(conversations)

( *, outputs: DSparkForwardOutput, loss_decay_gamma: Optional[float], ce_loss_alpha: float,

( *, model, batch: dict[str, torch.Tensor], ttt_length: int, step_loss_decay: float, )

( self, token_ids: torch.Tensor, hidden_states: Optional[torch.Tensor], )

(*, num_samples: int, rank: int, world_size: int)

(self, hidden_states: torch.Tensor)

(value)

(spec: DatasetSpec, args: argparse.Namespace)

(checkpoint_dir)

( tokenizer, messages, *, add_generation_prompt: bool, enable_thinking: bool | None = None

(self)

( *, seq_len: int, hidden_size: int, num_target_layers: int, )

( *, seq_len: int, hidden_size: int, num_target_layers: int, )

(cfg)

(*, output_dir: str, summaries, shard_map)

(output_path, error_path)

( self, *, dataset_name: str, metric_summary: dict[str, int | list[int]],

()

( model: DSparkModel, *, draft_input_ids: torch.Tensor, position_ids: torch.Tensor, past_k

( self, *, input_ids: torch.Tensor, stop_token_ids: list[int] | None, )

()

()

(target_config)

(target_config)

(self)

()

(self)

(self)

()

(spec: DatasetSpec, args: argparse.Namespace)

(spec: DatasetSpec, args: argparse.Namespace)

(spec: DatasetSpec, args: argparse.Namespace)

( data_name: str, dataset_root: str = DEFAULT_DATASET_ROOT, )

(spec: DatasetSpec, args: argparse.Namespace)

( *, resume_checkpoint_dir: str, draft_model, device, precision_dtype, global_rank: in

(args: argparse.Namespace)

(cache_dir: str)

( *, resume_checkpoint_dir: str, optimizer, global_rank: int, world_size: int, local_b

(self)

()

()

()

()

(self)

(batch, device)

( self, *, proposal: ConfidenceProposal, verification: VerificationResult,

( self, conversation, max_length, )

()

()

()

()

()

()

(dataset_dir: Path, row: dict)

( self, hidden_states: torch.Tensor, prev_token_ids: Optional[torch.Tensor] = None,

(output_dir: str)

( record, tokenizer, chat_template, max_length, )

(args)

(self, metrics_row: dict[str, object])

(self)

()

(self, hidden_states: torch.Tensor)

(self, hidden_states: torch.Tensor)

(*, output_dir: str, rank_dir: str, summary, shard_map)

( self, *, metrics_row: dict[str, object], confidence_row: dict, args_

(self)

(dataset_name: str)

(spec: DatasetSpec, args: argparse.Namespace)

(target_model, tokenizer)

(self, batch)

( *, target_model, input_ids: torch.Tensor, attention_mask: torch.Tensor, target_layer_ids

(src, dst)

( self, base_logits: torch.Tensor, *, first_prev_token_ids: torch.Tensor,

( target_probs: torch.Tensor, draft_probs: torch.Tensor, )

(*, train_config, checkpoint_dir: str)

(self, trainable: bool)

(self, trainable: bool)

(self, trainable: bool)

(self)

(self)

(row: dict)

(args: argparse.Namespace)

(args)

(args)

(*, cache_dir: str, manifest)

(*, train_dataset, draft_model, target_model_name_or_path)

(turns: list[str], row_number: int)

( self, *, metrics_row: dict[str, object], confidence_row: dict, args_

(dataset, output_path: Path)

( self, *, input_ids: torch.Tensor, attention_mask: torch.Tensor, loss

(*, output_dir: str, manifest)

(dataset, output_path: Path)

(self, features: List[Dict])

(self, features: List[Dict])

(self)

(self)

(self, item)

(self, idx)

(self, index: int)