| Property | Value |
|---|---|
| Parameter Count | 176 Billion |
| License | Modified Apache 2.0 with RAIL restrictions |
| Developer | SambaNova Systems & Together Computer |
| Base Model | BLOOM |
What is BLOOMChat-176B-v1?
BLOOMChat-176B-v1 is a state-of-the-art multilingual chat model developed by SambaNova Systems and Together Computer. Built upon the BLOOM architecture, this 176B parameter model has been instruction-tuned specifically for conversation and question-answering tasks across multiple languages. The model represents a significant advancement in open-source multilingual AI capabilities, combining the robust foundation of BLOOM with enhanced conversational abilities.
Implementation Details
The model was trained using SambaNova's Reconfigurable Dataflow Unit (RDU) architecture, utilizing a carefully curated training process that included instruction tuning on the OIG dataset, Dolly 2.0, and Oasst1. The training procedure involved specific hyperparameters including AdamW optimizer, cosine learning rate scheduling, and a global batch size of 128.
- Training utilized both bf16 and int8 precision options
- Implements specific ChatML formatting with human/bot tags
- Supports multiple deployment frameworks including Hugging Face Transformers
Core Capabilities
- Multilingual conversation and question-answering
- Context-aware responses across various languages
- Advanced text generation with customizable parameters
- Support for multiple deployment options including GPU and RDU implementations
Frequently Asked Questions
Q: What makes this model unique?
BLOOMChat-176B-v1 stands out for its combination of massive scale (176B parameters), multilingual capabilities, and specialized instruction tuning for conversational AI. It's one of the largest open-source multilingual chat models available.
Q: What are the recommended use cases?
The model is best suited for commercial and research applications in multilingual environments, including chatbots, question-answering systems, and content generation. However, it should not be used for mission-critical applications or important automated pipelines due to potential limitations and biases.