language¶
A causal-language-model name over TransformersModel (deprecated).
TransformersModel already generates through the model and returns token
ids (see its generate),
so LanguageModel adds nothing of its own — it just pins the
text-generation task and takes tokenizer_kwargs at load. Prefer
TransformersModel(repo_id, task="text-generation") directly; this name is kept
for backwards compatibility and warns on construction.
Bases: TransformersModel
Deprecated: a TransformersModel
pinned to the text-generation task.
Use TransformersModel(repo_id, task="text-generation") instead. Everything
this class offered — generate returning token ids, pipe running the
pipeline, trace, scan — lives on TransformersModel now; the
only extra here is tokenizer_kwargs (apply the same settings to
model.tokenizer yourself, padding_side being the usual one).