Tomhn/

phi-4-FP8-Dynamic-fixed

This is a quantization of the phi-4.

The phi-4 model is a cutting-edge open-source LLM developed using a diverse mix of synthetic datasets, curated public domain web content, and acquired academic resources, including books and Q&A datasets. This deliberate data selection ensures the training of compact yet highly capable models with an emphasis on quality and advanced reasoning. To further enhance its performance, phi-4 underwent a rigorous alignment process that included supervised fine-tuning and direct preference optimization, resulting in precise instruction adherence and robust safety measures.

This model provides an accuracy recovery of 99.68%.

We did not check for data contamination. Evaluation was done using Eval. Harness with limit=1000.

Install vLLM and run the server:

Access the model:

curl http://localhost:8000/v1/completions     -H "Content-Type: application/json"     -d ' {
        "model": "cortecs/phi-4-FP8-Dynamic",
        "prompt": "San Francisco is a"
    } '

⚡ This model is optimized to handle heavy workloads providing a total throughput of ️4623 tokens per second using one NVIDIA L40S ⚡