outlines · Issues· 171 open

LlamaCppTokenizer drops every token whose piece exceeds the 32-byte buffer: llama_token_to_piece reports an undersized buffer with a negative return, so the n > size retry is unreachable

`types.datetime` rejects the ISO 8601 `T` separator (`datetime.isoformat()` output does not match)

outlines with gemma4 output wrong answer

DSL regex for List[T] and variadic Tuple[T, ...] forbids the empty collection, diverging from Dict and the JSON-schema path

DSL container types render bool as Python True/False, producing invalid JSON and diverging from the JSON-schema path

outlines.from_openai with custom base_url /v1 for a Chat Completions host

DSL containers leave date/time/datetime unquoted, producing output json.loads rejects

Optional[T] generates Python None instead of JSON null inside DSL containers

Built-in Regex terms are not JSON-quoted inside containers (list[types.email] generates [[email protected]])

SGLangTypeAdapter silently drops whitespace_pattern from JsonSchema (vLLM backend handles it correctly)

mlxlm.generate_batch uses falsy guard (if output_type:) instead of identity check (if output_type is not None:)

JsonSchema and CFG are unhashable: defining __eq__ without __hash__ sets __hash__ = None

applications: Application does not detect output_type changes between calls

Template.from_string leaks stray blank lines/whitespace when a trailing blank line is over-indented or tab-indented

caching: cache_disabled() context manager is not safe in asyncio contexts