--- license: apache-2.0 language: - en datasets: - snorkelai/snorkel-curated-instruction-tuning inference: parameters: temperature: 0.7 top_p: 0.7 top_k: 50 max_new_tokens: 512 pipeline_tag: text-generation --- # RedPajama-7B-Chat-Curated The model is created by fine-tuning the RedPajama Base model on [snorkelai/snorkel-curated-instruction-tuning](https://huggingface.co/datasets/snorkelai/snorkel-curated-instruction-tuning) to enhance chatting ability further with high-quality instruction-response pairs. For a more comprehensive understanding of our methodology, please visit our blog - [How we built a better GenAI with programmatic data development](snorkel.ai/how-we-built-a-better-genai-with-programmatic-data-development). - Chat-Curated Version: [snorkelai/RedPajama-7B-Chat-Curated](https://huggingface.co/snorkelai/RedPajama-7B-Chat-Curated) - Instruction Tuning Dataset: [snorkelai/snorkel-curated-instruction-tuning](https://huggingface.co/datasets/snorkelai/snorkel-curated-instruction-tuning) - Base Model: [RedPajama-INCITE-7B-Base](https://huggingface.co/togethercomputer/RedPajama-INCITE-7B-Base) ## Model Details - **Developed by**: Snorkel AI. - **Model type**: Language Model - **Language(s)**: English - **License**: Apache 2.0 - **Model Description**: A 6.9B parameter pretrained language model. # Quick Start Please note that the model requires `transformers` version >= 4.25.1. To prompt the chat model, use the following format: ``` : [Chat] : ``` ## GPU Inference This requires a GPU with 16GB memory. ```python import torch import transformers from transformers import AutoTokenizer, AutoModelForCausalLM ``` Example with RedPajama-7B-Chat-Curated ```python # Example 1 using RedPajama-7B-Chat-Curated tokenizer = AutoTokenizer.from_pretrained("snorkelai/RedPajama-7B-Chat-Curated") model = AutoModelForCausalLM.from_pretrained("snorkelai/RedPajama-7B-Chat-Curated", torch_dtype=torch.float16) model = model.to('cuda:0') ## inference prompt = ": Give me a list of food recommendations to try in Japan.\n:" inputs = tokenizer(prompt, return_tensors='pt').to(model.device) input_length = inputs.input_ids.shape[1] outputs = model.generate( **inputs, max_new_tokens=512, do_sample=True, temperature=0.7, top_p=0.7, top_k=50, ) token = outputs.sequences[0, input_length:] output_str = tokenizer.decode(token) print(output_str) """ Here are some food recommendations to try in Japan: 1. Sushi: Japan is famous for its sushi, a dish of vinegared rice with various ingredients such as seafood, vegetables, and meat. 2. Tempura: Another popular Japanese dish, tempura is a vegetable or seafood batter that is deep-fried. 3. Yakitori: Grilled chicken skewers are a common street food in Japan. 4. Ramen: A noodle soup dish that is a staple food in Japan. 5. Bentos: A Japanese boxed lunch that is a convenient and affordable option for travelers. 6. Fugu: A type of fish that contains highly poisonous pufferfish. It is considered a delicacy in Japan and is only prepared by trained professionals. 7. Sukiyaki: A Japanese hot pot dish made with beef and vegetables. 8. Takoyaki: A Japanese dumpling dish filled with octopus and pickled ginger. 9. Gyoza: Chinese-style dumplings that are a popular street food in Japan. 10. Matcha: A type of green tea that is often served as a latte or in desserts. """ ``` Comparing with RedPajama-INCITE-7B-Chat ```python # Example 1 using RedPajama-INCITE-7B-Chat tokenizer = AutoTokenizer.from_pretrained("togethercomputer/RedPajama-INCITE-7B-Chat") model = AutoModelForCausalLM.from_pretrained("togethercomputer/RedPajama-INCITE-7B-Chat", torch_dtype=torch.float16) model = model.to('cuda:0') ## inference prompt = ": Give me a list of food recommendations to try in Japan.\n:" inputs = tokenizer(prompt, return_tensors='pt').to(model.device) input_length = inputs.input_ids.shape[1] outputs = model.generate( **inputs, max_new_tokens=512, do_sample=True, temperature=0.7, top_p=0.7, top_k=50, ) token = outputs.sequences[0, input_length:] output_str = tokenizer.decode(token) print(output_str) """ - Sushi - Ramen - Tempura - Yakitori - Sujuki - Gyoza - Takoyaki - Okonomiyaki - Baccala Tokkumon - Takoyaki Mayo - Karaage - Gohan bowl - California Maki """ ``` To do GPU Inference in Int8 or CPU inference, please refer to the `togethercomputer/RedPajama-INCITE-7B-Chat` documentation. ## Training **Training Data** Please refer to [snorkelai/snorkel-curated-instruction-tuning](https://huggingface.co/datasets/snorkelai/snorkel-curated-instruction-tuning) **Training Procedure** - **Hardware:** 8 A100 - **Optimizer:** Adam - **Gradient Accumulations**: 1 - **Num of Tokens:** 3.8M tokens - **Learning rate:** 1e-5 - **Batch size:** 64 ## License/Attribution **Copyright (2023) Snorkel AI, Inc.** This model was developed at [Snorkel AI](https://snorkel.ai/) and its use is subject to the Apache 2.0 license. This work comes with the collaboration with Together Computer. Please further licensing, please refer to the dataset licenses: [snorkelai/snorkel-curated-instruction-tuning](https://huggingface.co/datasets/snorkelai/snorkel-curated-instruction-tuning/) # Uses ## Direct Use Excluded uses are described below. ### Misuse, Malicious Use, and Out-of-Scope Use It is the responsibility of the end user to ensure that the model is used in a responsible and ethical manner. #### Out-of-Scope Use `RedPajama-7B-Chat-Curated` is a language model and may not perform well for other use cases outside of its intended scope. For example, it may not be suitable for use in safety-critical applications or for making decisions that have a significant impact on individuals or society. It is important to consider the limitations of the model and to only use it for its intended purpose. #### Misuse and Malicious Use `RedPajama-7B-Chat-Curated` is designed for language modeling. Misuse of the model, such as using it to engage in illegal or unethical activities, is strictly prohibited and goes against the principles of the project. Using the model to generate content that is cruel to individuals is a misuse of this model. This includes, but is not limited to: - Generating fake news, misinformation, or propaganda - Promoting hate speech, discrimination, or violence against individuals or groups - Impersonating individuals or organizations without their consent - Engaging in cyberbullying or harassment - Defamatory content - Spamming or scamming - Sharing confidential or sensitive information without proper authorization - Violating the terms of use of the model or the data used to train it - Creating automated bots for malicious purposes such as spreading malware, phishing scams, or spamming ## Limitations `RedPajama-7B-Chat-Curated`, like other language models, has limitations that should be taken into consideration. For example, the model may not always provide accurate or relevant answers, particularly for questions that are complex, ambiguous, or outside of its training data. We therefore welcome contributions from individuals and organizations, and encourage collaboration towards creating a more robust and inclusive chatbot. ## Community Join us on [Snorkel AI Slack](snorkel.ai/slack)