Add LLM as Juddge evaluation metric

[NISH1001]

Major Changes

• Add evalem.nlp.metrics.llm.LLMAsJudgeMetric that uses LLM to evaluate and compare. Can be installed via [llm] namespace.

Minor Changes

• Add evalem._base.structures.SequenceType to encapsulate Union[list, tuple, set]
• Add imports to dunder init methods for easy imports of NLP metrics.
• Upgrade major dependencies

Usage

from evalem.nlp import LLMAsJudgeMetric

model = "ollama/llama3.2:3b"
api_base = "http://localhost:11434/v1"

references = [...]
predictions = [...]

LLMAsJudgeMetric(
    model=MODEL,
    api_base=API_BASE,
    api_key=os.environ.get("OPENAI_API_KEY"),
    # api_key=None,
    n_tries=3,
    prompt=PROMPT,
    debug=True,
).compute(
    references=references,
    predictions=predictions,
)