From fcd9052179f039f075b3e25a8c1aec8657fc98a0 Mon Sep 17 00:00:00 2001 From: mateo-berri <277851410+mateo-berri@users.noreply.github.com> Date: Tue, 1 Sep 2026 17:10:52 -0700 Subject: [PATCH] feat(proxy): honor model_info.display_name in the Anthropic-shaped /v1/models listing --- litellm/llms/anthropic/common_utils.py | 17 ++-- .../proxy/common_utils/model_listing_utils.py | 25 +++++- litellm/proxy/proxy_server.py | 21 +++-- litellm/router.py | 20 +++++ .../proxy/proxy_server/test_routes_models.py | 78 +++++++++++++++++++ .../test_team_model_name_translation.py | 26 ++++++- tests/test_litellm/test_router.py | 65 ++++++++++++++++ 7 files changed, 240 insertions(+), 12 deletions(-) diff --git a/litellm/llms/anthropic/common_utils.py b/litellm/llms/anthropic/common_utils.py index c60ebd844ba..d23690976ad 100644 --- a/litellm/llms/anthropic/common_utils.py +++ b/litellm/llms/anthropic/common_utils.py @@ -1378,31 +1378,38 @@ def process_anthropic_headers(headers: httpx.Headers | dict) -> dict: return additional_headers -def _anthropic_model_entry(model: ModelInfoResponse, created_at: str) -> Mapping[str, object]: +def _anthropic_model_entry( + model: ModelInfoResponse, created_at: str, display_names: Mapping[str, str] +) -> Mapping[str, object]: return { # mutable-ok: JSON response body, serialized by the route and never mutated "type": "model", "id": model["id"], - "display_name": model["id"], + "display_name": display_names.get(model["id"], model["id"]), "created_at": created_at, "max_input_tokens": model.get("max_input_tokens"), "max_tokens": model.get("max_output_tokens"), } -def create_anthropic_model_list_response(models: Sequence[ModelInfoResponse]) -> Mapping[str, object]: +def create_anthropic_model_list_response( + models: Sequence[ModelInfoResponse], + display_names: Mapping[str, str] = MappingProxyType({}), +) -> Mapping[str, object]: """Build the Anthropic-native /v1/models envelope. Clients that send an anthropic-version header parse the Anthropic Models API shape (type/display_name/created_at plus has_more/first_id/last_id) and filter the list themselves, so every model is returned here. The token limits carry over from the OpenAI-shaped listing, named as the Messages API names them, and - are always present because the vendor shape declares them nullable, not optional + are always present because the vendor shape declares them nullable, not optional. + display_names maps a listed model id to a configured human-readable name; ids + without an entry fall back to the id itself, matching the vendor behavior """ created_at: Final = ( datetime.fromtimestamp(DEFAULT_MODEL_CREATED_AT_TIME, tz=timezone.utc).isoformat().replace("+00:00", "Z") ) data: Final = [ # mutable-ok: JSON response body, serialized by the route and never mutated - _anthropic_model_entry(model, created_at) for model in models + _anthropic_model_entry(model, created_at, display_names) for model in models ] return { # mutable-ok: JSON response body, serialized by the route and never mutated "data": data, diff --git a/litellm/proxy/common_utils/model_listing_utils.py b/litellm/proxy/common_utils/model_listing_utils.py index 9fd24162f7e..213a697b3dd 100644 --- a/litellm/proxy/common_utils/model_listing_utils.py +++ b/litellm/proxy/common_utils/model_listing_utils.py @@ -10,13 +10,36 @@ legacy internal names with `general_settings.use_team_public_model_name: false`. from __future__ import annotations -from collections.abc import Mapping +from collections.abc import Mapping, Sequence +from types import MappingProxyType from typing import TYPE_CHECKING, Final, cast if TYPE_CHECKING: from litellm.router import Router +def configured_display_names( + entries: Sequence[tuple[str, str]], + llm_router: Router | None, +) -> Mapping[str, str]: + """response_id -> configured `model_info.display_name` for the listing entries + that have one. + + Metadata is looked up by each entry's internal lookup id (so team-scoped rows + resolve), while the returned map is keyed by the public response id the + Anthropic-shaped listing is built from. Entries without a configured name are + omitted so the listing falls back to the id itself. + """ + if llm_router is None: + return MappingProxyType({}) + resolved: Final = ( + (response_id, llm_router.get_configured_display_name(lookup_id)) for response_id, lookup_id in entries + ) + return MappingProxyType( + {response_id: display_name for response_id, display_name in resolved if display_name is not None} + ) + + class TeamModelNameTranslator: """Translates internal team routing keys to their public names for the model listing/retrieve responses. Stateless; the live router and general_settings diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py index 2c600667283..4f172ca29b9 100644 --- a/litellm/proxy/proxy_server.py +++ b/litellm/proxy/proxy_server.py @@ -351,7 +351,10 @@ from litellm.proxy.common_utils.load_config_utils import ( get_file_contents_from_s3, ) from litellm.proxy.common_utils.model_deprecation import collect_model_deprecations -from litellm.proxy.common_utils.model_listing_utils import TeamModelNameTranslator +from litellm.proxy.common_utils.model_listing_utils import ( + TeamModelNameTranslator, + configured_display_names, +) from litellm.proxy.common_utils.openai_endpoint_utils import ( remove_sensitive_info_from_deployment, ) @@ -10193,7 +10196,8 @@ async def model_list( # The internal routing key drives the metadata/fallback lookup, while the # public name is what the client sees as the model id. model_data = [] - for response_id, lookup_id in TeamModelNameTranslator.listing_entries(all_models, llm_router, settings): + admin_entries: Final = TeamModelNameTranslator.listing_entries(all_models, llm_router, settings) + for response_id, lookup_id in admin_entries: model_info = create_model_info_response( model_id=lookup_id, provider="openai", @@ -10206,7 +10210,10 @@ async def model_list( if wants_anthropic_format: admin_listing: Final = cast(Sequence[ModelInfoResponse], model_data) # cast-ok: rows built above - return create_anthropic_model_list_response(admin_listing) + return create_anthropic_model_list_response( + admin_listing, + display_names=configured_display_names(admin_entries, llm_router), + ) return dict( data=model_data, @@ -10237,7 +10244,8 @@ async def model_list( # The internal routing key drives the metadata/fallback lookup, while the # public name is what the client sees as the model id. model_data = [] - for response_id, lookup_id in TeamModelNameTranslator.listing_entries(all_models, llm_router, settings): + entries: Final = TeamModelNameTranslator.listing_entries(all_models, llm_router, settings) + for response_id, lookup_id in entries: model_info = create_model_info_response( model_id=lookup_id, provider="openai", @@ -10250,7 +10258,10 @@ async def model_list( if wants_anthropic_format: listing: Final = cast(Sequence[ModelInfoResponse], model_data) # cast-ok: rows built above - return create_anthropic_model_list_response(listing) + return create_anthropic_model_list_response( + listing, + display_names=configured_display_names(entries, llm_router), + ) return dict( data=model_data, diff --git a/litellm/router.py b/litellm/router.py index 471a1116f44..6245f8a6a03 100644 --- a/litellm/router.py +++ b/litellm/router.py @@ -9607,6 +9607,26 @@ class Router: coerce_token_limit(model_info.get("max_output_tokens")), ) + def get_configured_display_name(self, model_name: str) -> "str | None": + """ + Return the display_name explicitly configured in a concrete deployment's + model_info for model_name, via O(1) index lookup. + + Returns None for wildcard-expanded or unknown names, and treats a + non-string or empty configured value as absent rather than failing the + listing. Like get_configured_token_limits, this never triggers pattern + matching or deep copies, so it is safe to call per listed model on the + /v1/models hot path. + """ + deployment: Final = self.get_deployment_by_model_group_name(model_group_name=model_name) + if deployment is None: + return None + + display_name: Final = deployment.model_info.get("display_name") + if isinstance(display_name, str) and display_name.strip(): + return display_name + return None + def get_deployment_credentials_with_provider( self, model_id: str, team_id: str | None = None ) -> dict[str, Any] | None: diff --git a/tests/test_litellm/proxy/proxy_server/test_routes_models.py b/tests/test_litellm/proxy/proxy_server/test_routes_models.py index 2b126b1ea95..bc6106a06f8 100644 --- a/tests/test_litellm/proxy/proxy_server/test_routes_models.py +++ b/tests/test_litellm/proxy/proxy_server/test_routes_models.py @@ -45,6 +45,7 @@ def patched_models(monkeypatch): deployment = MagicMock() deployment.litellm_params.model = "gpt-4" router.get_deployment_by_model_group_name = MagicMock(return_value=deployment) + router.get_configured_display_name = MagicMock(return_value=None) monkeypatch.setattr(proxy_server, "llm_router", router) monkeypatch.setattr(proxy_server, "prisma_client", MagicMock()) @@ -187,6 +188,83 @@ def test_anthropic_format_carries_router_configured_token_limits(client, auth_as assert (claude["max_input_tokens"], claude["max_tokens"]) == (500000, 4096) +@pytest.mark.parametrize("path", ["/v1/models", "/models"]) +def test_anthropic_format_uses_configured_display_name(client, auth_as, patched_models, path): + """A deployment's ``model_info.display_name`` becomes the Anthropic-native + ``display_name`` so Claude Code's picker shows a clean name while the id keeps + routing; models without one keep the id fallback, and the OpenAI-shaped + listing carries no display_name either way.""" + + def _configured(model_name): + return "Kimi K3" if model_name == "gpt-4" else None + + patched_models.get_configured_display_name = MagicMock(side_effect=_configured) + + with auth_as(): + anthropic_response = client.get(path, headers={"anthropic-version": "2023-06-01"}) + openai_response = client.get(path) + + assert anthropic_response.status_code == 200 + gpt_4, claude = anthropic_response.json()["data"] + assert (gpt_4["id"], gpt_4["display_name"]) == ("gpt-4", "Kimi K3") + assert (claude["id"], claude["display_name"]) == ("claude-sonnet", "claude-sonnet") + + assert openai_response.status_code == 200 + openai_models = openai_response.json()["data"] + assert [m["id"] for m in openai_models] == ["gpt-4", "claude-sonnet"] + assert all("display_name" not in m for m in openai_models) + + +@pytest.mark.parametrize("params", [{}, {"scope": "expand"}]) +def test_anthropic_display_name_resolved_via_internal_team_key( + client, auth_as, patched_models, monkeypatch, params +): + """For a team-scoped row the configured display name must be looked up by the + internal routing key while the entry itself is keyed by the public name, so + the clean name lands on the id the client actually sees.""" + from litellm.proxy import utils as proxy_utils + from litellm.proxy.auth import model_checks + + internal_name = "model_name_team-1_c0ffee" + + patched_models.get_model_list = MagicMock( + return_value=[ + { + "model_name": internal_name, + "model_info": { + "team_id": "team-1", + "team_public_model_name": "gpt-4-team", + }, + } + ] + ) + patched_models.get_model_names = MagicMock(return_value=[internal_name]) + patched_models.get_configured_display_name = MagicMock( + side_effect=lambda model_name: "Team GPT" if model_name == internal_name else None + ) + + async def _fake_get_available_models_for_user(**kwargs): + return [internal_name] + + monkeypatch.setattr( + proxy_utils, + "get_available_models_for_user", + _fake_get_available_models_for_user, + ) + monkeypatch.setattr( + model_checks, "get_complete_model_list", lambda **kwargs: [internal_name] + ) + + with auth_as(): + response = client.get( + "/v1/models", params=params, headers={"anthropic-version": "2023-06-01"} + ) + + assert response.status_code == 200 + (entry,) = response.json()["data"] + assert (entry["id"], entry["display_name"]) == ("gpt-4-team", "Team GPT") + + @pytest.mark.parametrize("path", ["/v1/models", "/models"]) def test_get_models_invalid_scope_returns_400(client, auth_as, patched_models, path): """Pins: ``GET /v1/models``, ``GET /models`` (error path: invalid scope).""" diff --git a/tests/test_litellm/proxy/proxy_server/test_team_model_name_translation.py b/tests/test_litellm/proxy/proxy_server/test_team_model_name_translation.py index aa35fd64f18..0fb9b1a6d88 100644 --- a/tests/test_litellm/proxy/proxy_server/test_team_model_name_translation.py +++ b/tests/test_litellm/proxy/proxy_server/test_team_model_name_translation.py @@ -19,7 +19,10 @@ from litellm.proxy._types import ( LitellmUserRoles, UserAPIKeyAuth, ) -from litellm.proxy.common_utils.model_listing_utils import TeamModelNameTranslator +from litellm.proxy.common_utils.model_listing_utils import ( + TeamModelNameTranslator, + configured_display_names, +) from litellm.proxy.proxy_server import ( _get_proxy_model_info, _translate_model_name_for_response, @@ -1391,6 +1394,27 @@ def test_resolve_public_name_respects_legacy_flag(): ) +def test_configured_display_names_keyed_by_response_id(): + """The map is keyed by the public response id while the router lookup uses + the internal routing key, and entries without a configured name are omitted.""" + router = MagicMock() + router.get_configured_display_name = MagicMock( + side_effect=lambda model_name: "Team Sonnet" if model_name == "model_name_team-abc-123_4a6b8" else None + ) + + assert configured_display_names( + entries=[ + ("team-claude-sonnet", "model_name_team-abc-123_4a6b8"), + ("gpt-4o", "gpt-4o"), + ], + llm_router=router, + ) == {"team-claude-sonnet": "Team Sonnet"} + + +def test_configured_display_names_empty_without_router(): + assert configured_display_names(entries=[("gpt-4o", "gpt-4o")], llm_router=None) == {} + + @pytest.mark.asyncio async def test_retrieve_model_by_public_name_returns_200(monkeypatch): """Regression: `GET /v1/models/{public_name}` must NOT 404. The listing diff --git a/tests/test_litellm/test_router.py b/tests/test_litellm/test_router.py index 44c1cdbff06..f4ea9b03a80 100644 --- a/tests/test_litellm/test_router.py +++ b/tests/test_litellm/test_router.py @@ -7271,6 +7271,71 @@ def test_get_configured_token_limits_coerces_numeric_strings(): assert router.get_configured_token_limits("quoted-limits-model") == (32000, 8000) +def test_get_configured_display_name_reads_deployment_model_info(): + router = litellm.Router( + model_list=[ + { + "model_name": "Kimi K3-claude-compatible", + "litellm_params": {"model": "openai/some-unmapped-model"}, + "model_info": {"display_name": "Kimi K3"}, + } + ] + ) + + assert router.get_configured_display_name("Kimi K3-claude-compatible") == "Kimi K3" + + +def test_get_configured_display_name_returns_none_for_unset_or_unknown(): + router = litellm.Router( + model_list=[ + { + "model_name": "no-display-model", + "litellm_params": {"model": "openai/some-unmapped-model"}, + } + ] + ) + + assert router.get_configured_display_name("no-display-model") is None + assert router.get_configured_display_name("not-a-real-model") is None + + +def test_get_configured_display_name_skips_wildcard_pattern_matching(): + router = litellm.Router( + model_list=[ + { + "model_name": "bedrock/*", + "litellm_params": {"model": "bedrock/*"}, + "model_info": {"display_name": "Bedrock"}, + } + ] + ) + + with patch.object( + router.pattern_router, "route", side_effect=AssertionError("pattern route called") + ): + assert ( + router.get_configured_display_name("bedrock/anthropic.claude-3-5-sonnet-20240620-v1:0") + is None + ) + + +def test_get_configured_display_name_treats_malformed_values_as_absent(): + malformed = ["", " ", 12345, ["Kimi K3"], {"name": "Kimi K3"}, True] + router = litellm.Router( + model_list=[ + { + "model_name": f"bad-display-{i}", + "litellm_params": {"model": "openai/some-unmapped-model"}, + "model_info": {"display_name": bad}, + } + for i, bad in enumerate(malformed) + ] + ) + + for i in range(len(malformed)): + assert router.get_configured_display_name(f"bad-display-{i}") is None + + @pytest.mark.asyncio async def test_acreate_batch_disable_fallbacks_surfaces_owning_provider_error(): router = litellm.Router(