上传文件至Gemini 1.5 Vertex AI时触发400前置检查失败错误排查

尝试通过Google Cloud的Vertex AI向Gemini 1.5模型上传文件，尽管文件限制为30MB，但11MB的文件触发错误：400 Precondition check failed，仅返回该错误信息，请求协助排查问题原因。

代码实现 #

async def async_get_model_output(prompt, filepaths, content_type):
    
    vertexai.init(project=os.getenv("PROJECT_ID"), location="us-central1")
    model = GenerativeModel(model_name="gemini-1.5-flash-001")
    #Figure 
    parts = [Part.from_uri(filepath["gcs_link"], mime_type=filepath["mimetype"]) for filepath in filepaths]
    parts.append(prompt)
    # Create a GenerationConfig object with the specified parameters
    generation_config = GenerationConfig(
        temperature=0.3,
        top_p=0.3,
        max_output_tokens = 5000
        # You can add other parameters here as needed
    )
#BLOCK_ONLY_HIGH
    safety_settings = [
        SafetySetting(
        category=HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
        threshold=HarmBlockThreshold.BLOCK_ONLY_HIGH,
    ),
    SafetySetting(
        category=HarmCategory.HARM_CATEGORY_HARASSMENT,
        threshold=HarmBlockThreshold.BLOCK_ONLY_HIGH,
    ),
    SafetySetting(
        category=HarmCategory.HARM_CATEGORY_HATE_SPEECH,
        threshold=HarmBlockThreshold.BLOCK_ONLY_HIGH,
    ),
    SafetySetting(
        category=HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT,
        threshold=HarmBlockThreshold.BLOCK_ONLY_HIGH,
    ),
        # Add other categories as needed
    ]

    for attempt in range(3):
        print(f'Attempt {attempt + 1} to get model output')
        try:
            token_count = model.count_tokens(parts)
            print(f"Token count: {token_count}")
            print(f"##{content_type}## DEBUG: INVOKING VERTEX AI")
            response = await asyncio.to_thread(
                model.generate_content,
                parts,
                generation_config=generation_config,
                safety_settings=safety_settings
            )
            logger.debug(f"Received response: {response}")
            json_match = re.search(r'\{[\s\S]*\}', response.text)

            if json_match:
                json_string = json_match.group(0)
                try:
                    study_guide_data = json.loads(json_string)
                    print(f"##{content_type}## DEBUG CONTENT SUCCESS: Successfully Generated content and parsed JSON")
                    return json_string
                except json.JSONDecodeError as e:
                    print(f"##{content_type}##JSON decode error: {e}")
                    print(f"##{content_type}##Extracted JSON string:", json_string)
                    # Continue to next attempt
            else:
                print(f"##{content_type}## No JSON object found in the response")
        except Exception as e:
            error_message = str(e)
            truncated_message = error_message[:197] + "..." if len(error_message) > 200 else error_message
            print(f"##{content_type}## Unexpected error during attempt {attempt + 1}: {error_message}")
        
        # If we reach here, the current attempt failed. We'll continue to the next one.
        print(f"##{content_type}## Attempt {attempt + 1} failed. Moving to next attempt.")

        
        if attempt < 2:  # Don't sleep after the last attempt
            print(f"##{content_type}## Waiting 1 seconds before next attempt...")
            await asyncio.sleep(1)
    # If we've exhausted all attempts without success
    raise Exception(f"##{content_type}## Failed to get valid JSON after 3 attempts")

可能的排查方向 #

• GCS权限验证：确认GCS存储桶已为Vertex AI服务账号（service-[PROJECT_NUMBER]@gcp-sa-vertexai.iam.gserviceaccount.com）分配roles/storage.objectViewer角色，确保模型能读取文件。
• MIME类型一致性检查：核对filepath["mimetype"]与实际文件类型是否匹配，比如PDF需设为application/pdf，Word文档设为application/vnd.openxmlformats-officedocument.wordprocessingml.document，错误类型会触发预检查失败。
• 文件完整性验证：重新上传文件到GCS，或尝试用本地文件测试，排除文件损坏导致的预检查失败。
• 区域匹配检查：确认GCS存储桶区域与Vertex AI初始化的us-central1区域一致，跨区域访问可能存在限制。
• Token计数超限排查：查看代码中token_count的输出值，Gemini 1.5 Flash输入Token上限为1M，部分文件（如长文本、高分辨率图片）转Token后可能超限。
• API配额检查：在GCP控制台的配额页面，确认generativeModels.generateContent等相关API的请求次数未超限。

内容的提问来源于stack exchange，提问作者Avik Samanta