ML разработчик

Obscura Compute is a participant in the Gonka decentralized AI network that leverages high-performance GPU infrastructure to maximize mining rewards.

• Implement advanced inference optimizations such as speculative decoding, quantization, and attention modifications to maximize mining weight;
• Fine-tune Docker configurations for various GPU models based on available registry;
• Develop custom optimization strategies that balance throughput and quality;
• Create and maintain custom Docker images optimized for specific GPU architectures;
• Design and implement systems for stable and scalable mining of Gonka and other protocols;
• Develop optimized images for Tenstorrent AI ASICs to expand the hardware ecosystem;
• Migrate Python code and vLLM implementations to new vLLM images and adapt them for specific GPU cards.

• Proven experience with large language model optimization techniques;
• Strong understanding of transformer architectures and attention mechanisms;
• Proficiency with PyTorch, CUDA, and GPU optimization techniques;
• Experience with vLLM, FlashInfer, or similar inference optimization frameworks;
• Familiarity with Docker containerization and GPU workload management;
• Nice to have: Experience with Claude Code Max, previous experience with Gonka or similar decentralized AI networks, background in competitive ML or distributed systems optimization, experience with NVIDIA GPU architectures (B200/B300/H200/H100/A100), knowledge of Tenstorrent AI ASICs or other specialized AI hardware.

• Opportunity to work with cutting-edge AI infrastructure;
• High performance-based bonuses tied to achieved weight improvements;
• Potential for a position with a percentage of mining profits;
• Access to high-end GPU hardware for experimentation.