BLOOMZ&mT0
FreeMultilingual instruction-tuned language model
FreeFree tier
Inputs: textOutputs: text
About BLOOMZ&mT0
BLOOMZ is a multilingual instruction-tuned language model developed by BigScience. It is fine-tuned from the BLOOM model to follow human instructions across many languages. The model supports tasks such as text generation, natural language inference, and coreference resolution, as evidenced by its performance on benchmarks like Winogrande, XWinograd, ANLI, SuperGLUE, and XNLI. It is available on Hugging Face as an open-source model.
Key Features
Fine-tuned from BLOOM to follow instructions across multiple languages
Supports text generation tasks
Capable of coreference resolution (e.g., Winogrande, XWinograd)
Handles natural language inference (ANLI, SuperGLUE, XNLI)
Available for free as an open-source model on Hugging Face
Pros & Cons
Pros
- Strong multilingual performance across diverse languages (English, French, Japanese, Portuguese, Russian, Chinese, Arabic, Bulgarian, etc.)
- Fine-tuned for instruction following, making it more aligned with user prompts
- Open-source and freely available for download and customization
Cons
- Requires significant computational resources to run (large model size)
- May not be as capable as larger contemporary models for complex reasoning
- Bias and limitations inherent in the base BLOOM model
Best For
Multilingual instruction following and text generationNatural language understanding tasks such as coreference resolution and inferenceResearch and experimentation with large language models
FAQ
What languages does BLOOMZ support?
Based on evaluation datasets, BLOOMZ supports English, French, Japanese, Portuguese, Russian, Chinese, Arabic, Bulgarian, and likely many others from the BLOOM training.
What tasks can BLOOMZ perform?
BLOOMZ excels at text generation, coreference resolution, and natural language inference, as demonstrated on Winogrande, XWinograd, ANLI, SuperGLUE, and XNLI benchmarks.
Is BLOOMZ free to use?
Yes, BLOOMZ is an open-source model available for free on Hugging Face.