Blog.Cloudflare Cloudflare Launches Clef Decision Models with RL Fine-Tuning
Article Content
- •Cloudflare released Clef and Clef-flash decision models as open-source.
- •The models improve classification speed and accuracy for various inputs.
- •Reinforcement learning fine-tuning is available for user-specific applications.
Cloudflare has introduced its Clef and Clef-flash decision models as open-source under the Apache 2.0 license. These models are designed to make programmatic decisions based on input data, significantly improving classification speed and accuracy compared to traditional large language models. Clef has been tested by Cloudflare's Threat Intelligence team to classify website domains, achieving a classification time of 2.2 seconds, compared to 4.7 seconds for their fastest LLM. The models are hosted on Workers AI, and users can fine-tune them using reinforcement learning for specific workloads. Cloudflare aims to enhance threat intelligence workflows and automate decision-making processes. The models are available for experimentation on Hugging Face.
Ask AI about this cluster
Answers cite the sources they use