Ensure models.llamacpp Doesn't Have Implicit max_tokens

[lapp0]

Fixes #973

Problem

In #973 an invalid pattern was being generated, however it was only invalid because it was incomplete. in the LlamaCpp.generate() call llama_cpp_params["max_tokens"] was unset, however the model was still finishing for reason "length":

[{'text': '{ "name": "Grace Lee", "age":32, "', 'index': 0, 'logprobs': None, 'finish_reason': 'length'}]                                                                                                         

In turns out the max tokens was being implicitly set to 30.

Solution

Allow 2**30 tokens if max_length is unset /None, because llama_cpp_params requires an int for the max_tokens field.