Mutaz Al Awamleh

🦌+🦅 2x Faster • 3x Smaller • 6x Savings • <1% Loss | NVFP4 Quantized

SOTA LLM Inference | Latest CUDA 13.0 | Blackwell 6-20x | H100/A100 2-6x faster

🦌 Qwen3-VL-32B NVFP4 | Vision-Language | 20GB (was 62GB) | <0.3% accuracy loss

Nemotron3-30B NVFP4 Quantized - First NVFP4 for NVIDIA Nemotron-3

Zero-Config TensorRT-LLM | Pre-loaded Nemotron-3-Nano-30B | By Mutaz Al Awamleh - ELK-AI

Devstral-Small-2-24B FP8 - Blackwell-optimized Mistral coding model

NVIDIA Nemotron-VL-12B NVFP4 - Blackwell-optimized multimodal vision-language model

Qwen3-VL-8B-Thinking NVFP4 W4A16 - First NVFP4 quantization for Blackwell

Qwen3-VL-4B-Thinking NVFP4 W4A16 - First NVFP4 quantization for Blackwell

Qwen3-VL-2B-Thinking NVFP4 W4A16 - First NVFP4 quantization for Blackwell

vLLM Blackwell NVFP4 Optimized - Blackwell-optimized vLLM for NVFP4/FP8

NVFP4 Quantization Pipeline | CUDA 13.0 | 72% Memory Reduction | <0.3% Loss