Hugging Face describes a torch-neuronx model-support Space that lists Trainium compatibility results, stack versions and runnable scripts, giving developers a starting point for checking a model before attempting deployment. Hugging Face
Hugging Face limits current testing to Transformers text-generation and image-text-to-text models fitting within four Trainium2 devices, or 16 cores. Its checks generate 32 tokens: single-core outputs are compared with CPU outputs, while sharded outputs receive a fluency check from a small judge model. These are vendor-reported checks, not independent confirmation. Hugging Face
Hugging Face explicitly says a passing result establishes neither throughput nor longer-generation performance. Those deployment characteristics remain unverified by the described checks; Hugging Face also notes that omitted models may simply lack testing priority rather than be incompatible. Hugging Face