What causes "Empty response from {self.name} (model={self.model})" in MemPalace/mempalace?

Retry once — empty generations are often transient sampling artifacts

MemPalace/mempalace · error · LLMError

Empty response from {self.name} (model={self.model})

Error message

Empty response from {self.name} (model={self.model})

What it means

LLMError raised by OpenAICompatProvider.classify() when choices[0].message.content exists but is empty. The request and parsing succeeded; the model simply generated zero content tokens.

Source

Thrown at mempalace/llm_client.py:360

            "model": self.model,
            "messages": [
                {"role": "system", "content": system},
                {"role": "user", "content": user},
            ],
            "temperature": 0.1,
        }
        if json_mode:
            body["response_format"] = {"type": "json_object"}
        headers = {}
        if self.api_key:
            headers["Authorization"] = f"Bearer {self.api_key}"
        data = _http_post_json(self._resolve_url(), body, headers=headers, timeout=self.timeout)
        try:
            text = data["choices"][0]["message"]["content"]
        except (KeyError, IndexError, TypeError) as e:
            raise LLMError(f"Unexpected response shape: {e}") from e
        if not text:
            raise LLMError(f"Empty response from {self.name} (model={self.model})")
        return LLMResponse(text=text, model=self.model, provider=self.name, raw=data)


# ==================== ANTHROPIC ====================


class AnthropicProvider(LLMProvider):
    name = "anthropic"
    DEFAULT_ENDPOINT = "https://api.anthropic.com"
    API_VERSION = "2023-06-01"

    def __init__(
        self,
        model: str,
        api_key: Optional[str] = None,
        endpoint: Optional[str] = None,
        timeout: int = 120,
        **_: object,

View on GitHub (pinned to 06cb6987f0)

Solutions

Retry once — empty generations are often transient sampling artifacts
Relax json_mode or lower temperature to get actual content
Use an instruct-tuned model with solid JSON support
Raise max_tokens/context so the model has room for a final answer

Defensive patterns

Strategy: retry

Try / catch

from mempalace.llm_client import LLMError

for attempt in range(2):
    try:
        return provider.classify(system, user, json_mode=True)
    except LLMError as e:
        if "Empty response" in str(e) and attempt == 0:
            continue  # empty generation — retry once
        raise

Prevention

Prefer json-mode-capable instruct models for structured tasks
Set temperature low (the provider already uses 0.1) and prompts explicit
Log raw responses on failure to distinguish empty content from shape issues

When it happens

Trigger: Models that return only tool calls or reasoning with no final content; content filtered/empty under response_format json_object when the model cannot produce valid JSON; degenerate outputs from too-small num_ctx or max_tokens.

Common situations: json_mode=True with models that ignore or choke on JSON instructions; reasoning models spending the entire budget on hidden reasoning; nearly-empty prompts producing one blank token.

Related errors

AI-assisted analysis of MemPalace/mempalace@06cb6987f0 (2026-08-15). Data as JSON: /api/errors/7a9d802073d98a12. Report an issue: GitHub.