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 Modelhuggingface:Qwen/Qwen3.6-35B-A3B-Instruct
Enginellama.cpp
QuantizationTQ3_4S
Platformcuda
VRAM Required12GB 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...