Safurai/Safurai-Csharp-34B

TEXT GENERATIONConcurrent Unit Cost:2Model Size:34BQuant:FP8Context Size:16kPublished:Oct 3, 2023License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Safurai/Safurai-Csharp-34B is a 34 billion parameter CodeLlama-based model developed by Safurai, fine-tuned using QLoRA on 13 billion tokens of C# evolved Q&A data. This model is specifically optimized for C# code generation and understanding, achieving 56% at pass@1 with n=5 on the MultiPL-E benchmark for C#. It is designed to excel in C# programming tasks, offering high performance for developers working with the language.

Loading preview...

Safurai-Csharp-34B Overview

Safurai-Csharp-34B is a 34 billion parameter language model developed by Safurai, built upon the codellama/CodeLlama-34b-hf architecture. It has been meticulously fine-tuned using QLoRA (4-bit precision) on an extensive dataset comprising 13 billion tokens of C# evolved Q&A data.

Key Capabilities & Performance

  • C# Code Generation: The model is highly specialized for generating C# code, leveraging its targeted training data.
  • State-of-the-Art Benchmarking: It achieved a notable 56% at pass@1 with n=5 on the MultiPL-E code LLM benchmark specifically for C#, indicating strong performance in code completion and generation tasks.
  • Efficient Training: The model was trained in approximately 7 hours and 40 minutes on 2 x NVIDIA A100 PCIe 80GB GPUs, utilizing a sequence_len of 4096 and flash_attention for efficiency.

Good For

  • C# Developers: Ideal for developers and teams requiring a powerful language model specifically tailored for C# programming.
  • Code Generation & Assistance: Excels in scenarios demanding accurate and contextually relevant C# code generation, auto-completion, and problem-solving.
  • Research & Development: Useful for exploring advanced C# language models and integrating them into development workflows.