Transformers Engine Schema

Engine version: 5.14.1
Discovered at: 2026-08-06T13:06:39.990799+00:00
Discovery method: inspect.signature(from_pretrained) + GenerationConfig().to_dict() with docstring-Args type recovery
Schema version: 1.0.0

Summary: 11 engine parameters, 72 sampling parameters.

Discovery limitations

engine_params - from_pretrained accepts **kwargs; kwargs are not in the signature (documented kwargs live in the class docstring only) Affected fields: AutoModelForCausalLM.from_pretrained.**model_args, AutoModelForCausalLM.from_pretrained.**kwargs, PreTrainedModel.from_pretrained.**model_args, PreTrainedModel.from_pretrained.**kwargs

sampling_params - GenerationConfig field is None-defaulted and undocumented in the docstring Args block; type='unknown' Affected fields: use_mtp, cache_config, bad_words_ids, exponential_decay_length_penalty, suppress_tokens, begin_suppress_tokens, sequence_bias, watermarking_config, eos_token_id, compile_config, continuous_batching_config, low_memory, penalty_alpha, dola_layers, diversity_penalty, num_beam_groups, constraints, force_words_ids, prefill_chunk_size, _from_model_config

Engine Parameters

Field	Type	Default	Description
`config`	`PreTrainedConfig	str	PathLike
`cache_dir`	`str	PathLike	None`
`ignore_mismatched_sizes`	`bool`	`false`
`force_download`	`bool`	`false`
`local_files_only`	`bool`	`false`
`token`	`str	bool	None`
`revision`	`str`	`main`
`use_safetensors`	`bool	None`	-
`weights_only`	`bool`	`true`
`fusion_config`	`dict[str, bool	dict[str, Any]]	None`
`disable_mmap`	`bool	None`	-

Sampling Parameters

Field	Type	Default	Description
`max_length`	`int`	-
`max_new_tokens`	`int`	-
`min_length`	`int`	-
`min_new_tokens`	`int`	-
`early_stopping`	`bool	Literal['never']`	-
`max_time`	`float`	-
`stop_strings`	`str`	-
`do_sample`	`bool`	-
`num_beams`	`int`	-
`use_mtp`	`unknown`	-
`use_cache`	`bool`	-
`cache_implementation`	`str`	-
`cache_config`	`unknown`	-
`max_cache_len`	`int`	-
`temperature`	`float`	-
`top_k`	`int`	-
`top_p`	`float`	-
`min_p`	`float`	-
`top_h`	`float`	-
`typical_p`	`float`	-
`epsilon_cutoff`	`float`	-
`eta_cutoff`	`float`	-
`repetition_penalty`	`float`	-
`encoder_repetition_penalty`	`float`	-
`length_penalty`	`float`	-
`no_repeat_ngram_size`	`int`	-
`bad_words_ids`	`unknown`	-
`renormalize_logits`	`bool`	-
`forced_bos_token_id`	`int`	-
`forced_eos_token_id`	`int`	-
`remove_invalid_values`	`bool`	-
`exponential_decay_length_penalty`	`unknown`	-
`suppress_tokens`	`unknown`	-
`begin_suppress_tokens`	`unknown`	-
`sequence_bias`	`unknown`	-
`token_healing`	`bool`	-
`guidance_scale`	`float`	-
`watermarking_config`	`unknown`	-
`num_return_sequences`	`int`	-
`output_attentions`	`bool`	-
`output_hidden_states`	`bool`	-
`output_scores`	`bool`	-
`output_logits`	`bool`	-
`return_dict_in_generate`	`bool`	-
`pad_token_id`	`int`	-
`bos_token_id`	`int`	-
`eos_token_id`	`unknown`	-
`encoder_no_repeat_ngram_size`	`int`	-
`decoder_start_token_id`	`int`	-
`is_assistant`	`bool`	-
`num_assistant_tokens`	`int`	-
`num_assistant_tokens_schedule`	`str`	-
`assistant_confidence_threshold`	`float`	-
`prompt_lookup_num_tokens`	`int`	-
`max_matching_ngram_size`	`int`	-
`assistant_early_exit`	`int`	-
`assistant_lookbehind`	`int`	-
`target_lookbehind`	`int`	-
`assistant_ensemble_weight`	`float`	-
`compile_config`	`unknown`	-
`disable_compile`	`bool`	-
`continuous_batching_config`	`unknown`	-
`low_memory`	`unknown`	-
`penalty_alpha`	`unknown`	-
`dola_layers`	`unknown`	-
`diversity_penalty`	`unknown`	-
`num_beam_groups`	`unknown`	-
`constraints`	`unknown`	-
`force_words_ids`	`unknown`	-
`prefill_chunk_size`	`unknown`	-
`_from_model_config`	`unknown`	-
`transformers_version`	`str`	`5.14.1`

Engine Parameters​

Sampling Parameters​

Engine Parameters

Sampling Parameters