Unduh PaLM rlhf pytorch - Unduh kode sumber PaLM rlhf pytorch

posting blog chatgpt resmi

PaLM + RLHF - Pytorch (hapus)

Implementasi RLHF (Reinforcement Learning with Human Feedback) di atas arsitektur PaLM. Mungkin saya akan menambahkan fungsi pengambilan juga, ala RETRO

Jika Anda tertarik untuk mereplikasi sesuatu seperti ChatGPT secara terbuka, silakan pertimbangkan untuk bergabung dengan Laion

Penerus potensial: Optimasi Preferensi Langsung - semua kode dalam repo ini menjadi ~ kerugian entropi silang biner, <5 loc. Begitu banyak untuk model Reward dan PPO

Pertanyaan Umum

Apakah ini berisi model inferensi?

Tidak ada model yang terlatih. Ini hanyalah kapal dan peta keseluruhannya. Kita masih membutuhkan jutaan dolar komputasi + data untuk mencapai titik yang tepat dalam ruang parameter dimensi tinggi. Meski begitu, Anda memerlukan pelaut profesional (seperti Robin Rombach dari Stable Diffusion yang terkenal) untuk benar-benar memandu kapal melewati masa-masa sulit hingga saat itu.

Masyarakat

CarperAI telah mengerjakan kerangka kerja RLHF untuk model bahasa besar selama berbulan-bulan sebelum rilis ChatGPT.

Yannic Kilcher juga sedang mengerjakan implementasi open source

AI Coffeebreak dengan Letitia | Emporium Kode | Emporium Kode Bagian 2

Apresiasi

Stability.ai atas sponsor yang murah hati untuk mengerjakan penelitian kecerdasan buatan yang mutakhir
? Hugging Face dan CarperAI karena menulis postingan blog Illustrating Reinforcement Learning from Human Feedback (RLHF), dan CarperAI juga untuk perpustakaan akselerasi mereka
@kisseternity dan @taynoel84 untuk peninjauan kode dan menemukan bug
Enrico untuk mengintegrasikan Flash Attention dari Pytorch 2.0

Memasang

$ pip install palm-rlhf-pytorch

Penggunaan

Latih PaLM pertama, seperti transformator autoregresif lainnya

 import torch
from palm_rlhf_pytorch import PaLM

palm = PaLM (
    num_tokens = 20000 ,
    dim = 512 ,
    depth = 12 ,
    flash_attn = True # https://arxiv.org/abs/2205.14135
). cuda ()

seq = torch . randint ( 0 , 20000 , ( 1 , 2048 )). cuda ()

loss = palm ( seq , return_loss = True )
loss . backward ()

# after much training, you can now generate sequences

generated = palm . generate ( 2048 ) # (1, 2048)

Kemudian latih model penghargaan Anda, dengan masukan manusia yang dikurasi. Dalam makalah aslinya, mereka tidak bisa mendapatkan model hadiah untuk disempurnakan dari transformator yang telah dilatih sebelumnya tanpa overfitting, namun saya tetap memberikan opsi untuk menyempurnakan dengan LoRA , karena ini masih merupakan penelitian terbuka.

 import torch
from palm_rlhf_pytorch import PaLM , RewardModel

palm = PaLM (
    num_tokens = 20000 ,
    dim = 512 ,
    depth = 12 ,
    causal = False
)

reward_model = RewardModel (
    palm ,
    num_binned_output = 5 # say rating from 1 to 5
). cuda ()

# mock data

seq = torch . randint ( 0 , 20000 , ( 1 , 1024 )). cuda ()
prompt_mask = torch . zeros ( 1 , 1024 ). bool (). cuda () # which part of the sequence is prompt, which part is response
labels = torch . randint ( 0 , 5 , ( 1 ,)). cuda ()

# train

loss = reward_model ( seq , prompt_mask = prompt_mask , labels = labels )
loss . backward ()

# after much training

reward = reward_model ( seq , prompt_mask = prompt_mask )

Kemudian Anda akan meneruskan transformator dan model hadiah Anda ke RLHFTrainer

 import torch
from palm_rlhf_pytorch import PaLM , RewardModel , RLHFTrainer

# load your pretrained palm

palm = PaLM (
    num_tokens = 20000 ,
    dim = 512 ,
    depth = 12
). cuda ()

palm . load ( './path/to/pretrained/palm.pt' )

# load your pretrained reward model

reward_model = RewardModel (
    palm ,
    num_binned_output = 5
). cuda ()

reward_model . load ( './path/to/pretrained/reward_model.pt' )

# ready your list of prompts for reinforcement learning

prompts = torch . randint ( 0 , 256 , ( 50000 , 512 )). cuda () # 50k prompts

# pass it all to the trainer and train

trainer = RLHFTrainer (
    palm = palm ,
    reward_model = reward_model ,
    prompt_token_ids = prompts
)

trainer . train ( num_episodes = 50000 )

# then, if it succeeded...
# generate say 10 samples and use the reward model to return the best one

answer = trainer . generate ( 2048 , prompt = prompts [ 0 ], num_samples = 10 ) # (<= 2048,)

Semua yang harus dilakukan

Kutipan

 @article { Stiennon2020LearningTS ,
    title   = { Learning to summarize from human feedback } ,
    author  = { Nisan Stiennon and Long Ouyang and Jeff Wu and Daniel M. Ziegler and Ryan J. Lowe and Chelsea Voss and Alec Radford and Dario Amodei and Paul Christiano } ,
    journal = { ArXiv } ,
    year    = { 2020 } ,
    volume  = { abs/2009.01325 }
}

 @inproceedings { Chowdhery2022PaLMSL ,
    title   = { PaLM: Scaling Language Modeling with Pathways } ,
    author  = {Aakanksha Chowdhery and Sharan Narang and Jacob Devlin and Maarten Bosma and Gaurav Mishra and Adam Roberts and Paul Barham and Hyung Won Chung and Charles Sutton and Sebastian Gehrmann and Parker Schuh and Kensen Shi and Sasha Tsvyashchenko and Joshua Maynez and Abhishek Rao and Parker Barnes and Yi Tay and Noam M. Shazeer and Vinodkumar Prabhakaran and Emily Reif and Nan Du and Benton C. Hutchinson and Reiner Pope and James Bradbury and Jacob Austin and Michael Isard and Guy Gur-Ari and Pengcheng Yin and Toju Duke and Anselm Levskaya and Sanjay Ghemawat and Sunipa Dev and Henryk Michalewski and Xavier Garc{'i}a and Vedant Misra and Kevin Robinson and Liam Fedus and Denny Zhou and Daphne Ippolito and David Luan and Hyeontaek Lim and Barret Zoph and Alexander Spiridonov and Ryan Sepassi and David Dohan and Shivani Agrawal and Mark Omernick and Andrew M. Dai and Thanumalayan Sankaranarayana Pillai and Marie Pellat and Aitor Lewkowycz and Erica Oliveira Moreira and Rewon Child and Oleksandr Polozov and Katherine Lee and Zongwei Zhou and Xuezhi Wang and Brennan Saeta and Mark Diaz and Orhan Firat and Michele Catasta and Jason Wei and Kathleen S. Meier-Hellstern and Douglas Eck and Jeff Dean and Slav Petrov and Noah Fiedel},
    year    = { 2022 }
}

 @article { Hu2021LoRALA ,
    title   = { LoRA: Low-Rank Adaptation of Large Language Models } ,
    author  = { Edward J. Hu and Yelong Shen and Phillip Wallis and Zeyuan Allen-Zhu and Yuanzhi Li and Shean Wang and Weizhu Chen } ,
    journal = { ArXiv } ,
    year    = { 2021 } ,
    volume  = { abs/2106.09685 }
}

 @inproceedings { Sun2022ALT ,
    title     = { A Length-Extrapolatable Transformer } ,
    author    = { Yutao Sun and Li Dong and Barun Patra and Shuming Ma and Shaohan Huang and Alon Benhaim and Vishrav Chaudhary and Xia Song and Furu Wei } ,
    year      = { 2022 }
}

 @misc { gilmer2023intriguing
    title  = { Intriguing Properties of Transformer Training Instabilities } ,
    author = { Justin Gilmer, Andrea Schioppa, and Jeremy Cohen } ,
    year   = { 2023 } ,
    status = { to be published - one attention stabilization technique is circulating within Google Brain, being used by multiple teams }
}

 @inproceedings { dao2022flashattention ,
    title   = { Flash{A}ttention: Fast and Memory-Efficient Exact Attention with {IO}-Awareness } ,
    author  = { Dao, Tri and Fu, Daniel Y. and Ermon, Stefano and Rudra, Atri and R{'e}, Christopher } ,
    booktitle = { Advances in Neural Information Processing Systems } ,
    year    = { 2022 }
}

 @misc { Rubin2024 ,
    author  = { Ohad Rubin } ,
    url     = { https://medium.com/ @ ohadrubin/exploring-weight-decay-in-layer-normalization-challenges-and-a-reparameterization-solution-ad4d12c24950 }
}

 @inproceedings { Yuan2024FreePR ,
    title   = { Free Process Rewards without Process Labels } ,
    author  = { Lifan Yuan and Wendi Li and Huayu Chen and Ganqu Cui and Ning Ding and Kaiyan Zhang and Bowen Zhou and Zhiyuan Liu and Hao Peng } ,
    year    = { 2024 } ,
    url     = { https://api.semanticscholar.org/CorpusID:274445748 }
}