CabraMixtral-8x7b / README.md
nicolasdec's picture
Update README.md
afdc737 verified
|
raw
history blame
4.07 kB
metadata
language:
  - pt
  - en
license: cc
tags:
  - text-generation-inference
  - transformers
  - mistral
  - mixtral
  - gguf
  - brazil
  - brasil
  - portuguese

BotBot Cabra Mixtral 8x7b

Esse modelo é um finetune do Mixtral 8x7b com o dataset Cabra 30k. Esse modelo é optimizado para português. Ele apresenta melhoria em varios benchmarks brasileiros em comparação com o modelo base.

Conheça os nossos outros modelos: Cabra.

dataset: Cabra 30k

Dataset interno para finetuning. Vamos lançar em breve.

Quantização / GGUF

Colocamos diversas versões (GGUF) quantanizadas no branch "quantanization".

Exemplo

<s> [INST] who is Elon Musk? [/INST]Elon Musk é um empreendedor, inventor e capitalista americano. Ele é o fundador, CEO e CTO da SpaceX, CEO da Neuralink e fundador do The Boring Company. Musk também é o proprietário do Twitter.</s>

Uso

O modelo é destinado, por agora, a fins de pesquisa. As áreas e tarefas de pesquisa possíveis incluem:

  • Pesquisa sobre modelos gerativos.
  • Investigação e compreensão das limitações e viéses de modelos gerativos.

Proibido para uso comercial. Somente Pesquisa.

Evals

Tasks Version Filter n-shot Metric Value Stderr assin2_rte 1.1 all 15 f1_macro 0.9095 ± 0.0041 all 15 acc 0.9097 ± 0.0041 assin2_sts 1.1 all 15 pearson 0.7763 ± 0.0068 all 15 mse 0.4610 ± N/A bluex 1.1 all 3 acc 0.6412 ± 0.0103 exam_id__UNICAMP_2021_2 3 acc 0.5882 ± 0.0397 exam_id__USP_2023 3 acc 0.7045 ± 0.0397 exam_id__UNICAMP_2020 3 acc 0.6545 ± 0.0371 exam_id__UNICAMP_2023 3 acc 0.7442 ± 0.0384 exam_id__UNICAMP_2018 3 acc 0.5926 ± 0.0386 exam_id__UNICAMP_2022 3 acc 0.6154 ± 0.0451 exam_id__USP_2019 3 acc 0.6250 ± 0.0442 exam_id__USP_2021 3 acc 0.6346 ± 0.0387 exam_id__UNICAMP_2021_1 3 acc 0.5000 ± 0.0427 exam_id__USP_2022 3 acc 0.6531 ± 0.0393 exam_id__UNICAMP_2024 3 acc 0.5556 ± 0.0428 exam_id__USP_2024 3 acc 0.8537 ± 0.0319 exam_id__USP_2018 3 acc 0.6296 ± 0.0379 exam_id__USP_2020 3 acc 0.6071 ± 0.0375 exam_id__UNICAMP_2019 3 acc 0.7000 ± 0.0374 enem 1.1 all 3 acc 0.7810 ± 0.0063 exam_id__2013 3 acc 0.7685 ± 0.0236 exam_id__2010 3 acc 0.8205 ± 0.0205 exam_id__2012 3 acc 0.8276 ± 0.0202 exam_id__2016_2 3 acc 0.7886 ± 0.0213 exam_id__2016 3 acc 0.8017 ± 0.0209 exam_id__2022 3 acc 0.6541 ± 0.0238 exam_id__2009 3 acc 0.8087 ± 0.0212 exam_id__2015 3 acc 0.8067 ± 0.0208 exam_id__2017 3 acc 0.7759 ± 0.0223 exam_id__2014 3 acc 0.7798 ± 0.0230 exam_id__2023 3 acc 0.7037 ± 0.0228 exam_id__2011 3 acc 0.8632 ± 0.0183 faquad_nli 1.1 all 15 f1_macro 0.7893 ± 0.0137 all 15 acc 0.8554 ± 0.0097 hatebr_offensive_binary 1.0 all 25 f1_macro 0.7800 ± 0.0080 all 25 acc 0.7879 ± 0.0077 oab_exams 1.5 all 3 acc 0.5549 ± 0.0061 exam_id__2014-14 3 acc 0.6375 ± 0.0311 exam_id__2010-01 3 acc 0.4471 ± 0.0312 exam_id__2016-20a 3 acc 0.5000 ± 0.0323 exam_id__2013-10 3 acc 0.5750 ± 0.0318 exam_id__2010-02 3 acc 0.5500 ± 0.0288 exam_id__2011-04 3 acc 0.5750 ± 0.0319 exam_id__2013-12 3 acc 0.6000 ± 0.0316 exam_id__2017-22 3 acc 0.6000 ± 0.0316 exam_id__2015-16 3 acc 0.5625 ± 0.0320 exam_id__2012-09 3 acc 0.4416 ± 0.0326 exam_id__2016-19 3 acc 0.5641 ± 0.0326 exam_id__2017-23 3 acc 0.5500 ± 0.0321 exam_id__2013-11 3 acc 0.5500 ± 0.0321 exam_id__2012-08 3 acc 0.5500 ± 0.0323 exam_id__2011-03 3 acc 0.5152 ± 0.0290 exam_id__2015-17 3 acc 0.6538 ± 0.0310 exam_id__2015-18 3 acc 0.6125 ± 0.0314 exam_id__2014-15 3 acc 0.6538 ± 0.0310 exam_id__2018-25 3 acc 0.5750 ± 0.0320 exam_id__2012-06a 3 acc 0.5750 ± 0.0318 exam_id__2017-24 3 acc 0.4875 ± 0.0323 exam_id__2011-05 3 acc 0.5250 ± 0.0323 exam_id__2012-06 3 acc 0.5750 ± 0.0319 exam_id__2016-21 3 acc 0.4875 ± 0.0322 exam_id__2012-07 3 acc 0.5000 ± 0.0323 exam_id__2016-20 3 acc 0.5500 ± 0.0322 exam_id__2014-13 3 acc 0.5875 ± 0.0316 portuguese_hate_speech_binary 1.0 all 25 f1_macro 0.6954 ± 0.0114 all 25 acc 0.7086 ± 0.0110