generation_config.json

{
  "_from_model_config": null,
  "assistant_confidence_threshold": null,
  "assistant_early_exit": null,
  "assistant_lookbehind": null,
  "bad_words_ids": null,
  "begin_suppress_tokens": null,
  "bos_token_id": 2,
  "cache_config": null,
  "cache_implementation": null,
  "compile_config": null,
  "constraints": null,
  "continuous_batching_config": null,
  "decoder_start_token_id": null,
  "disable_compile": null,
  "diversity_penalty": null,
  "do_sample": true,
  "dola_layers": null,
  "early_stopping": null,
  "encoder_no_repeat_ngram_size": null,
  "encoder_repetition_penalty": null,
  "eos_token_id": 1,
  "epsilon_cutoff": null,
  "eta_cutoff": null,
  "exponential_decay_length_penalty": null,
  "force_words_ids": null,
  "forced_bos_token_id": null,
  "forced_eos_token_id": null,
  "guidance_scale": null,
  "is_assistant": null,
  "length_penalty": null,
  "low_memory": null,
  "max_length": null,
  "max_matching_ngram_size": null,
  "max_new_tokens": null,
  "max_time": null,
  "min_length": null,
  "min_new_tokens": null,
  "min_p": null,
  "no_repeat_ngram_size": null,
  "num_assistant_tokens": null,
  "num_assistant_tokens_schedule": null,
  "num_beam_groups": null,
  "num_beams": null,
  "num_return_sequences": null,
  "output_attentions": null,
  "output_hidden_states": null,
  "output_logits": null,
  "output_scores": null,
  "pad_token_id": 0,
  "penalty_alpha": null,
  "prefill_chunk_size": null,
  "prompt_lookup_num_tokens": null,
  "remove_invalid_values": null,
  "renormalize_logits": null,
  "repetition_penalty": null,
  "return_dict_in_generate": null,
  "sequence_bias": null,
  "stop_strings": null,
  "suppress_tokens": null,
  "target_lookbehind": null,
  "temperature": 1.0,
  "token_healing": null,
  "top_h": null,
  "top_k": 64,
  "top_p": 0.95,
  "transformers_version": "5.5.0",
  "trust_remote_code": false,
  "typical_p": null,
  "use_cache": null,
  "watermarking_config": null
}