YogeLiu/doc_processer
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix:glm max tokens

Browse Source
This commit is contained in:
liuyuanchuang
2026-02-07 21:38:41 +08:00
parent 1a4d54ce34
commit 5504bbbf1e
2 changed files with 65 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files

71
app/services/ocr_service.py
View File

@@ -532,50 +532,51 @@ class GLMOCRService(OCRServiceBase):
Returns:
Dict with 'latex', 'markdown', 'mathml', 'mml' keys.
Raises:
RuntimeError: If recognition fails (preserves original exception for fallback handling).
"""
try:
# Add padding to image
padded_image = self.image_processor.add_padding(image)
# Add padding to image
padded_image = self.image_processor.add_padding(image)
# Encode image to base64
success, encoded_image = cv2.imencode(".png", padded_image)
if not success:
raise RuntimeError("Failed to encode image")
# Encode image to base64
success, encoded_image = cv2.imencode(".png", padded_image)
if not success:
raise RuntimeError("Failed to encode image")
image_base64 = base64.b64encode(encoded_image.tobytes()).decode("utf-8")
image_url = f"data:image/png;base64,{image_base64}"
image_base64 = base64.b64encode(encoded_image.tobytes()).decode("utf-8")
image_url = f"data:image/png;base64,{image_base64}"
# Call OpenAI-compatible API with formula recognition prompt
prompt = "Formula Recognition:"
messages = [{"role": "user", "content": [{"type": "image_url", "image_url": {"url": image_url}}, {"type": "text", "text": prompt}]}]
# Call OpenAI-compatible API with formula recognition prompt
prompt = "Formula Recognition:"
messages = [{"role": "user", "content": [{"type": "image_url", "image_url": {"url": image_url}}, {"type": "text", "text": prompt}]}]
response = self.openai_client.chat.completions.create(
model="glm-ocr",
messages=messages,
temperature=0.0,
)
# Don't catch exceptions here - let them propagate for fallback handling
response = self.openai_client.chat.completions.create(
model="glm-ocr",
messages=messages,
temperature=0.0,
)
markdown_content = response.choices[0].message.content
markdown_content = response.choices[0].message.content
# Process LaTeX delimiters
if markdown_content.startswith(r"\[") or markdown_content.startswith(r"\("):
markdown_content = markdown_content.replace(r"\[", "$$").replace(r"\(", "$$")
markdown_content = markdown_content.replace(r"\]", "$$").replace(r"\)", "$$")
elif not markdown_content.startswith("$$") and not markdown_content.startswith("$"):
markdown_content = f"$${markdown_content}$$"
# Process LaTeX delimiters
if markdown_content.startswith(r"\[") or markdown_content.startswith(r"\("):
markdown_content = markdown_content.replace(r"\[", "$$").replace(r"\(", "$$")
markdown_content = markdown_content.replace(r"\]", "$$").replace(r"\)", "$$")
elif not markdown_content.startswith("$$") and not markdown_content.startswith("$"):
markdown_content = f"$${markdown_content}$$"
# Apply postprocessing
markdown_content = _postprocess_markdown(markdown_content)
convert_result = self.converter.convert_to_formats(markdown_content)
# Apply postprocessing
markdown_content = _postprocess_markdown(markdown_content)
convert_result = self.converter.convert_to_formats(markdown_content)
return {
"latex": convert_result.latex,
"mathml": convert_result.mathml,
"mml": convert_result.mml,
"markdown": markdown_content,
}
except Exception as e:
raise RuntimeError(f"GLM formula recognition failed: {e}") from e
return {
"latex": convert_result.latex,
"mathml": convert_result.mathml,
"mml": convert_result.mml,
"markdown": markdown_content,
}
def recognize(self, image: np.ndarray) -> dict:
"""Recognize content using GLM-4V.