Back to registry
CUDA12GB VRAM
arnav080/qwen3.6-35b-moe-turboquant-tq3-AJKV
Optimized Qwen 3.6 35B MoE recipe using the experimental llama.cpp-tq3 (TurboQuant) engine, TQ3_4S quant, and hybrid offload.
Configuration Specifications
| Base Model | huggingface:Qwen/Qwen3.6-35B-A3B-Instruct |
| Engine | llama.cpp |
| Quantization | TQ3_4S |
| Platform | cuda |
| VRAM Required | 12GB minimum |
Telemetry Benchmarks
No runs recorded yet
Be the first to benchmark this recipe! Run the CLI command from your terminal:
bloc run arnav080/qwen3.6-35b-moe-turboquant-tq3-AJKV
qwen3.6-35b-moe-turboquant-tq3-AJKV.yaml
Loading workspace editor...