_ltr_cmn_Hani_8000
xnli_de: flexitok/reindexed-flexitok-bpe_deu_Latn_8000
xnli_el: flexitok/reindexed-flexitok-bpe_ell_Grek_8000
xnli_en: flexitok/reindexed-flexitok-bpe_fw_edu_8000
xnli_es: flexitok/reindexed-flexitok-bpe_spa_Latn_8000
xnli_fr: flexitok/reindexed-flexitok-bpe_fra_Latn_8000
xnli_ru: flexitok/reindexed-flexitok-bpe_rus_Cyrl_8000
xnli_tr: flexitok/reindexed-flexitok-bpe_ltr_tur_Latn_8000_v2
xnli_vi: flexitok/reindexed-flexitok-bpe_vie_Latn_8000
regex:toksuite_turkish*: flexitok/reindexed-flexitok-bpe_ltr_tur_Latn_8000_v2
regex:toksuite_english*: flexitok/reindexed-flexitok-bpe_fw_edu_8000
regex:toksuite_farsi*: flexitok/reindexed-flexitok-bpe_fas_Arab_8000
regex:toksuite_italian*: flexitok/reindexed-flexitok-bpe_ita_Latn_8000
regex:toksuite_chinese*: flexitok/reindexed-flexitok-bpe_ltr_cmn_Hani_8000
regex:toksuite_math*: flexitok/reindexed-flexitok-bpe_fw_edu_8000
regex:toksuite_stem*: flexitok/reindexed-flexitok-bpe_fw_edu_8000
flores200_arb_Arab: flexitok/reindexed-flexitok-bpe_arb_Arab_8000
flores200_ces_Latn: flexitok/reindexed-flexitok-bpe_ces_Latn_8000
flores200_cmn_Hani: flexitok/reindexed-flexitok-bpe_ltr_cmn_Hani_8000
flores200_dan_Latn: flexitok/reindexed-flexitok-bpe_dan_Latn_8000
flores200_deu_Latn: flexitok/reindexed-flexitok-bpe_deu_Latn_8000
flores200_ell_Grek: flexitok/reindexed-flexitok-bpe_ell_Grek_8000
flores200_fas_Arab: flexitok/reindexed-flexitok-bpe_fas_Arab_8000
flores200_fra_Latn: flexitok/reindexed-flexitok-bpe_fra_Latn_8000
flores200_fw_edu: flexitok/reindexed-flexitok-bpe_fw_edu_8000
flores200_eng_Latn: flexitok/reindexed-flexitok-bpe_fw_edu_8000
flores200_hun_Latn: flexitok/reindexed-flexitok-bpe_hun_Latn_8000
flores200_ind_Latn: flexitok/reindexed-flexitok-bpe_ind_Latn_8000
flores200_ita_Latn: flexitok/reindexed-flexitok-bpe_ita_Latn_8000
flores200_jpn_Jpan: flexitok/reindexed-flexitok-bpe_ltr_jpn_Jpan_8000_v2
flores200_nld_Latn: flexitok/reindexed-flexitok-bpe_nld_Latn_8000
flores200_pol_Latn: flexitok/reindexed-flexitok-bpe_pol_Latn_8000
flores200_por_Latn: flexitok/reindexed-flexitok-bpe_por_Latn_8000
flores200_rus_Cyrl: flexitok/reindexed-flexitok-bpe_rus_Cyrl_8000
flores200_spa_Latn: flexitok/reindexed-flexitok-bpe_spa_Latn_8000
flores200_swe_Latn: flexitok/reindexed-flexitok-bpe_swe_Latn_8000
flores200_tur_Latn: flexitok/reindexed-flexitok-bpe_ltr_tur_Latn_8000_v2
flores200_vie_Latn: flexitok/reindexed-flexitok-bpe_vie_Latn_8000
xcopa_id: flexitok/reindexed-flexitok-bpe_ind_Latn_8000
xcopa_it: flexitok/reindexed-flexitok-bpe_ita_Latn_8000
xcopa_tr: flexitok/reindexed-flexitok-bpe_ltr_tur_Latn_8000_v2
xcopa_vi: flexitok/reindexed-flexitok-bpe_vie_Latn_8000
xcopa_zh: flexitok/reindexed-flexitok-bpe_ltr_cmn_Hani_8000
xstorycloze_ar: flexitok/reindexed-flexitok-bpe_arb_Arab_8000
xstorycloze_en: flexitok/reindexed-flexitok-bpe_fw_edu_8000
xstorycloze_es: flexitok/reindexed-flexitok-bpe_spa_Latn_8000
xstorycloze_id: flexitok/reindexed-flexitok-bpe_ind_Latn_8000
xstorycloze_ru: flexitok/reindexed-flexitok-bpe_rus_Cyrl_8000
xstorycloze_zh: flexitok/reindexed-flexitok-bpe_ltr_cmn_Hani_8000
paws_de: flexitok/reindexed-flexitok-bpe_deu_Latn_8000
paws_en: flexitok/reindexed-flexitok-bpe_fw_edu_8000
paws_es: flexitok/reindexed-flexitok-bpe_spa_Latn_8000
paws_fr: flexitok/reindexed-flexitok-bpe_fra_Latn_8000
paws_ja: flexitok/reindexed-flexitok-bpe_ltr_jpn_Jpan_8000_v2
paws_zh: flexitok/reindexed-flexitok-bpe_ltr_cmn_Hani_8000
hellaswag_ar: flexitok/reindexed-flexitok-bpe_arb_Arab_8000
hellaswag_da: flexitok/reindexed-flexitok-bpe_dan_Latn_8000
hellaswag_de: flexitok/reindexed-flexitok-bpe_deu_Latn_8000
hellaswag_fr: flexitok/reindexed-flexitok-bpe_fra_Latn_8000
hellaswag_hu: flexitok/reindexed-flexitok-bpe_hun_Latn_8000
hellaswag_id: flexitok/reindexed-flexitok-bpe_ind_Latn_8000
hellaswag_it: flexitok/reindexed-flexitok-bpe_ita_Latn_8000
hellaswag_nl: flexitok/reindexed-flexitok-bpe_nld_Latn_8000
hellaswag_pt: flexitok/reindexed-flexitok-bpe_por_Latn_8000
hellaswag_ru: flexitok/reindexed-flexitok-bpe_rus_Cyrl_8000
hellaswag_sv: flexitok/reindexed-flexitok-bpe_swe_Latn_8000
hellaswag_vi: flexitok/reindexed-flexitok-bpe_vie_Latn_8000
hellaswag_es: flexitok/reindexed-flexitok-bpe_spa_Latn_8000

Xet Storage Details

Size:
14.4 kB
·
Xet hash:
797cd25554bd0a0abc7f169ab0c5b1585e80337c783d780cda92d231c4ab6ee7

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.