Transformers Engine Schema

Engine version: 4.57.3
Discovered at: 2026-05-06T22:57:22+02:00
Discovery method: inspect.signature(from_pretrained) + GenerationConfig().to_dict()
Schema version: 1.0.0

Summary: 9 engine parameters, 67 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 has no type annotations; None defaults yield type='unknown' Affected fields: max_new_tokens, min_new_tokens, max_time, stop_strings, cache_implementation, cache_config, return_legacy_cache, prefill_chunk_size, min_p, bad_words_ids, forced_bos_token_id, forced_eos_token_id, exponential_decay_length_penalty, suppress_tokens, begin_suppress_tokens, sequence_bias, guidance_scale, watermarking_config, output_logits, pad_token_id, bos_token_id, eos_token_id, decoder_start_token_id, prompt_lookup_num_tokens, max_matching_ngram_size, assistant_early_exit, low_memory, penalty_alpha, dola_layers, constraints, force_words_ids

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`

Sampling Parameters

Field	Type	Default
`max_length`	`int`	`20`
`max_new_tokens`	`unknown`	—
`min_length`	`int`	`0`
`min_new_tokens`	`unknown`	—
`early_stopping`	`bool`	`false`
`max_time`	`unknown`	—
`stop_strings`	`unknown`	—
`do_sample`	`bool`	`false`
`num_beams`	`int`	`1`
`use_cache`	`bool`	`true`
`cache_implementation`	`unknown`	—
`cache_config`	`unknown`	—
`return_legacy_cache`	`unknown`	—
`prefill_chunk_size`	`unknown`	—
`temperature`	`float`	`1.0`
`top_k`	`int`	`50`
`top_p`	`float`	`1.0`
`min_p`	`unknown`	—
`typical_p`	`float`	`1.0`
`epsilon_cutoff`	`float`	`0.0`
`eta_cutoff`	`float`	`0.0`
`repetition_penalty`	`float`	`1.0`
`encoder_repetition_penalty`	`float`	`1.0`
`length_penalty`	`float`	`1.0`
`no_repeat_ngram_size`	`int`	`0`
`bad_words_ids`	`unknown`	—
`renormalize_logits`	`bool`	`false`
`forced_bos_token_id`	`unknown`	—
`forced_eos_token_id`	`unknown`	—
`remove_invalid_values`	`bool`	`false`
`exponential_decay_length_penalty`	`unknown`	—
`suppress_tokens`	`unknown`	—
`begin_suppress_tokens`	`unknown`	—
`sequence_bias`	`unknown`	—
`token_healing`	`bool`	`false`
`guidance_scale`	`unknown`	—
`watermarking_config`	`unknown`	—
`num_return_sequences`	`int`	`1`
`output_attentions`	`bool`	`false`
`output_hidden_states`	`bool`	`false`
`output_scores`	`bool`	`false`
`output_logits`	`unknown`	—
`return_dict_in_generate`	`bool`	`false`
`pad_token_id`	`unknown`	—
`bos_token_id`	`unknown`	—
`eos_token_id`	`unknown`	—
`encoder_no_repeat_ngram_size`	`int`	`0`
`decoder_start_token_id`	`unknown`	—
`is_assistant`	`bool`	`false`
`num_assistant_tokens`	`int`	`20`
`num_assistant_tokens_schedule`	`str`	`constant`
`assistant_confidence_threshold`	`float`	`0.4`
`prompt_lookup_num_tokens`	`unknown`	—
`max_matching_ngram_size`	`unknown`	—
`assistant_early_exit`	`unknown`	—
`assistant_lookbehind`	`int`	`10`
`target_lookbehind`	`int`	`10`
`disable_compile`	`bool`	`false`
`low_memory`	`unknown`	—
`penalty_alpha`	`unknown`	—
`dola_layers`	`unknown`	—
`diversity_penalty`	`float`	`0.0`
`num_beam_groups`	`int`	`1`
`constraints`	`unknown`	—
`force_words_ids`	`unknown`	—
`_from_model_config`	`bool`	`false`
`transformers_version`	`str`	`4.57.3`

Engine Parameters​

Sampling Parameters​

Engine Parameters

Sampling Parameters