[bug-fix] GLM OCR Patch Merger context_dim (#37962)

Signed-off-by: JaredforReal <w13431838023@gmail.com>
This commit is contained in:
Jared Wen
2026-03-26 05:11:21 -07:00
committed by GitHub
parent dcdc145893
commit 757eafcf37
2 changed files with 14 additions and 4 deletions
+6 -1
View File
@@ -38,7 +38,10 @@ import torch.nn as nn
import torch.nn.functional as F
from einops import rearrange
from transformers import BatchFeature, Glm4vProcessor
from transformers.models.glm4v.configuration_glm4v import Glm4vVisionConfig
from transformers.models.glm4v.configuration_glm4v import (
Glm4vTextConfig,
Glm4vVisionConfig,
)
from transformers.models.glm4v.image_processing_glm4v import (
Glm4vImageProcessor,
smart_resize,
@@ -604,6 +607,7 @@ class Glm4vVisionEmbeddings(nn.Module):
class Glm4vVisionTransformer(nn.Module):
def __init__(
self,
text_config: Glm4vTextConfig,
vision_config: Glm4vVisionConfig,
norm_eps: float = 1e-6,
quant_config: QuantizationConfig | None = None,
@@ -1424,6 +1428,7 @@ class Glm4vForConditionalGeneration(
with self._mark_tower_model(vllm_config, {"image", "video"}):
self.visual = Glm4vVisionTransformer(
config.text_config,
config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-5),
quant_config=quant_config,