metadata

license: afl-3.0
language:
  - en
widget:
  - text: >-
      <prefix>You are a helpful chatbot name Stan</prefix><human>What's my
      name?<bot>
    example_title: Who am I?
  - text: >-
      <prefix>You are a helpful chatbot name Stan</prefix><human>How to make a
      campfire<bot>
    example_title: Tutorial
datasets:
  - Dahoas/instruct-synthetic-prompt-responses
  - openai/webgpt_comparisons
  - squad_v2
  - samsum
  - allenai/soda
  - xsum
  - multi_news
  - cnn_dailymail
  - scitldr
  - billsum
pipeline_tag: text-generation
tags:
  - finance

Supervised Finetuning demonstration

Models are finetuned on generated conversation curated from the Open Assistant.

This model was finetune for only 2,000 iterations, uploaded for ease of sharing only.

Mixing reward model with sampling

We can use reward model to rank the best answer using this example code:

import torch
from transformers import AutoModelForSequenceClassification
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("theblackcat102/galactica-1.3b-v2")
model = AutoModelForCausalLM.from_pretrained("theblackcat102/galactica-1.3b-v2").eval().half().cuda()

reward_name = "OpenAssistant/reward-model-deberta-v3-large"
rank_model, rank_tokenizer = AutoModelForSequenceClassification.from_pretrained(reward_name), AutoTokenizer.from_pretrained(reward_name)
rank_model = rank_model.eval().half().cuda()

questions = ["<prefix>You are a helpful chatbot call Agnes</prefix><human>How do I make a resume?<bot>"]
for question in questions:
    inputs = tokenizer(question, return_tensors="pt", padding=True).to(0)
    if 'token_type_ids' in inputs:
        inputs.pop('token_type_ids')
    outputs = model.generate(**inputs, do_sample=True,
        top_k=60,
        max_length=220,
        num_return_sequences=80, 
        early_stopping=True
    )
    print(question)

    results = []
    for i, beam_output in enumerate(outputs):
        output = tokenizer.decode(beam_output, truncate_before_pattern=[r"\n\n^#", "^'''", "\n\n\n"])
        question, answer = output.split('<bot>', maxsplit=1)
        answer = answer.split('</s>')[0].replace('<|endoftext|>', '').lstrip().split('<bot>')[0]
        rank_inputs = rank_tokenizer(question, answer, return_tensors="pt", padding=True, max_length=512, truncation=True).to(1)
        score = rank_model(**rank_inputs).logits[0].cpu().detach()
        results.append((answer, score, output))
    full_results[question] = results
    sorted_result = sorted(results, key=lambda x:x[1], reverse=True)
    total_scores += sorted_result[0][1].item()
    print('score',sorted_result[0][1].item())
    print('-----Best rank-----')
    print(sorted_result[0][0])
    print('-------------------')

This work is done under the Open Assistant initiative which democratize open source AI assistant. Feel free to join discord and contribute to github!

Thanks to BASIC lab for compute resource. BASIC Lab is an academic research lab which focuses in multi-modality learning and data mining domain.