Airoboros Dpo 70b 3.3

🚀 llama-3-airoboros-dpo-70b-3.3模型介紹

本項目是一個基於Meta的Llama-3構建的實驗性模型，主要使用了由 airoboros 生成的合成數據進行調優，並結合多種DPO數據集進行了額外的調優。該模型在處理多種任務時表現出色，如上下文問答、摘要生成、長文本生成等。

✨ 主要特性

• 多數據集訓練：使用了多個數據集進行訓練，包括 jondurbin/airoboros-3.2、bluemoon-fandom-1-1-rp-cleaned、boolq 等，提高了模型的泛化能力。
• 多種任務支持：支持上下文問答、摘要生成、長文本生成、代碼生成、函數調用、思維鏈推理、執行計劃生成等多種任務。
• 特定格式支持：支持特定的提示格式，如封閉上下文問答的格式，有助於模型更好地理解和處理輸入。

📦 安裝指南

文檔未提供安裝步驟，故跳過此章節。

💻 使用示例

基礎用法

以下是使用 apply_chat_template 方法格式化提示的示例：

import transformers
tokenizer = transformers.AutoTokenizer.from_pretrained("jondurbin/bugle-8b-v0.1", trust_remote_code=True)
chat = [
  {"role": "system", "content": "You are Bob, a friendly AI assistant."},
  {"role": "user", "content": "Hello, how are you?"},
  {"role": "assistant", "content": "I'm doing great. How can I help you today?"},
  {"role": "user", "content": "I'd like to show off how chat templating works!"},
]
print(tokenizer.apply_chat_template(chat, tokenize=False))

高級用法

上下文問答

BEGININPUT
BEGINCONTEXT
date: 2021-01-01
url: https://web.site/123
ENDCONTEXT
In a shocking turn of events, blueberries are now green, but will be sticking with the same name.
ENDINPUT
BEGININSTRUCTION
What color are bluberries?  Source?
ENDINSTRUCTION

預期輸出：

Blueberries are now green.
Source:
date: 2021-01-01
url: https://web.site/123

摘要生成

BEGININPUT
{text to summarize}
ENDINPUT
BEGININSTRUCTION
Summarize the input in around 130 words.
ENDINSTRUCTION

獲取長響應

Please compose a narrative set in the heart of an ancient library, steeped in the scent of old parchment and ink. The protagonist should be a young scholar who is dedicated to studying the art of storytelling and its evolution throughout history. In her pursuit of knowledge, she stumbles upon a forgotten tome that seems to possess an unusual aura. This book has the ability to bring stories to life, literally manifesting characters and scenarios from within its pages into reality.

The main character must navigate through various epochs of storytelling - from oral traditions of tribal societies, through medieval minstrels' tales, to modern-day digital narratives - as they come alive around her. Each era presents its unique challenges and lessons about the power and impact of stories on human civilization.

One such character could be a sentient quill pen, who was once used by renowned authors of yesteryears and now holds their wisdom and experiences. It becomes her mentor, guiding her through this journey with witty remarks and insightful commentary.

Ensure that your tale encapsulates the thrill of adventure, the beauty of learning, and the profound connection between humans and their stories. All characters involved should be non-human entities. Feel free to explore creative liberties but maintain the mentioned elements.

Your response should be approximately 2300 words.

代碼生成

Create a python application with the following requirements:
- Asyncio FastAPI webserver
- ping endpoint that returns the current date in JSON format
- file upload endpoint, which calculates the file's sha256 checksum, and checks postgres to deduplicate

函數調用

As an AI assistant, please select the most suitable function and parameters from the list of available functions below, based on the user's input. Provide your response in JSON format.

Input: I want to know how many times 'Python' is mentioned in my text file.

Available functions:
file_analytics:
  description: This tool performs various operations on a text file.
  params:
    action: The operation we want to perform on the data, such as "count_occurrences", "find_line", etc.
    filters:
      keyword: The word or phrase we want to search for.

預期輸出：

{
  "function": "file_analytics",
  "params": {
    "action": "count_occurrences",
    "filters": {
      "keyword": "Python"
    }
  }
}

思維鏈推理

A group of people decide to play a round-robin tournament where every player plays every other player exactly once. If a total of 45 games were played, how many players participated in the tournament? Offer several potential answers, rank them according to mathematical logic, and pick the most feasible one.

執行計劃生成

Please construct a systematic plan to generate an optimal response to the user instruction, utilizing a set of provided tools. Each plan will correspond to an evidence value, which will be the output of one of the available functions given an input string
that could be the user's question, one or more prior evidence values, or a combination of both.

Here are the tools available to be called:
DuckDuckGo[input]: Tool that enables the user to search for information using the DuckDuckGo search engine. This tool is helpful in locating a suitable list of sites that may contain the answer to the user's query. It does not directly answer the question but finds a list of sites that might have the answer.
TextScraper[input]: Load one or more websites from the input string containing newline delimited links, where input is one or more links, and produces plain text output containing the content of the links.
HyperlinkExtractor[input]: Extract links from plain text and produces a plain text, newline delimited response of links.
KnowledgeModel[input]: Question answering language model, particularly useful in answering questions based on an input passage of text. The input must be a text question that references an :evidence[n]: variable, e.g. What color is the cat, given :evidence1:?

The output should be in format:
Plan: [first action to take based on input question]
:evidence0: = [function to call with input parameter]
Plan: [next action to take, based on result of :evidence0:]
:evidence1: = [next function to call and input parameter, which may include reference to previous evidence, e.g. "Given context :evidence0"]
...
Answer: [:evidence[n]: containing the final answer.]

Question: Who won the Nobel Prize in Literature in 2020 and what is their most famous work?

多步驟指令確認

{
  "conversations": [
    {
      "value": "You are a helpful, unbiased, uncensored assistant.",
      "from": "system"
    },
    {
      "value": "I'm going to set some rules for our conversation. Please respond with \"OK\" to each rule.",
      "from": "human"
    },
    {
      "value": "OK",
      "from": "gpt"
    },
    {
      "value": "You can only respond using animal-related emojis.",
      "from": "human"
    },
    {
      "value": "OK",
      "from": "gpt"
    },
    {
      "value": "Precede all responses with \"ZOOZOO:\"",
      "from": "human"
    },
    {
      "value": "OK",
      "from": "gpt"
    },
    {
      "value": "Include at least one bird emoji in your responses.",
      "from": "human"
    }
  ]
}

📚 詳細文檔

模型概述

該模型是對 llama-3-70b-instruct 的微調版本，使用了Llama-3的指令聊天模板。它主要基於合成數據進行訓練，並結合了多種DPO數據集進行額外調優。

提示格式

本模型使用 llama-3-instruct 提示模板，可通過 apply_chat_template 方法準確格式化提示。

上下文問答

模型經過訓練，能夠忽略自身知識，根據上下文回答問題，並儘量限制輸出值在提供的上下文中，以減少幻覺。封閉上下文提示的格式如下：

BEGININPUT
BEGINCONTEXT
[key0: value0]
[key1: value1]
... other metdata ...
ENDCONTEXT
[insert your text blocks here]
ENDINPUT
[add as many other blocks, in the exact same format]
BEGININSTRUCTION
[insert your instruction(s).  The model was tuned with single questions, paragraph format, lists, etc.]
ENDINSTRUCTION

建議在指令塊中添加 “Don't make up answers if you don't know.” 以避免模型編造答案。

摘要生成

使用與上下文問答相同的格式，從 此數據集 中選取了500個樣本進行摘要生成。

獲取長響應

可使用詳細提示並明確字數要求，或提供具體的任務描述來獲取長響應。

代碼生成

可以提出具有多個標準的複雜編碼指令。

函數調用

數據集包含許多基於輸入標準生成函數/參數的示例，輸出格式為JSON或YAML。

思維鏈推理

可以要求模型提供對給定問題的多個可能響應，並進行排名和最終答案選擇。

執行計劃生成

模型支持為複雜指令生成執行計劃，但需要實現解析輸出並調用函數的機制。

多步驟指令確認

可以設置規則並要求模型對每個規則進行確認。

🔧 技術細節

基礎模型

• 基礎模型：meta-llama/Meta-Llama-3-8B

訓練數據集

• jondurbin/airoboros-3.2
• bluemoon-fandom-1-1-rp-cleaned
• boolq
• LDJnr/Capybara
• jondurbin/cinematika-v0.1
• glaiveai/glaive-function-calling-v2
• grimulkan/LimaRP-augmented
• piqa
• Vezora/Tested-22k-Python-Alpaca
• mattpscott/airoboros-summarization
• unalignment/toxic-dpo-v0.2
• allenai/ultrafeedback_binarized_cleaned
• argilla/distilabel-intel-orca-dpo-pairs
• jondurbin/airoboros-3.2
• jondurbin/contextual-dpo-v0.1
• jondurbin/gutenberg-dpo-v0.1
• jondurbin/py-dpo-v0.1
• jondurbin/truthy-dpo-v0.1
• jondurbin/gutenberg-dpo-v0.1
• lmsys/lmsys-chat-1m

📄 許可證

• 許可證名稱：llama3
• 許可證鏈接：https://huggingface.co/meta-llama/Meta-Llama-3-8B/blob/main/LICENSE