Skip to content

server: fix json_schema response_format handling #4

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
superbarne merged 2 commits into from
Mar 15, 2026
Merged

server: fix json_schema response_format handling #4

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4336918
Initial plan
Copilot Mar 15, 2026
9680d0d
server: fix json_schema response_format handling (merge ggml-org/llam…
Copilot Mar 15, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 7 additions & 2 deletions tools/server/server-common.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -927,10 +927,15 @@ json oaicompat_chat_params_parse(
if (response_type == "json_object") {
json_schema = json_value(response_format, "schema", json::object());
} else if (response_type == "json_schema") {
// https://platform.openai.com/docs/api-reference/chat/create#chat-create-response_format
// OpenAI expects: response_format.json_schema.schema
auto schema_wrapper = json_value(response_format, "json_schema", json::object());
json_schema = json_value(schema_wrapper, "schema", json::object());
if (!schema_wrapper.contains("schema")) {
throw std::invalid_argument("response_format type \"json_schema\" requires \"json_schema.schema\" to be set");
}
json_schema = schema_wrapper.at("schema");
} else if (!response_type.empty() && response_type != "text") {
throw std::invalid_argument("response_format type must be one of \"text\" or \"json_object\", but got: " + response_type);
throw std::invalid_argument("response_format type must be one of \"text\", \"json_object\", or \"json_schema\", but got: " + response_type);
}
}

Expand Down
31 changes: 31 additions & 0 deletions tools/server/tests/unit/test_chat_completion.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -176,12 +176,18 @@ def test_apply_chat_template():
({"type": "json_object", "schema": {"const": "42"}}, 6, "\"42\""),
({"type": "json_object", "schema": {"items": [{"type": "integer"}]}}, 10, "[ -3000 ]"),
({"type": "json_schema", "json_schema": {"schema": {"const": "foooooo"}}}, 10, "\"foooooo\""),
# json_schema with name field (OpenAI-style)
({"type": "json_schema", "json_schema": {"name": "test", "schema": {"const": "bar"}, "strict": True}}, 6, "\"bar\""),
({"type": "json_object"}, 10, "(\\{|John)+"),
({"type": "sound"}, 0, None),
# invalid response format (expected to fail)
({"type": "json_object", "schema": 123}, 0, None),
({"type": "json_object", "schema": {"type": 123}}, 0, None),
({"type": "json_object", "schema": {"type": "hiccup"}}, 0, None),
# json_schema missing required json_schema.schema field (should fail)
({"type": "json_schema", "json_schema": {"name": "test"}}, 0, None),
({"type": "json_schema", "json_schema": {}}, 0, None),
({"type": "json_schema"}, 0, None),
])
def test_completion_with_response_format(response_format: dict, n_predicted: int, re_content: str | None):
global server
Expand All @@ -203,6 +209,31 @@ def test_completion_with_response_format(response_format: dict, n_predicted: int
assert "error" in res.body


@pytest.mark.parametrize("response_format,expected_error_message", [
# json_schema type requires json_schema.schema to be set
({"type": "json_schema", "json_schema": {"name": "test"}}, "json_schema.schema"),
({"type": "json_schema", "json_schema": {}}, "json_schema.schema"),
({"type": "json_schema"}, "json_schema.schema"),
# invalid response_format type should mention valid options
({"type": "invalid_type"}, "json_schema"),
])
def test_response_format_error_messages(response_format: dict, expected_error_message: str):
"""Test that invalid response_format configurations return helpful error messages."""
global server
server.start()
res = server.make_request("POST", "/chat/completions", data={
"max_tokens": 10,
"messages": [
{"role": "user", "content": "test"},
],
"response_format": response_format,
})
assert res.status_code == 400
assert "error" in res.body
assert expected_error_message in res.body["error"]["message"], \
f"Expected '{expected_error_message}' in error message, got: {res.body['error']['message']}"


@pytest.mark.parametrize("jinja,json_schema,n_predicted,re_content", [
(False, {"const": "42"}, 6, "\"42\""),
(True, {"const": "42"}, 6, "\"42\""),
Expand Down