Auto-dispatching wrapper that loads a causal (left-to-right) LM.
Resolves names like "gpt" / "gpt2_small" to their concrete
{Family}LMHeadModel subclass. The model is suitable for
next-token prediction and autoregressive generation via
CausalLMMixin.generate.
Notes
The forward returns a CausalLMOutput with logits shaped
(B, T, vocab_size) and optionally a past_key_values tuple for
cache-friendly decoding. Models tagged "causal-lm" typically
expose config.vocab_size / pad_token_id / bos_token_id /
eos_token_id.
Examples
>>> from lucid.models import AutoModelForCausalLM
>>> model = AutoModelForCausalLM.from_pretrained("gpt")
>>> type(model).__name__
'GPTLMHeadModel'