GELI: Adapted LLM (Llama-2-7b-chat-hf) for Social Conversations Via Facial Expressions
This is the official model card for the fine-tuned Llama-2 model that was produced from Global Reward to Local Rewards: Multimodal-Guided Decomposition for Improving Dialogue Agents presented at EMNLP 2024 (Oral). Check our the paper Here To load the model, please follow the following script. We rely on the Llama-2 tokenizer.
lm_tokenizer = LlamaTokenizer.from_pretrained("meta-llama/Llama-2-7b-chat-hf")
model = AutoModelForCausalLMWithValueHead("dondongwonlee/GELI")
model.eval()
CANDOR Attribution:
Overview: This language model was trained on the CANDOR provided by BetterUp, Inc., and is subject to their specific licensing terms. The dataset was obtained and used in accordance with the BetterUp, Inc. Terms of Use, and all users of this model must comply with these terms. For reference, the dataset's licensing details and restrictions are outlined below.
Dataset Attribution: This model is trained on data sourced from BetterUp, Inc. Any insights, analyses, or outputs derived from this model are, therefore, inherently influenced by this dataset. We acknowledge BetterUp, Inc. as the original creator and provider of this dataset. For further information on the dataset and licensing terms, please refer to [BetterUp, Inc.’s official documentation(https://betterup-data-requests.herokuapp.com/).
Terms of Use: This model and any derivative works or analyses thereof are subject to the following terms as required by BetterUp, Inc.:
Research Use Only: This model is intended exclusively for legitimate academic and/or scientific research. No outputs, analyses, reviews, or derivative works derived from this model may be used for commercial or for-profit purposes.
No Redistribution of Dataset: This model’s training data, whether in original or modified form, may not be re-published or re-shared without explicit permission from BetterUp, Inc.
No Identification or Tracking of Individuals: This model must not be used to personally identify, locate, or gather any information about individuals from the dataset, beyond the information explicitly provided in the data itself.
Protection of Personally Identifiable Information: If any personally identifiable information is inadvertently included in the dataset, it must not be used, analyzed, shared, or published in any form.
License and Usage Limitations: This model retains the license terms associated with the original BetterUp, Inc. dataset. Users of this model must agree to fully abide by these terms of use, as stipulated by BetterUp, Inc. If you have any questions regarding the dataset's license or its permissible uses, please contact BetterUp, Inc. or refer to their official licensing documentation.
Llama 2 Attribution:
Overview: This model is an adaptation of Meta's LLAMA 2. LLAMA 2 was originally developed and released by Meta, and you can find the official model and license details here. This version has been adapted or fine-tuned to be more suitable for social conversations.
Attribution to Meta: This model is based on LLAMA 2, developed by Meta and released under Meta's LLAMA 2 Community License Agreement. All rights to the original LLAMA 2 model architecture, parameters, and other core features are retained by Meta. This adapted model respects the guidelines and requirements specified in Meta’s license.
Intended Use and Limitations: While LLAMA 2 is licensed for research and certain commercial uses, please review and adhere to Meta's LLAMA 2 Community License Agreement to ensure compliant usage. Use cases outside the permissible scope of Meta's license are not supported by this adaptation.
License: This model retains the original license terms of LLAMA 2. If you have questions about the legal implications of using or sharing this model, please refer to Meta’s license or consult with a legal professional.