From fbe3f0120a5e786a1459c982c72185311c78a276 Mon Sep 17 00:00:00 2001 From: Harry Mellor <19981378+hmellor@users.noreply.github.com> Date: Fri, 27 Feb 2026 14:13:27 +0000 Subject: [PATCH] Revert "Add GlmOcrConfig for GLM-OCR model type recognition" (#35512) --- vllm/transformers_utils/config.py | 1 - vllm/transformers_utils/configs/__init__.py | 4 - vllm/transformers_utils/configs/glm_ocr.py | 91 --------------------- 3 files changed, 96 deletions(-) delete mode 100644 vllm/transformers_utils/configs/glm_ocr.py diff --git a/vllm/transformers_utils/config.py b/vllm/transformers_utils/config.py index f5adb171b4..00129d52ea 100644 --- a/vllm/transformers_utils/config.py +++ b/vllm/transformers_utils/config.py @@ -82,7 +82,6 @@ _CONFIG_REGISTRY: dict[str, type[PretrainedConfig]] = LazyConfigDict( deepseek_v32="DeepseekV3Config", flex_olmo="FlexOlmoConfig", funaudiochat="FunAudioChatConfig", - glm_ocr="GlmOcrConfig", hunyuan_vl="HunYuanVLConfig", isaac="IsaacConfig", kimi_linear="KimiLinearConfig", diff --git a/vllm/transformers_utils/configs/__init__.py b/vllm/transformers_utils/configs/__init__.py index 761f96a574..541bc4de61 100644 --- a/vllm/transformers_utils/configs/__init__.py +++ b/vllm/transformers_utils/configs/__init__.py @@ -28,8 +28,6 @@ _CLASS_TO_MODULE: dict[str, str] = { "FlexOlmoConfig": "vllm.transformers_utils.configs.flex_olmo", "FunAudioChatConfig": "vllm.transformers_utils.configs.funaudiochat", "FunAudioChatAudioEncoderConfig": "vllm.transformers_utils.configs.funaudiochat", - "GlmOcrConfig": "vllm.transformers_utils.configs.glm_ocr", - "GlmOcrVisionConfig": "vllm.transformers_utils.configs.glm_ocr", "HunYuanVLConfig": "vllm.transformers_utils.configs.hunyuan_vl", "HunYuanVLTextConfig": "vllm.transformers_utils.configs.hunyuan_vl", "HunYuanVLVisionConfig": "vllm.transformers_utils.configs.hunyuan_vl", @@ -85,8 +83,6 @@ __all__ = [ "FlexOlmoConfig", "FunAudioChatConfig", "FunAudioChatAudioEncoderConfig", - "GlmOcrConfig", - "GlmOcrVisionConfig", "HunYuanVLConfig", "HunYuanVLTextConfig", "HunYuanVLVisionConfig", diff --git a/vllm/transformers_utils/configs/glm_ocr.py b/vllm/transformers_utils/configs/glm_ocr.py deleted file mode 100644 index 43656d276e..0000000000 --- a/vllm/transformers_utils/configs/glm_ocr.py +++ /dev/null @@ -1,91 +0,0 @@ -# SPDX-License-Identifier: Apache-2.0 -# SPDX-FileCopyrightText: Copyright contributors to the vLLM project -from __future__ import annotations - -from typing import Any - -from transformers.configuration_utils import PretrainedConfig - - -class GlmOcrVisionConfig(PretrainedConfig): - model_type = "glm_ocr_vision" - - def __init__( - self, - hidden_size: int = 1024, - depth: int = 24, - num_heads: int = 16, - attention_bias: bool = True, - intermediate_size: int = 4096, - hidden_act: str = "silu", - hidden_dropout_prob: float = 0.0, - initializer_range: float = 0.02, - image_size: int = 336, - in_channels: int = 3, - patch_size: int = 14, - out_hidden_size: int = 1536, - rms_norm_eps: float = 1e-5, - spatial_merge_size: int = 2, - temporal_patch_size: int = 2, - **kwargs: Any, - ): - super().__init__(**kwargs) - self.hidden_size = hidden_size - self.depth = depth - self.num_heads = num_heads - self.attention_bias = attention_bias - self.intermediate_size = intermediate_size - self.hidden_act = hidden_act - self.hidden_dropout_prob = hidden_dropout_prob - self.initializer_range = initializer_range - self.image_size = image_size - self.in_channels = in_channels - self.patch_size = patch_size - self.out_hidden_size = out_hidden_size - self.rms_norm_eps = rms_norm_eps - self.spatial_merge_size = spatial_merge_size - self.temporal_patch_size = temporal_patch_size - - -class GlmOcrConfig(PretrainedConfig): - model_type = "glm_ocr" - - def __init__( - self, - text_config: dict | None = None, - vision_config: dict | None = None, - image_start_token_id: int = 59256, - image_end_token_id: int = 59257, - video_start_token_id: int = 59258, - video_end_token_id: int = 59259, - image_token_id: int = 59280, - video_token_id: int = 59281, - **kwargs: Any, - ): - super().__init__(**kwargs) - self.image_start_token_id = image_start_token_id - self.image_end_token_id = image_end_token_id - self.video_start_token_id = video_start_token_id - self.video_end_token_id = video_end_token_id - self.image_token_id = image_token_id - self.video_token_id = video_token_id - self.vision_config = GlmOcrVisionConfig(**(vision_config or {})) - - if isinstance(text_config, dict): - from transformers import AutoConfig - - model_type = text_config.get("model_type", "chatglm") - self.text_config = AutoConfig.for_model(model_type, **text_config) - elif text_config is None: - from transformers import AutoConfig - - self.text_config = AutoConfig.for_model("chatglm") - else: - self.text_config = text_config - - def get_text_config(self) -> PretrainedConfig: - return self.text_config - - def save_pretrained(self, save_directory, **kwargs): - self._auto_class = None - super().save_pretrained(save_directory, **kwargs)