Fix: Correct base model from Qwen3-2B to Qwen3-8B

Upload sharded model (9x2GB shards, continuous batching, neuronxcc 2.21)

initial commit