RLM Arbitrary Data Demo

This toy environment exercises the custom data serialization path in RLMEnv.

The environment supports multiple context_dtype options so you can test different serializers from a single entrypoint:

• text → string (default serializer)
• list → builtin serializer (random list of ints)
• tuple → builtin serializer (random tuple of ints)
• nested_list → builtin serializer (nested list of ints)
• nested_dict → builtin serializer (nested dict/list of ints)
• mixed → builtin serializer (mixed builtin containers)
• large_list → builtin serializer (larger list of ints)
• polars → custom serializer/deserializer in this environment

Run:

prime eval run rlm-arbitrary-data-demo -n 1

To pick a dtype explicitly:

prime eval run rlm-arbitrary-data-demo -n 1 --env-arg context_dtype=polars