{ "inference": { "gpt2": { "instanceType": "ml.g4dn.xlarge", "numShard": 1 }, "google/flan-t5-xxl": { "instanceType": "ml.g5.12xlarge", "numShard": 4 }, "bigscience/bloom-7b1": { "instanceType": "ml.g5.12xlarge", "numShard": 4 }, "EleutherAI/gpt-neo-2.7B": { "instanceType": "ml.g5.xlarge", "numShard": 1 }, "tiiuae/falcon-40b-instruct": { "instanceType": "ml.g5.12xlarge", "numShard": 4 }, "EleutherAI/gpt-neox-20b": { "instanceType": "ml.g5.12xlarge", "numShard": 4 }, "bigcode/starcoder": { "instanceType": "ml.g5.12xlarge", "numShard": 4 }, "openlm-research/open_llama_13b": { "instanceType": "ml.g5.12xlarge", "numShard": 4 }, "amazon/FalconLite": { "instanceType": "ml.g5.12xlarge", "numShard": 4 } }, "training":{} }