Bulk Create Dataset Logs

Bulk create dataset logs from array of unified format data

Endpoint: POST /api/datasets/{dataset_id}/logs/bulk/

Args (POST body): - logs (array, required): List of log objects in unified format Each log object contains: - input (any): The input data (required) - can be any type (dict, list, string, etc.) - output (any, optional): The output data - can be any type - metadata (object, optional): Additional metadata fields (model, log_type, etc.) - metrics (object, optional): Metric fields (tokens, cost, latency, etc.)

Example Request (Recommended - Top-level expected_output): json { "logs": [ { "input": "What is your return policy?", "expected_output": "You can return within 30 days", "metadata": {"category": "support"} }, { "input": "How do I reset my password?", "expected_output": "Click Forgot Password on login page", "metadata": {"category": "support"} } ] }

Example Request (Legacy - Nested expected_output, auto-extracted): Frontend parses CSV where expected_output is nested in input: json { "logs": [ { "input": { "user_query": "What is your return policy?", "expected_output": "You can return within 30 days", "category": "support" } } ] } Note: Nested expected_output is automatically extracted to top-level field.

Example Request (Direct API usage with expected_output): json { "logs": [ { "input": "What is AI?", "expected_output": "AI is artificial intelligence", "output": "", "metadata": {"category": "qa", "model": "gpt-4"} }, { "input": [{"role": "user", "content": "Hello"}], "expected_output": "A friendly greeting", "output": {"role": "assistant", "content": "Hi there!"}, "metrics": {"tokens": 10, "cost": 0.0001} } ] }

Field Descriptions: - input: The input to be processed (can be string, dict, or array) - expected_output: Expected/ground truth output for evaluation (optional) - output: Actual output from LLM or system (populated during experiments) - metadata: Additional context fields - metrics: Performance metrics (tokens, cost, latency)

Returns (POST 201): json { "success_count": 95, "error_count": 5, "errors": [ {"index": 3, "error": "Invalid input format"}, {"index": 7, "error": "Missing required field"} ] }

Notes: - For UI users: Frontend parses CSV and sends array in unified format - For API users: Send JSON array directly, no CSV conversion needed - Each row of CSV becomes an "input" object in the dataset log - Plan limits are enforced (current dataset log count + new logs <= limit) - Errors are returned for individual logs that fail, successful ones are still created