| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | |
| 4 | import os |
| 5 | |
| 6 | from transformers import AutoConfig, DeepseekV2Config, PreTrainedConfig |
| 7 | |
| 8 | from vllm.transformers_utils.utils import without_trust_remote_code |
| 9 | |
| 10 | |
| 11 | class EAGLEConfig(PreTrainedConfig): |
| 12 | model_type = "eagle" |
| 13 | |
| 14 | def __init__( |
| 15 | self, |
| 16 | model: PreTrainedConfig | dict | None = None, |
| 17 | truncated_vocab_size: int | None = None, |
| 18 | method: str | None = "eagle", |
| 19 | **kwargs, |
| 20 | ): |
| 21 | model_config: PreTrainedConfig | DeepseekV2Config | None |
| 22 | if isinstance(model, dict): |
| 23 | model_config = AutoConfig.for_model(**model) |
| 24 | else: |
| 25 | model_config = model |
| 26 | |
| 27 | for k, v in kwargs.items(): |
| 28 | if k != "architectures" and k != "model_type" and hasattr(model_config, k): |
| 29 | setattr(model_config, k, v) |
| 30 | |
| 31 | self.model = model_config |
| 32 | |
| 33 | if self.model is None: |
| 34 | self.truncated_vocab_size = None |
| 35 | else: |
| 36 | self.truncated_vocab_size = ( |
| 37 | self.model.vocab_size |
| 38 | if truncated_vocab_size is None |
| 39 | else truncated_vocab_size |
| 40 | ) |
| 41 | |
| 42 | # Eagle model name should follow naming convention of |
| 43 | # LlamaForCausalLM -> EagleLlamaForCausalLM |
| 44 | # LlamaForCausalLM -> Eagle3LlamaForCausalLM |
| 45 | # LlamaForCausalLMEagle3 -> LlamaForCausalLMEagle3 |
| 46 | if method == "eagle": |
| 47 | assert self.model is not None, ( |
| 48 | "model should not be None when method is eagle" |
| 49 | ) |
| 50 | kwargs["architectures"] = [ |
| 51 | f"Eagle{arch}" if not arch.startswith("Eagle") else arch |
| 52 | for arch in self.model.architectures |
| 53 | ] |
| 54 | |
| 55 | elif method == "eagle3": |
| 56 | assert self.model is not None, ( |
| 57 | "model should not be None when method is eagle3" |
| 58 | ) |
| 59 | kwargs["architectures"] = [ |
| 60 | arch |
| 61 | if arch.startswith("Eagle3") or arch.endswith("Eagle3") |
| 62 | else f"Eagle3{arch}" |
| 63 | for arch in self.model.architectures |
| 64 | ] |
| 65 | elif method == "dflash": |
| 66 | assert self.model is not None, ( |
| 67 | "model should not be None when method is dflash" |
| 68 | ) |
| 69 | kwargs["architectures"] = [ |
| 70 | arch |
| 71 | if arch.startswith("DFlash") |
| 72 | or arch.endswith("DFlash") |
| 73 | or arch == "LiLiCorrDraftModel" |
| 74 | else f"DFlash{arch}" |
| 75 | for arch in self.model.architectures |
| 76 | ] |
| 77 | else: |
| 78 | raise ValueError( |
| 79 | f"Invalid method {method}. Supported methods are " |
| 80 | "eagle, eagle3, and dflash." |
| 81 | ) |
| 82 | |
| 83 | super().__init__(**kwargs) |
| 84 | |
| 85 | if self.model is not None: |
| 86 | for k, v in self.model.to_dict().items(): |
| 87 | if k not in kwargs: |
| 88 | setattr(self, k, v) |
| 89 | |
| 90 | @classmethod |
| 91 | def from_pretrained( |
| 92 | cls, |
| 93 | pretrained_model_name_or_path: str | os.PathLike, |
| 94 | **kwargs, |
| 95 | ) -> "EAGLEConfig": |
| 96 | config_dict, kwargs = cls.get_config_dict( |
| 97 | pretrained_model_name_or_path, **without_trust_remote_code(kwargs) |
| 98 | ) |
| 99 | return cls.from_dict(config_dict, **kwargs) |
| 100 | |
| 101 | def to_json_string(self, use_diff: bool = True) -> str: |
| 102 | # we override use_diff to False as initializing |
| 103 | # EAGLEConfig with default arguments is not supported |
| 104 | del use_diff |
| 105 | return super().to_json_string(use_diff=False) |