jpacifico/Chocolatine-14B-Instruct-DPO-v1.3

TEXT GENERATIONConcurrency Cost:1Model Size:14.7BQuant:FP8Ctx Length:32kPublished:Dec 22, 2024License:mitArchitecture:Transformer0.0K Open Weights Cold

jpacifico/Chocolatine-14B-Instruct-DPO-v1.3 is a 14.7 billion parameter instruction-tuned causal language model developed by jpacifico, fine-tuned from Microsoft's Phi-4 architecture. This model excels in French language tasks and general capabilities, demonstrating significant performance improvements over its base model and previous versions on benchmarks like MT-Bench-French. With a context window of up to 16,000 tokens, it offers strong performance for its size, particularly noted for its efficiency on the OpenLLM Leaderboard.

Loading preview...

Chocolatine-14B-Instruct-DPO-v1.3: Enhanced French and General LLM

Chocolatine-14B-Instruct-DPO-v1.3 is a 14.7 billion parameter language model developed by jpacifico, built upon Microsoft's Phi-4 architecture. It has undergone DPO (Direct Preference Optimization) fine-tuning using the jpacifico/french-orca-dpo-pairs-revised dataset, which has significantly boosted its performance in French and overall capabilities.

Key Capabilities & Performance

  • Superior French Performance: Outperforms its base model, Phi-4, and previous Chocolatine versions on the MT-Bench-French benchmark, demonstrating strong multi-turn conversational abilities in French.
  • OpenLLM Leaderboard Recognition: Achieves a notable average score of 42.42 on the OpenLLM Leaderboard, making it the best-performing Phi-4 based model. It is also highlighted for its energy efficiency, with a low carbon footprint (1.70kgCo2).
  • Extended Context Window: Supports a context length of up to 16,000 tokens, allowing for processing longer inputs and generating more coherent, extended responses.
  • Multilingual Support: While primarily enhanced for French, the model also supports English.

When to Use This Model

  • French Language Applications: Ideal for chatbots, content generation, and conversational AI systems requiring high proficiency in French.
  • Resource-Efficient Deployment: Suitable for scenarios where strong performance is needed from a 14B parameter model, especially considering its reported efficiency.
  • Instruction-Following Tasks: Excels in tasks requiring adherence to instructions due to its DPO fine-tuning.

Limitations

This model series is a demonstration of effective fine-tuning. It does not include any built-in moderation mechanisms.