gpt-inference-task.json

{"$defs": {"GPTGenerationConfig": {"properties": {"max_new_tokens": {"title": "Max New Tokens", "type": "integer"}, "do_sample": {"title": "Do Sample", "type": "boolean"}, "num_beams": {"title": "Num Beams", "type": "integer"}, "temperature": {"title": "Temperature", "type": "number"}, "typical_p": {"title": "Typical P", "type": "number"}, "top_k": {"title": "Top K", "type": "integer"}, "top_p": {"title": "Top P", "type": "number"}, "repetition_penalty": {"title": "Repetition Penalty", "type": "number"}, "num_return_sequences": {"title": "Num Return Sequences", "type": "integer"}}, "title": "GPTGenerationConfig", "type": "object"}, "Message": {"properties": {"role": {"enum": ["system", "user", "assistant"], "title": "Role", "type": "string"}, "content": {"title": "Content", "type": "string"}}, "required": ["role", "content"], "title": "Message", "type": "object"}}, "properties": {"model": {"minLength": 1, "title": "Model", "type": "string"}, "messages": {"items": {"$ref": "#/$defs/Message"}, "title": "Messages", "type": "array"}, "generation_config": {"anyOf": [{"$ref": "#/$defs/GPTGenerationConfig"}, {"type": "null"}], "default": null}, "seed": {"default": 0, "title": "Seed", "type": "integer"}, "dtype": {"default": "auto", "enum": ["float16", "bfloat16", "float32", "auto"], "title": "Dtype", "type": "string"}, "quantize_bits": {"anyOf": [{"enum": [4, 8], "type": "integer"}, {"type": "null"}], "default": null, "title": "Quantize Bits"}}, "required": ["model", "messages"], "title": "GPTTaskArgs", "type": "object"}