wortegaLM / README.md
AlexWortega's picture
Update README.md
3aeabeb
|
raw
history blame
1.92 kB
metadata
license: apache-2.0
datasets:
  - IlyaGusev/rulm
inference:
  parameters:
    min_length: 20
    max_new_tokens: 250
    top_k: 50
    top_p: 0.9
    early_stopping: true
    no_repeat_ngram_size: 2
    use_cache: true
    repetition_penalty: 1.5
    length_penalty: 0.8
    num_beams: 2
language:
  - ru
library_name: transformers
pipeline_tag: text-generation
tags:
  - finance
  - code

WortegaLM 109m

Model Summary

Это GPTneo like модель обученная с нуля на сете в 95gb кода, хабра, пикабу, новостей(порядка 12B токенов) Она умеет решать примитивные задачи, не пригодна для ZS FS, но идеальна как модель для студенческих проектов

Quick Start






from transformers import AutoTokenizer, AutoModelForCausalLM,


tokenizer = AutoTokenizer.from_pretrained('AlexWortega/wortegaLM',padding_side='left')
device = 'cuda'
model = AutoModelForCausalLM.from_pretrained('AlexWortega/wortegaLM')
model.resize_token_embeddings(len(tokenizer))
model.to(device)



def generate_seqs(q,model, k=2):
    gen_kwargs = {
        "min_length": 20,
        "max_new_tokens": 100,
        "top_k": 50,
        "top_p": 0.7,
        "do_sample": True,  
        "early_stopping": True,
        "no_repeat_ngram_size": 2,
        "eos_token_id": tokenizer.eos_token_id,
        "pad_token_id": tokenizer.eos_token_id,
        "use_cache": True,
        "repetition_penalty": 1.5,  
        "length_penalty": 1.2,  
        "num_beams": 4,
        "num_return_sequences": k
    }
    
    t = tokenizer.encode(q, add_special_tokens=False, return_tensors='pt').to(device)
    g = model.generate(t, **gen_kwargs)
    generated_sequences = tokenizer.batch_decode(g, skip_special_tokens=False)
    
    return  generated_sequences