metadata

license: llama2
tags:
  - facebook
  - meta
  - pytorch
  - llama
  - llama-2
  - Storywriter
model_type: llama
model-index:
  - name: GOAT-70B-Storytelling
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: AI2 Reasoning Challenge (25-Shot)
          type: ai2_arc
          config: ARC-Challenge
          split: test
          args:
            num_few_shot: 25
        metrics:
          - type: acc_norm
            value: 68.77
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=GOAT-AI/GOAT-70B-Storytelling
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: HellaSwag (10-Shot)
          type: hellaswag
          split: validation
          args:
            num_few_shot: 10
        metrics:
          - type: acc_norm
            value: 87.74
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=GOAT-AI/GOAT-70B-Storytelling
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU (5-Shot)
          type: cais/mmlu
          config: all
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 69.92
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=GOAT-AI/GOAT-70B-Storytelling
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: TruthfulQA (0-shot)
          type: truthful_qa
          config: multiple_choice
          split: validation
          args:
            num_few_shot: 0
        metrics:
          - type: mc2
            value: 53.53
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=GOAT-AI/GOAT-70B-Storytelling
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: Winogrande (5-shot)
          type: winogrande
          config: winogrande_xl
          split: validation
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 83.5
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=GOAT-AI/GOAT-70B-Storytelling
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GSM8k (5-shot)
          type: gsm8k
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 40.79
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=GOAT-AI/GOAT-70B-Storytelling
          name: Open LLM Leaderboard

GOAT-70B-Storytelling model

GOAT-70B-Storytelling model trained by GOAT.AI lab as a core model for an autonomous story-writing agent.

GOAT-Storytelling-Agent

This agent facilitates the generation of high-quality, cohesive, and captivating narratives, including stories and books. It achieves this by utilizing inputs such as plot outlines, character profiles, their interrelationships, and other relevant details. Examples are provided below.

Model description

Base Architecture: LLaMA 2 70B
License: llama2
Context window length: 4096 tokens

Training details

Training was performed on a GPU cluster of 64xH100s. FSDP ZeRO-3 sharding is employed for efficient training. We instruction finetune on a dataset of 18K examples for one epoch with batch size of 336, AdamW optimizer with learning rate 1e-5.

Learn more

Blogpost: GOAT-Storytelling: Arbitrarily Long Story Writing Agent
GitHub: here
Generated examples: here

Uses

The main purpose of GOAT-70B-Storytelling is to generate books, novels, movie scripts and etc. as an agent in coping with our GOAT-Storytelling-Agent. It is specifically designed for storywriters.

Usage

Usage can be either self-hosted via transformers or used with Spaces

import torch

from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = "GOAT-AI/GOAT-70B-Storytelling"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16
)

Currently, we support LLM endpoint generation, where you need to send a post request to the generation endpoint (we recommend using Text Generation Inference by HuggingFace).

Here is how you can utilize the model via GOAT-Storytelling-Agent:

from goat_storytelling_agent.storytelling_agent import StoryAgent

backend_uri = # Text generation endpoint
writer = StoryAgent(backend_uri, form='novel')
novel_scenes = writer.generate_story('treasure hunt in a jungle')

License

GOAT-70B-Storytelling model is based on Meta's LLaMA-2-70b-hf, and using own datasets.

GOAT-70B-Storytelling model weights are available under LLAMA-2 license.

Risks and Biases

GOAT-70B-Storytelling model can produce factually incorrect output and should not be relied on to deliver factually accurate information. Therefore, the GOAT-70B-Storytelling model could possibly generate wrong, biased, or otherwise offensive outputs.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	67.38
AI2 Reasoning Challenge (25-Shot)	68.77
HellaSwag (10-Shot)	87.74
MMLU (5-Shot)	69.92
TruthfulQA (0-shot)	53.53
Winogrande (5-shot)	83.50
GSM8k (5-shot)	40.79