harry0703/MoneyPrinterTurbo · error · ValueError

[{llm_provider}] returned empty response

Error message

[{llm_provider}] returned empty response

What it means

Raised in the litellm branch when litellm.completion() returns a falsy response object. litellm normally raises its own exceptions on failure, so a falsy return is an SDK-contract anomaly — most often seen with drop_params=True silently stripping parameters on an incompatible provider, or mocked litellm in tests.

Source

Thrown at app/services/llm.py:323

            )
            return _extract_chat_completion_text(response, llm_provider)

        if adapter == "litellm":
            import litellm

            if not model_name:
                raise ValueError(
                    f"{llm_provider}: model_name is not set, please set it in the config.toml file."
                )

            response = litellm.completion(
                model=model_name,
                messages=[{"role": "user", "content": prompt}],
                drop_params=True,
            )

            if not response:
                raise ValueError(f"[{llm_provider}] returned empty response")
            if not getattr(response, "choices", None):
                raise ValueError(f"[{llm_provider}] returned empty response")

            return _extract_chat_completion_text(response, llm_provider)

        if adapter == "azure":
            # Azure OpenAI SDK 使用 `azure_endpoint` 和 `api_version` 生成专用请求地址，
            # 不能继续复用下面普通 OpenAI-compatible 的 `base_url` 初始化逻辑。
            # 这里在 Azure 分支内完成请求并立即返回，避免客户端被后续 fallback
            # 覆盖，导致用户配置的 Azure 凭证通过校验但实际请求没有被使用。
            logger.info(f"requesting azure chat completion, model: {model_name}")
            client = AzureOpenAI(
                api_key=api_key,
                api_version=api_version,
                azure_endpoint=base_url,
            )
            response = client.chat.completions.create(
                model=model_name, messages=[{"role": "user", "content": prompt}]

View on GitHub (pinned to 1f9f19c202)

Solutions

1. Check litellm logs/verbose output (litellm.set_verbose=True or check logger) for the routing failure that preceded the empty return
2. Pin a known-good litellm version: the model-string routing contract changed between major versions; reinstall in a clean venv
3. Verify the model string format for the target provider (e.g. 'groq/llama-3.1-8b-instant') and that the matching credentials env vars (GROQ_API_KEY, etc.) are set
4. In tests, make the litellm.completion mock return a real ModelResponse with choices

Example fix

# before
response = litellm.completion(model=model_name, messages=[...], drop_params=True)
if not response:
    raise ValueError(f"[{llm_provider}] returned empty response")

# after
response = litellm.completion(model=model_name, messages=[...], drop_params=True)
if not response:
    raise ValueError(
        f"[{llm_provider}] litellm returned empty response for model "
        f"'{model_name}'; check provider prefix and credentials"
    )

import litellm
if not hasattr(litellm, "completion"):
    raise RuntimeError("litellm not installed")
# verify model routes: litellm.get_model_info or a 1-token dry run
resp = litellm.completion(model=model_name, messages=[{"role":"user","content":"hi"}], max_tokens=1)
assert resp and resp.choices

except ValueError: — check the model string and credentials once, then fail; retrying an empty return from the same model usually repeats

When it happens

Trigger: litellm.completion(model=..., messages=[...], drop_params=True) evaluating False under `if not response:` — e.g. a litellm version returning None on certain internal routing failures, unsupported model/provider strings falling through, or a test mock returning None.

Common situations: Upgrading litellm major versions (its return contract and exception behavior changed across releases); passing a model string litellm cannot resolve to any provider; unit tests patching litellm.completion without a return value; edge failure modes inside litellm's router.

Related errors

• {llm_provider}: unsupported llm provider
• {llm_provider}: api_key is not set, please set it in the con
• {llm_provider}: model_name is not set, please set it in the
• {llm_provider}: base_url is not set, please set it in the co
• {llm_provider}: {field.config_suffix} is not set, please set

AI-assisted analysis of harry0703/MoneyPrinterTurbo@1f9f19c202 (2026-08-14). Data as JSON: /api/errors/41577e53477edc63. Report an issue: GitHub.