renderers 0.1.12.dev10__tar.gz → 0.1.12.dev11__tar.gz

{renderers-0.1.12.dev10 → renderers-0.1.12.dev11}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: renderers
-Version: 0.1.12.dev10
+Version: 0.1.12.dev11
 Summary: Chat template renderers — deterministic message-to-token conversion for LLM training
 License-Expression: Apache-2.0
 License-File: LICENSE

{renderers-0.1.12.dev10 → renderers-0.1.12.dev11}/renderers/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '0.1.12.dev10'
-__version_tuple__ = version_tuple = (0, 1, 12, 'dev10')
+__version__ = version = '0.1.12.dev11'
+__version_tuple__ = version_tuple = (0, 1, 12, 'dev11')
 
 __commit_id__ = commit_id = None

{renderers-0.1.12.dev10 → renderers-0.1.12.dev11}/renderers/client.py

@@ -224,8 +224,8 @@ async def generate(
     mm_placeholders, kwargs_data) before POSTing. The serializer imports
     ``vllm.*`` lazily so text-only consumers never pay for the import.
     With ``process_multimodal=False``, rendering skips image processing and
-    the request carries ``content_parts`` instead; vLLM must return the
-    expanded prompt as ``prompt_token_ids``.
+    the request carries ``content_parts`` with ``return_token_ids=true``;
+    vLLM returns the expanded prompt and multimodal placeholder ranges.
 
     ``max_prompt_len`` controls the pre-flight overflow check. When the
     rendered prompt is strictly longer than the cap, the request is never
@@ -325,6 +325,7 @@ async def generate(
     )
     if content_parts:
         body["content_parts"] = content_parts
+        body["return_token_ids"] = True
     if features is not None:
         body["features"] = features
     if cache_salt is not None:

{renderers-0.1.12.dev10 → renderers-0.1.12.dev11}/tests/test_client.py

@@ -240,6 +240,7 @@ def test_generate_process_multimodal_false_sends_content_parts():
 
     body = client.calls[0]["body"]
     assert body["content_parts"] == [{"type": "image_url", "url": image_url}]
+    assert body["return_token_ids"] is True
     assert "features" not in body
     assert result["renderer_prompt_ids"] == [1, 2, 3]
     assert result["prompt_ids"] == [1, 2, 2, 3]