fix: align SDK with server #218 (max_completion_tokens, models.retrieve, streaming fence strip)

Author: Abhinavexist

README.md

@@ -124,8 +124,8 @@ URLs and base64 work; raw `bytes` do **not** (must be base64-encoded — this SD
 ## Good to know
 
 - Interfaze implements `chat.completions` and `models`; other OpenAI endpoints are not exposed.
-- `temperature` ≤ 1, `max_tokens` ≤ 32000, `top_p` ≤ 1 (above → 400). Use `max_tokens` (not
-  `max_completion_tokens`) to bound output.
+- `temperature` ≤ 1, `max_tokens` ≤ 32000, `top_p` ≤ 1 (above → 400). Both `max_tokens` and
+  `max_completion_tokens` bound output (`max_tokens` wins if both are set).
 - `n`, `seed`, `stop`, penalties, `logprobs`, `tool_choice`, `top_k` are ignored by Interfaze.
 - The underlying OpenAI client is available at `interfaze.openai`.
 

src/interfaze/_chat.py

@@ -1,6 +1,5 @@
 from __future__ import annotations
 
-import re
 from typing import Any, Dict, Iterable, List, Literal, Optional, Union, cast, overload
 
 from openai import AsyncOpenAI, AsyncStream, OpenAI, Stream
@@ -10,19 +9,9 @@
 from ._errors import InterfazeError
 from ._guard import guard_tag
 from ._schema import empty_task_schema
-from ._stream import AsyncInterfazeStream, InterfazeStream
+from ._stream import AsyncInterfazeStream, InterfazeStream, strip_json_fence
 from ._types import GuardCode, InterfazeChatCompletion, TaskName
 
-_FENCE = re.compile(r"^```(?:json)?\s*|\s*```$", re.IGNORECASE)
-
-
-def strip_json_fence(content: str) -> str:
-    """Interfaze wraps ``json_object`` content in a ```json fence; unwrap it."""
-    t = content.strip()
-    if not t.startswith("```"):
-        return content
-    return _FENCE.sub("", t).strip()
-
 
 def _is_non_empty_schema(rf: Any) -> bool:
     schema = (rf or {}).get("json_schema", {}).get("schema", {}) if isinstance(rf, dict) else {}
@@ -154,8 +143,8 @@ def stream(
         response_format: Optional[Dict[str, Any]] = None,
         **kwargs: Any,
     ) -> InterfazeStream:
-        kw, _ = self._kwargs(messages, model, task, guard, response_format, kwargs)
-        return InterfazeStream(self._client, kw)
+        kw, strip = self._kwargs(messages, model, task, guard, response_format, kwargs)
+        return InterfazeStream(self._client, kw, strip)
 
 
 class AsyncCompletions(_CompletionsBase):
@@ -219,8 +208,8 @@ def stream(
         response_format: Optional[Dict[str, Any]] = None,
         **kwargs: Any,
     ) -> AsyncInterfazeStream:
-        kw, _ = self._kwargs(messages, model, task, guard, response_format, kwargs)
-        return AsyncInterfazeStream(self._client, kw)
+        kw, strip = self._kwargs(messages, model, task, guard, response_format, kwargs)
+        return AsyncInterfazeStream(self._client, kw, strip)
 
 
 class Chat:

src/interfaze/_stream.py

@@ -30,6 +30,17 @@ def strip_side_channels(content: str) -> Tuple[str, Optional[str], Optional[List
     return text.strip(), ("\n".join(thinks) if thinks else None), (pre or None)
 
 
+_FENCE = re.compile(r"^```(?:json)?\s*|\s*```$", re.IGNORECASE)
+
+
+def strip_json_fence(content: str) -> str:
+    """Interfaze wraps ``json_object`` content in a ```json fence; unwrap it."""
+    t = content.strip()
+    if not t.startswith("```"):
+        return content
+    return _FENCE.sub("", t).strip()
+
+
 _SIDE_OPEN = ("<think>", "<precontext>")
 _SIDE_CLOSE = {"<think>": "</think>", "<precontext>": "</precontext>"}
 
@@ -127,8 +138,10 @@ def accumulate(self, chunk: ChatCompletionChunk) -> None:
             if tc.function and tc.function.arguments:
                 acc["arguments"] += tc.function.arguments
 
-    def build(self) -> InterfazeChatCompletion:
+    def build(self, strip_fence: bool = False) -> InterfazeChatCompletion:
         text, reasoning, precontext = strip_side_channels(self.content)
+        if strip_fence:
+            text = strip_json_fence(text)
         tool_calls = [
             {"id": t["id"], "type": "function", "function": {"name": t["name"], "arguments": t["arguments"]}}
             for t in self.tool_calls.values()
@@ -156,9 +169,10 @@ def build(self) -> InterfazeChatCompletion:
 class InterfazeStream:
     """Sync streaming helper — iterate chunks, then ``get_final_completion()``."""
 
-    def __init__(self, client: OpenAI, kwargs: Dict[str, Any]) -> None:
+    def __init__(self, client: OpenAI, kwargs: Dict[str, Any], strip_fence: bool = False) -> None:
         self._client = client
         self._kwargs = kwargs
+        self._strip_fence = strip_fence
         self._state = _State()
         self._started = False
         self._done = False
@@ -215,15 +229,16 @@ def get_final_completion(self) -> InterfazeChatCompletion:
             raise InterfazeError(
                 "Call get_final_completion() after fully iterating, or instead of iterating."
             )
-        return self._state.build()
+        return self._state.build(self._strip_fence)
 
 
 class AsyncInterfazeStream:
     """Async streaming helper — ``async for`` chunks, then ``await get_final_completion()``."""
 
-    def __init__(self, client: AsyncOpenAI, kwargs: Dict[str, Any]) -> None:
+    def __init__(self, client: AsyncOpenAI, kwargs: Dict[str, Any], strip_fence: bool = False) -> None:
         self._client = client
         self._kwargs = kwargs
+        self._strip_fence = strip_fence
         self._state = _State()
         self._started = False
         self._done = False
@@ -279,4 +294,4 @@ async def get_final_completion(self) -> InterfazeChatCompletion:
             raise InterfazeError(
                 "Call get_final_completion() after fully iterating, or instead of iterating."
             )
-        return self._state.build()
+        return self._state.build(self._strip_fence)

tests/test_models.py

@@ -0,0 +1,41 @@
+from __future__ import annotations
+
+import httpx
+import pytest
+import respx
+from openai import NotFoundError
+
+from interfaze import Interfaze
+
+MODELS_URL = "https://api.interfaze.ai/v1/models"
+MODEL = {"id": "interfaze-beta", "object": "model", "owned_by": "interfaze", "name": "Interfaze Beta"}
+MODELS_LIST = {"object": "list", "data": [MODEL]}
+NOT_FOUND = {
+    "error": {
+        "message": "The model 'nope' does not exist",
+        "type": "invalid_request_error",
+        "code": "model_not_found",
+    }
+}
+
+
+@respx.mock
+def test_models_list():
+    respx.get(MODELS_URL).mock(return_value=httpx.Response(200, json=MODELS_LIST))
+    page = Interfaze(api_key="t").models.list()
+    assert [m.id for m in page] == ["interfaze-beta"]
+    assert page.data[0].owned_by == "interfaze"
+
+
+@respx.mock
+def test_models_retrieve():
+    respx.get(f"{MODELS_URL}/interfaze-beta").mock(return_value=httpx.Response(200, json=MODEL))
+    m = Interfaze(api_key="t").models.retrieve("interfaze-beta")
+    assert m.id == "interfaze-beta" and m.owned_by == "interfaze"
+
+
+@respx.mock
+def test_models_retrieve_not_found():
+    respx.get(f"{MODELS_URL}/nope").mock(return_value=httpx.Response(404, json=NOT_FOUND))
+    with pytest.raises(NotFoundError):
+        Interfaze(api_key="t").models.retrieve("nope")

tests/test_stream.py

@@ -1,12 +1,20 @@
 from __future__ import annotations
 
 import asyncio
+import json
 
 import respx
 from conftest import STREAM_CHUNKS, STREAM_THINK, _chunk, mock_sse
 
 from interfaze import AsyncInterfaze, Interfaze
 
+FENCED_JSON = [
+    _chunk({"content": "```json\n"}),
+    _chunk({"content": '{"city": "Tokyo"}'}),
+    _chunk({"content": "\n```"}),
+    _chunk({}, finish_reason="stop"),
+]
+
 
 @respx.mock
 def test_stream_iterates_and_accumulates():
@@ -104,3 +112,36 @@ async def go():
         return "".join([t async for t in stream.text_deltas()])
 
     assert asyncio.run(go()) == "Total is $12.34"
+
+
+@respx.mock
+def test_stream_json_object_strips_fence():
+    mock_sse(FENCED_JSON)
+    stream = Interfaze(api_key="t").chat.completions.stream(
+        messages=[{"role": "user", "content": "x"}], response_format={"type": "json_object"}
+    )
+    content = stream.get_final_completion().choices[0].message.content or ""
+    assert not content.lstrip().startswith("```")
+    assert json.loads(content)["city"] == "Tokyo"
+
+
+@respx.mock
+def test_stream_without_json_object_keeps_fence():
+    mock_sse(FENCED_JSON)
+    stream = Interfaze(api_key="t").chat.completions.stream(messages=[{"role": "user", "content": "x"}])
+    content = stream.get_final_completion().choices[0].message.content or ""
+    assert content.lstrip().startswith("```")
+
+
+@respx.mock
+def test_async_stream_json_object_strips_fence():
+    mock_sse(FENCED_JSON)
+
+    async def go():
+        stream = AsyncInterfaze(api_key="t").chat.completions.stream(
+            messages=[{"role": "user", "content": "x"}], response_format={"type": "json_object"}
+        )
+        return await stream.get_final_completion()
+
+    content = asyncio.run(go()).choices[0].message.content or ""
+    assert not content.lstrip().startswith("```")