Skip to content

Use Claude Code, Codex, VSCode, Pi, and OpenCode (and 6 other harnesses) for free (1.3B+ free tokens) from your terminal, app, IDE, or phone, and now from the browser with native browser sessions (multi-harness + multi-model) like OpenClaw (voice supported + ToS friendly)

Python

fix: detect NIM 500 context overflow as context_window_exceeded (#1280)

Joyeb Kothiya51cc4be3e47a
diff --git a/pyproject.toml b/pyproject.toml
index 0d18d2da..c0f4bcb7 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
[project]
name = "free-claude-code"
-version = "4.14.0"
+version = "4.14.1"
description = "Local proxy connecting coding agents to OpenAI-compatible AI providers"
readme = "README.md"
requires-python = ">=3.14.0"
diff --git a/src/free_claude_code/providers/nvidia_nim/client.py b/src/free_claude_code/providers/nvidia_nim/client.py
index 9cebd7a4..a658540c 100644
--- a/src/free_claude_code/providers/nvidia_nim/client.py
+++ b/src/free_claude_code/providers/nvidia_nim/client.py
@@ -135,15 +135,18 @@ class NvidiaNimProvider(OpenAIChatProvider):
return None
def _provider_failure_override(self, error: Exception) -> ExecutionFailure | None:
- """Classify NVIDIA-specific 400 responses by their actual semantics."""
- if not isinstance(error, openai.BadRequestError):
+ """Classify NVIDIA-specific 400/500 responses by their actual semantics."""
+ if not isinstance(error, openai.BadRequestError | openai.InternalServerError):
return None
- if getattr(error, "status_code", None) != 400:
+ status = getattr(error, "status_code", None)
+ if status not in (400, 500):
return None
bodies = _nim_error_bodies(error)
if any(_is_context_window_exhaustion(body) for body in bodies):
return context_window_exceeded_provider_failure()
- if any(_is_degraded_function(body) for body in bodies):
+ if isinstance(error, openai.BadRequestError) and any(
+ _is_degraded_function(body) for body in bodies
+ ):
return overloaded_provider_failure()
return None
diff --git a/tests/providers/test_nvidia_nim_degraded_retry.py b/tests/providers/test_nvidia_nim_degraded_retry.py
index 067334e8..1763608b 100644
--- a/tests/providers/test_nvidia_nim_degraded_retry.py
+++ b/tests/providers/test_nvidia_nim_degraded_retry.py
@@ -77,18 +77,23 @@ def _context_window_error(
),
param: str = "max_tokens",
nested: bool = False,
-) -> openai.BadRequestError:
+ status_code: int = 400,
+) -> openai.BadRequestError | openai.InternalServerError:
request = httpx.Request(
"POST", "https://integrate.api.nvidia.com/v1/chat/completions"
)
- response = httpx.Response(400, request=request)
+ response = httpx.Response(status_code, request=request)
error_body: dict[str, object] = {
"message": message,
"type": "BadRequestError",
"param": param,
- "code": 400,
+ "code": status_code,
}
body = {"error": error_body} if nested else error_body
+ if status_code == 500:
+ return openai.InternalServerError(
+ "Internal Server Error", response=response, body=body
+ )
return openai.BadRequestError("Bad Request", response=response, body=body)
@@ -232,6 +237,42 @@ async def test_negative_derived_max_tokens_is_context_window_failure(
assert "prompt is too long" not in failure.message
[email protected]("nested", [False, True])
+async def test_negative_derived_max_tokens_is_context_window_failure_on_500(
+ nested: bool,
+) -> None:
+ provider = _nim(_admission())
+
+ with (
+ patch.object(
+ provider._client.chat.completions,
+ "create",
+ new_callable=AsyncMock,
+ side_effect=_context_window_error(nested=nested, status_code=500),
+ ) as create,
+ pytest.raises(ExecutionFailure) as exc_info,
+ ):
+ [
+ event
+ async for event in provider.stream_response(
+ make_messages_request(), request_id="req_context_500"
+ )
+ ]
+
+ assert create.await_count == 1
+ failure = exc_info.value
+ assert failure.kind is FailureKind.CONTEXT_WINDOW_EXCEEDED
+ assert failure.status_code == 400
+ assert failure.retryable is False
+ assert "Mapped message: Provider input exceeds the model context window." in (
+ failure.message
+ )
+ assert "max_tokens must be at least 1, got -853" in failure.message
+ assert "Request ID: req_context_500" in failure.message
+ assert "prompt is too long" not in failure.message
+
+
@pytest.mark.parametrize(
("message", "param"),
[
diff --git a/uv.lock b/uv.lock
index d91ad02e..29e680f3 100644
--- a/uv.lock
+++ b/uv.lock
@@ -620,7 +620,7 @@ wheels = [
[[package]]
name = "free-claude-code"
-version = "4.14.0"
+version = "4.14.1"
source = { editable = "." }
dependencies = [
{ name = "aiohttp" },