Codacus
Content creator and educator specializing in local large language model (LLM) deployment, optimization, and resource-constrained inference. Known for tutorials on running high-parameter models on consumer-grade hardware.
Key Works & Demonstrations
- Achieving Fast 35B MoE AI Model Performance on 6GB VRAM with Llama.cpp (2026-05-10)
- Channel guide: “Running a 35B AI Model on 6GB VRAM, FAST (llama.cpp Guide)”
- Demonstrated inference of qwen-36-35b-a3b (35B parameters, mixture-of-experts architecture) on hardware with only 6GB VRAM
- Leveraged quantization and MoE sparsity to bypass VRAM limits.
- Bonsai 27B vs. Qwen 35B: LLM Performance and Replacement Feasibility Benchmarks (2026-07-22)
- Channel guide: “Can a 3.5GB model replace my 35B daily driver? (Bonsai 27B)”
- Benchmarked Bonsai 27B against Qwen 35B to evaluate trade-offs between model size, inference speed, and real-world applicability.
- Investigated feasibility of replacing high-parameter daily drivers with smaller, optimized models for specific use cases.
References
Source Notes
- 2026-07-22: Bonsai 27B vs. Qwen 35B: LLM Performance and Replacement Feasibility Benchmarks · ▶ source
- 2026-07-13: Developing Persistent, Intelligent Memory for Local AI with a Librarian System · ▶ source
- 2026-05-31: Budget GPU Local Coding Agent Performance Optimization Report · ▶ source
- 2026-05-10: Achieving Fast 35B MoE AI Model Performance on 6GB VRAM with Llama.cpp · ▶ source