metadata
license: mit
pipeline_tag: text-generation
library_name: transformers
language:
- en
- am
- ar
- as
- az
- be
- bg
- bn
- br
- bs
- ca
- cs
- cy
- da
- de
- el
- eo
- es
- et
- eu
- fa
- ff
- fi
- fr
- fy
- ga
- gd
- gl
- gn
- gu
- ha
- he
- hi
- hr
- ht
- hu
- hy
- id
- ig
- is
- it
- ja
- jv
- ka
- kk
- km
- kn
- ko
- ku
- ky
- la
- lg
- li
- ln
- lo
- lt
- lv
- mg
- mk
- ml
- mn
- mr
- ms
- my
- ne
- nl
- 'no'
- ns
- om
- or
- pa
- pl
- ps
- pt
- qu
- rm
- ro
- ru
- sa
- si
- sc
- sd
- sk
- sl
- so
- sq
- sr
- ss
- su
- sv
- sw
- ta
- te
- th
- tl
- tn
- tr
- ug
- uk
- ur
- uz
- vi
- wo
- xh
- yi
- yo
- zu
datasets:
- ontocord/fineweb-permissive-multilingual-2m
- distily/c4_multilingual_1M
- data-silence/sumnews
- xu-song/cc100-samples
- badrex/llm-emoji-dataset
- fblgit/simple-math
- Gusarich/math-expressions-1m
- neuralwork/arxiver
- christopher/rosetta-code
- nampdn-ai/tiny-codes
- JeanKaddour/minipile
- NousResearch/hermes-function-calling-v1
- simplescaling/s1K-1.1
- mlabonne/open-perfectblend
- allenai/tulu-3-sft-mixture
- rombodawg/Everything_Instruct_Multilingual
- open-r1/OpenR1-Math-220k
- open-thoughts/OpenThoughts-114k
- cognitivecomputations/dolphin-r1
- simplescaling/s1K-1.1
tags:
- chat
- core
- base
- instruct
- reason
tangled-alpha-0.10-core
time python -B prepare_core_datasets.py
i=0, min_len=0, max_len=1073741824, block_size=1025, chunk_size=16400000, len(dataset)=10913927, len(dataset) * block_size=11186775175
Total number of tokens in the optimized dataset '../core-data-0-0-1073741824-1025-16000' is 11186775175
i=1, min_len=1025, max_len=2049, block_size=2049, chunk_size=16392000, len(dataset)=893465, len(dataset) * block_size=1830709785
Total number of tokens in the optimized dataset '../core-data-1-1025-2049-2049-8000' is 1830709785
i=2, min_len=2049, max_len=4097, block_size=4097, chunk_size=16388000, len(dataset)=375104, len(dataset) * block_size=1536801088
Total number of tokens in the optimized dataset '../core-data-2-2049-4097-4097-4000' is 1536801088
i=3, min_len=4097, max_len=8193, block_size=8193, chunk_size=16386000, len(dataset)=177522, len(dataset) * block_size=1454437746
Total number of tokens in the optimized dataset '../core-data-3-4097-8193-8193-2000' is 1454437746
i=4, min_len=8193, max_len=16385, block_size=16385, chunk_size=16385000, len(dataset)=77725, len(dataset) * block_size=1273524125
Total number of tokens in the optimized dataset '../core-data-4-8193-16385-16385-1000' is 1273524125
i=5, min_len=16385, max_len=32769, block_size=32769, chunk_size=16384500, len(dataset)=22931, len(dataset) * block_size=751425939
Total number of tokens in the optimized dataset '../core-data-5-16385-32769-32769-500' is 751425939
i=6, min_len=32769, max_len=65537, block_size=65537, chunk_size=16384250, len(dataset)=4988, len(dataset) * block_size=326898556
Total number of tokens in the optimized dataset '../core-data-6-32769-65537-65537-250' is 326898556
i=7, min_len=65537, max_len=131073, block_size=131073, chunk_size=16384125, len(dataset)=1137, len(dataset) * block_size=149030001
Total number of tokens in the optimized dataset '../core-data-7-65537-131073-131073-125' is 149030001
42G ../core-data-0-0-1073741824-1025-16000
6.9G ../core-data-1-1025-2049-2049-8000
5.8G ../core-data-2-2049-4097-4097-4000
5.5G ../core-data-3-4097-8193-8193-2000
4.8G ../core-data-4-8193-16385-16385-1000
2.9G ../core-data-5-16385-32769-32769-500
1.3G ../core-data-6-32769-65537-65537-250
573M ../core-data-7-65537-131073-131073-125
CUDA_VISIBLE_DEVICES=0 CUDA_LAUNCH_BLOCKING=0 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True litgpt pretrain --config pretrain_core_model_0.yaml
# ...
Backup wandb:
mv wandb wandb-pretrain-core-0
Copy config:
cp ../config-0.json ../out/pretrain-core-0/final/config.json
Chat with model:
CUDA_VISIBLE_DEVICES=0 CUDA_LAUNCH_BLOCKING=0 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True litgpt chat ../out/pretrain-core-0/final
CUDA_VISIBLE_DEVICES=0 CUDA_LAUNCH_BLOCKING=0 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True time litgpt evaluate --tasks 'leaderboard' --out_dir '../evaluate/pretrain-core-0/leaderboard/' --batch_size '4' --dtype 'bfloat16' '../out/pretrain-core-0/final'