majid2230/crypto-supreme-27b

VISIONConcurrency Cost:2Model Size:27BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 11, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

majid2230/crypto-supreme-27b is a 27 billion parameter language model based on Qwen3.6-27B, enhanced with the R5 SUPREME distillation LoRA. This model is specifically optimized for cryptocurrency-related tasks, demonstrating a 34.2% win rate on clean test data and a 93.2% improvement in 60-day forward portfolio simulation. It is designed for deployment with vLLM and excels at tasks requiring Yes/No token logprob evaluation.

Loading preview...

Model Overview

majid2230/crypto-supreme-27b is a specialized 27 billion parameter language model built upon the Qwen3.6-27B architecture. Its core enhancement comes from the integration of the R5 SUPREME distillation LoRA (checkpoint-7500), which has been merged into the base model in bf16 format.

Key Capabilities

  • Cryptocurrency Optimization: Specifically fine-tuned for tasks within the cryptocurrency domain.
  • Performance: Achieves a 34.2% win rate on clean test data (compared to a base of 13.2%) when evaluated by Yes/No token logprobs on prompts ending with "Answer:".
  • Portfolio Simulation: Demonstrates a significant +93.2% improvement in 60-day forward portfolio simulation, indicating strong predictive or analytical capabilities for crypto assets.
  • Deployment: Designed for efficient serving with vLLM.

Use Cases

This model is particularly well-suited for applications requiring:

  • Cryptocurrency Analysis: Tasks involving market trends, predictions, or sentiment analysis within the crypto space.
  • Automated Trading Strategies: Potentially useful for informing or executing trading decisions based on its optimized performance in portfolio simulation.
  • Question Answering: Effective for Yes/No type questions, especially those related to cryptocurrency, given its evaluation methodology.

Finetunes
1 models