sgl-project/sglang · error · ValueError

schema_ is required for json_schema response format request.

Error message

schema_ is required for json_schema response format request.

What it means

A completions request declared response_format={"type":"json_schema"} but the nested json_schema object had no schema_ field (the actual JSON Schema payload). SGLang requires the schema to convert into sampling_params.json_schema for constrained decoding.

Source

Thrown at python/sglang/srt/entrypoints/openai/serving_completions.py:178

            "repetition_penalty": request.repetition_penalty,
            "regex": request.regex,
            "json_schema": request.json_schema,
            "ebnf": request.ebnf,
            "n": request.n,
            "no_stop_trim": request.no_stop_trim,
            "ignore_eos": request.ignore_eos,
            "skip_special_tokens": request.skip_special_tokens,
            "logit_bias": request.logit_bias,
            "custom_params": request.custom_params,
            "sampling_seed": request.seed,
        }

        # Handle response_format constraints
        if request.response_format and request.response_format.type == "json_schema":
            json_schema = request.response_format.json_schema
            schema = getattr(json_schema, "schema_", None)
            if schema is None:
                raise ValueError(
                    "schema_ is required for json_schema response format request."
                )
            sampling_params["json_schema"] = convert_json_schema_to_str(schema)
        elif request.response_format and request.response_format.type == "json_object":
            sampling_params["json_schema"] = '{"type": "object"}'
        elif (
            request.response_format and request.response_format.type == "structural_tag"
        ):
            sampling_params["structural_tag"] = convert_json_schema_to_str(
                request.response_format.model_dump(by_alias=True)
            )

        return sampling_params

    async def _handle_streaming_request(
        self,
        adapted_request: GenerateReqInput,
        request: CompletionRequest,

View on GitHub (pinned to 0132848349)

Solutions

1. Include the schema: response_format.json_schema.schema_ = {...}
2. If using OpenAI SDK types, populate ResponseFormatJsonSchema(schema_={...})
3. For loose object constraints, use type 'json_object' instead

Example fix

# before
{"response_format": {"type": "json_schema", "json_schema": {"name": "user"}}}
# after
{"response_format": {"type": "json_schema", "json_schema": {"name": "user", "schema_": {"type": "object", "properties": {"name": {"type": "string"}}, "required": ["name"]}}}}

if rf.get('type') == 'json_schema':
    assert rf['json_schema'].get('schema_'), 'schema_ required'

def valid_json_schema_rf(rf) -> bool:
    js = rf.get('json_schema') or {}
    return rf.get('type') != 'json_schema' or isinstance(js.get('schema_'), (dict, str)) and js.get('schema_') is not None

When it happens

Trigger: POST /v1/completions with response_format.type == 'json_schema' and json_schema.schema_ missing/None (e.g. only json_schema.name given).

Common situations: Clients building the request dict by hand and naming the schema field 'schema' instead of 'schema_'; partial request templating that leaves the schema out; version drift where clients sent the bare schema.

Understand the failure class

Background: Schema validation failed / invalid input schema: payload rejected because its shape doesn't match the expected schema — this error's family across 28 libraries.

Related errors

• Kimi K3 additional parameter schema accepts no values
• Kimi K3 tool parameters 'properties' must be an object
• Kimi K3 tool parameters 'required' must be a string list
• Kimi K3 required parameters are missing schemas: {sorted(mis
• Kimi K3 tool property schemas must be JSON schemas

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/ae68d8d7450fb3d8. Report an issue: GitHub.