Workflow Step with Gemini Model Not Finishing Properly, Missing finish_reason and Usage Data

[nguyenphan]

Self Checks

• [X] This is only for bug report, if you would like to ask a question, please head to Discussions.
• [X] I have searched for existing issues search for existing issues, including closed ones.
• [X] I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [X] [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:）
• [X] Please do not modify this template :) and fill in all the required fields.

Dify version

0.10.1

Cloud or Self Hosted

Cloud

Steps to reproduce

1. Create a workflow using Dify with a step that utilizes the Gemini model (gemini-1.5-pro-latest).
2. Configure the step to add translation notes based on an input list of translations and a reference image.
3. Run the workflow with the above step.
4. Observe that the step returns a status of succeeded, but the finish_reason is null and the usage data (e.g., total_tokens) is zero.

✔️ Expected Behavior

The workflow step should complete with a proper finish_reason and accurate usage data, similar to the behavior seen with other models like gpt4o-mini. The usage object should reflect the tokens used and provide correct pricing information.

❌ Actual Behavior

• The workflow step runs and returns a status of succeeded, but: • The finish_reason is null. • All values in the usage object (e.g., prompt_tokens, completion_tokens, total_tokens, total_price) are zero. • The response contains incomplete output, cutting off mid-sentence. • This issue occurs only when using the Gemini model (gemini-1.5-pro-latest). The same workflow runs correctly when using the gpt4o-mini model.

Example of the node output:

{
    "id": "391decfc-743c-4171-a7a5-993b92ea3b15",
    "index": 1,
    "predecessor_node_id": null,
    "node_id": "1729500375254",
    "node_type": "llm",
    "title": "Add translation notes",
    "inputs": null,
    "process_data": {
        "model_mode": "chat",
        "prompts": [
            {
                "role": "system",
                "text": "```xml\n<instruction>\n  <task_description>\n    ***  </task_description>\n\n  <instructions>\n   *** \n  </instructions>\n\n  <examples>\n    <example>\n      Input:\n      Translations:\n{text-1}\nLorem ipsum dolor sit amet, consectetur adipiscing elit.\n\n{text-2}\nSed do eiusmod tempor incididunt ut labore et dolore magna aliqua.\n\n      Output:\n{text-1}\nLorem ipsum dolor sit amet, consectetur adipiscing elit.\n[Note: The English text at the top should be translated concisely to fit within the design. Consider using a shorter tagline that captures the essence of the message, such as \\\"Adipiscing elit.\\\" or \\\"Dolor sit amet.\\\"]\n\n{text-2}\nSed do eiusmod tempor incididunt ut labore et dolore magna aliqua.\n[Note: No change needed]\n    </example>\n  </examples>\n</instruction>\n```",
                "files": []
            },
            {
                "role": "user",
                "text": "Image: refer to the uploaded image file\nCurrent translations:\n{text-1}\\nWater-based paint crib\\n\\n{text-2}\\nOrdinary wooden crib\\n\\n{text-3}\\nDangerous for babies to bite and put in mouth\\n\\n{text-4}\\nSingle function, not very practical\\n\\n{text-5}\\nIron folding bed\\n\\n{text-6}\\nComposite board bed\\n\\n{text-7}\\nProne to rust\\n\\n{text-8}\\nIron parts are prone to rust and endanger health\\n\\n{text-9}\\nContains formaldehyde\\n\\n{text-10}\\nThe bed board contains glue and formaldehyde, which seriously exceeds the standard\\n\\n{text-11}\\nEasy to shake\\n\\n{text-12}\\nShaky and unsafe, the bed body tube is thin\\n\\n{text-13}\\nEasy to deform\\n\\n{text-14}\\nProne to dampness, illness, not waterproof, easy to deform\\n\\n{text-15}\\nMultifunctional four-in-one\\nOne bed meets multiple needs",
                "files": []
            }
        ],
        "model_provider": "google",
        "model_name": "gemini-1.5-pro-latest"
    },
    "outputs": {
        "text": "{text-1}\nWater-based paint crib\n[Note: No change needed]\n\n{text-2}\nOrdinary wooden crib\n[Note:",
        "usage": {
            "prompt_tokens": 0,
            "prompt_unit_price": "0.0",
            "prompt_price_unit": "0.0",
            "prompt_price": "0.0",
            "completion_tokens": 0,
            "completion_unit_price": "0.0",
            "completion_price_unit": "0.0",
            "completion_price": "0.0",
            "total_tokens": 0,
            "total_price": "0.0",
            "currency": "USD",
            "latency": 0.0
        },
        "finish_reason": null
    },
    "status": "succeeded",
    "error": null,
    "elapsed_time": 1.6948861805722117,
    "execution_metadata": {
        "total_tokens": 0,
        "total_price": "0.0",
        "currency": "USD"
    },
    "extras": {},
    "created_at": 1729848840,
    "created_by_role": "account",
    "created_by_account": {
        "id": "3fe4603a-f8a6-4de9-a92d-111172caa548",
        "name": "Dify",
        "email": "[email protected]"
    },
    "created_by_end_user": null,
    "finished_at": 1729848840
}

This issue started occurring after the v0.10 update and only affects workflows using the Gemini model.