[Frontend] Add /v1/messages/render endpoint for the Anthropic Messages API (#45803)

Signed-off-by: HyunKyun Moon <mhg5303@gmail.com>
This commit is contained in:
Hyunkyun Moon
2026-08-26 11:33:59 +08:00
committed by GitHub
parent a5e0045433
commit f25c2692fe
5 changed files with 165 additions and 5 deletions
@@ -493,3 +493,115 @@ async def test_chat_completion_render_assistant_tokens_mask_with_generation_tags
assert detok_rest.status_code == 200
assert "Hi!" not in detok_rest.json()["prompt"]
assert "Bye" in detok_rest.json()["prompt"]
@pytest.mark.asyncio
async def test_messages_render_basic(client):
"""Test basic Anthropic Messages render endpoint."""
response = await client.post(
"/v1/messages/render",
json={
"model": MODEL_NAME,
"max_tokens": 16,
"messages": [{"role": "user", "content": "Render this Anthropic message."}],
},
)
assert response.status_code == 200
data = response.json()
# Single GenerateRequest, like chat render.
assert isinstance(data, dict)
assert "token_ids" in data
assert "sampling_params" in data
assert "model" in data
assert data["model"] == MODEL_NAME
token_ids = data["token_ids"]
assert isinstance(token_ids, list)
assert len(token_ids) > 0
assert all(isinstance(tid, int) for tid in token_ids)
assert token_ids[0] == 1 # BOS
@pytest.mark.asyncio
async def test_messages_render_system_and_multi_turn(client):
"""System field + multi-turn messages render to a single prompt."""
response = await client.post(
"/v1/messages/render",
json={
"model": MODEL_NAME,
"max_tokens": 16,
"system": "You are a helpful assistant.",
"messages": [
{"role": "user", "content": "Hello"},
{"role": "assistant", "content": "Hi! How can I help?"},
{"role": "user", "content": "What is 2 + 2?"},
],
},
)
assert response.status_code == 200
data = response.json()
assert isinstance(data, dict)
assert len(data["token_ids"]) > 0
assert data["token_ids"][0] == 1 # BOS
@pytest.mark.asyncio
async def test_messages_render_merges_inline_system(client):
"""Inline system messages merge into the leading system block.
Without a --chat-template arg the /v1/messages server path detects
merge_inline_system=True, so render must produce the same tokens as
the manually pre-merged request.
"""
inline = await client.post(
"/v1/messages/render",
json={
"model": MODEL_NAME,
"max_tokens": 16,
"system": "You are a helpful assistant.",
"messages": [
{"role": "user", "content": "Hello"},
{"role": "assistant", "content": "Hi! How can I help?"},
{"role": "system", "content": "Be brief."},
{"role": "user", "content": "What is 2 + 2?"},
],
},
)
assert inline.status_code == 200
merged = await client.post(
"/v1/messages/render",
json={
"model": MODEL_NAME,
"max_tokens": 16,
"system": "You are a helpful assistant.Be brief.",
"messages": [
{"role": "user", "content": "Hello"},
{"role": "assistant", "content": "Hi! How can I help?"},
{"role": "user", "content": "What is 2 + 2?"},
],
},
)
assert merged.status_code == 200
assert inline.json()["token_ids"] == merged.json()["token_ids"]
@pytest.mark.asyncio
async def test_messages_render_error_invalid_model(client):
"""Messages render with an invalid model returns an error."""
response = await client.post(
"/v1/messages/render",
json={
"model": "invalid-model-name",
"max_tokens": 16,
"messages": [{"role": "user", "content": "Hello"}],
},
)
assert response.status_code == 404
data = response.json()
assert "error" in data