bunnycore/Phi-4-RP-v0

TEXT GENERATIONPricing:Input $0.28 / Output $0.56Concurrent Unit Cost:1Model Size:14.7BQuant:FP8Context Size:32kPublished:Jan 10, 2025License:mitArchitecture:Transformer0.0K Open Weights Featherless Exclusive Cold

bunnycore/Phi-4-RP-v0 is a 14.7 billion parameter language model created by bunnycore, resulting from a linear merge of Undi95/Phi4-abliterated and bunnycore/Phi-4-RP. This model leverages the strengths of its base components, offering a 32768 token context length. It is designed for general language understanding and generation tasks, with its merged architecture aiming for balanced performance across various benchmarks.

Loading preview...

Overview

bunnycore/Phi-4-RP-v0 is a 14.7 billion parameter language model developed by bunnycore. It was created using a linear merge method via mergekit, combining two base models: Undi95/Phi4-abliterated and bunnycore/Phi-4-RP. This merging strategy aims to synthesize the capabilities of its constituent models.

Key Capabilities

  • Merged Architecture: Combines two distinct Phi-4 variants to potentially enhance overall performance.
  • Linear Merge Method: Utilizes a straightforward weighting approach for model integration.
  • BFloat16 Precision: The model is configured to use bfloat16 for efficient computation.

Performance Highlights

Evaluated on the Open LLM Leaderboard, bunnycore/Phi-4-RP-v0 achieved an average score of 37.32. Specific benchmark results include:

  • IFEval (0-Shot): 68.27
  • BBH (3-Shot): 54.84
  • MMLU-PRO (5-shot): 48.49

Good for

  • General Language Tasks: Suitable for a broad range of applications requiring language understanding and generation.
  • Experimentation with Merged Models: Developers interested in exploring the performance characteristics of linearly merged models.
  • Applications requiring a 14.7B parameter model: Offers a balance between model size and capability for various deployment scenarios.

Finetunes
2 models