The 12B ships as a multimodal class with no encoder loaded, and its sliding-window attention overflows Neuron's fused-attention SBUF. Three surgical fixes and it serves Paris at ~15 tok/s on one inf2.8xlarge.
The 12B ships as a multimodal class with no encoder loaded, and its sliding-window attention overflows Neuron's fused-attention SBUF. Three surgical fixes and it serves Paris at ~1...
Porting Gemma-4 31B (dense) to AWS Inferentia2: the tensor-parallel recipe that worked at 12B collapses at 31B, NxD ModelBuilder saves it — and then a passing validation lies to yo...
The MoE was the hard one: a dual-path FFN, a sparse expert loop that won't trace, and a bug where the device output was empty while the CPU reference was perfect and every unit tes...