ZeroAgency/Zero-Mistral-Small-24B-Instruct-2501

Hugging Face
TEXT GENERATIONPricing:Input $0.7 / Cached $0.04 / Output $1.16Concurrent Unit Cost:2Model Size:24BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Feb 15, 2025License:mitArchitecture:Transformer0.0K Open Weights Featherless Exclusive Warm

ZeroAgency/Zero-Mistral-Small-24B-Instruct-2501 is a 24 billion parameter instruction-tuned causal language model developed by ZeroAgency, based on mistralai/Mistral-Small-24B-Instruct-2501. It is primarily adapted and optimized for performance in both Russian and English languages, trained on the Vikhrmodels/GrandMaster-PRO-MAX dataset. This model excels in conversational tasks and demonstrates improved performance over its base model on Russian-language benchmarks, making it suitable for bilingual applications requiring strong instruction following.

Loading preview...

Overview

Zero-Mistral-Small-24B-Instruct-2501 is an enhanced 24 billion parameter instruction-tuned model from ZeroAgency, built upon the mistralai/Mistral-Small-24B-Instruct-2501 architecture. Its primary distinction lies in its adaptation and optimization for both Russian and English languages, achieved through supervised fine-tuning (SFT) on the GrandMaster-PRO-MAX dataset.

Key Capabilities & Performance

  • Bilingual Proficiency: Specifically tuned for high performance in both English and Russian, making it a strong candidate for multilingual applications.
  • Improved Benchmarks: Demonstrates notable improvements over its base model, Mistral-Small-24B-Instruct-2501, particularly on Russian-language benchmarks like Ru Arena General and Arena-Hard-Ru lm_eval, where it scores 87.43 and 77.5 respectively.
  • Instruction Following: Designed for conversational and chat applications, leveraging its instruction-tuned nature.
  • Function Calling: Supports advanced function/tool calling capabilities, as demonstrated in the provided usage examples.

When to Use This Model

  • Bilingual Applications: Ideal for use cases requiring robust performance in both Russian and English.
  • Conversational AI: Well-suited for chatbots, virtual assistants, and other dialogue-based systems.
  • Instruction Following Tasks: Excels in scenarios where precise adherence to instructions is critical.
  • Tool Use: Recommended for applications that benefit from function calling and integration with external tools.

Limitations

  • Running the 16-bit merged version on GPU requires approximately 55-60 GB of GPU RAM, which may be a consideration for deployment.

Finetunes
6 models