verifiers 0.4.1.dev30__tar.gz → 0.4.1.dev31__tar.gz

{verifiers-0.4.1.dev30 → verifiers-0.4.1.dev31}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifiers
-Version: 0.4.1.dev30
+Version: 0.4.1.dev31
 Summary: Verifiers: Environments for LLM Reinforcement Learning
 Project-URL: Homepage, https://github.com/primeintellect-ai/verifiers
 Project-URL: Documentation, https://github.com/primeintellect-ai/verifiers

{verifiers-0.4.1.dev30 → verifiers-0.4.1.dev31}/verifiers/v1/dialects/anthropic.py

@@ -24,6 +24,7 @@ from verifiers.v1.dialects.base import (
     parse_sse_event,
     provider_domains,
 )
+from verifiers.v1.errors import ProviderError
 from verifiers.v1.types import (
     AssistantMessage,
     ContentPart,
@@ -51,6 +52,18 @@ STOP_REASONS: dict[str, FinishReason] = {
     "tool_use": "tool_calls",
     "stop_sequence": "stop",
 }
+ERROR_TYPES = {
+    400: "invalid_request_error",
+    401: "authentication_error",
+    403: "permission_error",
+    404: "not_found_error",
+    413: "request_too_large",
+    429: "rate_limit_error",
+    502: "api_error",
+    529: "overloaded_error",
+}
+ERROR_STATUSES = {kind: status for status, kind in ERROR_TYPES.items()}
+
 # Claude may reorder mixed thinking block types between a response and its replay.
 # Native tool events share the final rank, preserving their relative order.
 THINKING_ORDER = {"redacted_thinking": 0, "thinking": 1}
@@ -356,6 +369,11 @@ class AnthropicStreamParser(StreamParser):
         if event is None:
             return
         kind = event.get("type")
+        if kind == "error":
+            # HTTP 200 streams can fail in-band; preserve the provider's cause and IDs.
+            detail = event.get("error") or {}
+            status = ERROR_STATUSES.get(detail.get("type"), 502)
+            raise ProviderError(json.dumps(event), status_code=status)
         if kind == "message_start":
             self.message = event.get("message") or {}
         elif kind == "content_block_start":
@@ -516,7 +534,14 @@ class AnthropicDialect(Dialect[AnthropicMessage]):
         # Anthropic's own keepalive; its SDKs skip it anywhere in the stream.
         return b'event: ping\ndata: {"type": "ping"}\n\n'
 
-    def stream_error(self, error: dict) -> bytes:
+    def stream_error(self, error: dict, *, status_code: int | None = None) -> bytes:
+        # Once HTTP 200 is committed, clients classify retries from the SSE error type.
+        if status_code is not None:
+            kind = ERROR_TYPES.get(
+                status_code,
+                "api_error" if status_code >= 500 else "invalid_request_error",
+            )
+            error = {**error, "error": {**error.get("error", {}), "type": kind}}
         # The Anthropic SDKs raise only on a named `error` event.
         return b"event: error\ndata: " + json.dumps(error).encode() + b"\n\n"
 

{verifiers-0.4.1.dev30 → verifiers-0.4.1.dev31}/verifiers/v1/dialects/base.py

@@ -318,7 +318,7 @@ class Dialect(ABC, Generic[RespT]):
         # Don't terminate an empty event; some SSE clients try to JSON-decode it.
         return b": keepalive\n"
 
-    def stream_error(self, error: dict) -> bytes:
+    def stream_error(self, error: dict, *, status_code: int | None = None) -> bytes:
         """An `error_body` as an SSE event, for a failure after the stream is committed.
         OpenAI SDKs raise on any event carrying `error`."""
         return b"data: " + json.dumps(error).encode() + b"\n\n"

{verifiers-0.4.1.dev30 → verifiers-0.4.1.dev31}/verifiers/v1/dialects/responses.py

@@ -687,7 +687,7 @@ class ResponsesDialect(Dialect[OpenAIResponse]):
         payload = {"type": kind, "sequence_number": 0, "response": _KEEPALIVE_RESPONSE}
         return f"event: {kind}\ndata: {json.dumps(payload)}\n\n".encode()
 
-    def stream_error(self, error: dict) -> bytes:
+    def stream_error(self, error: dict, *, status_code: int | None = None) -> bytes:
         # `response.failed` is what Responses clients (codex) act on; the `error` key is what
         # the OpenAI SDK raises on.
         message = error["error"]["message"]

{verifiers-0.4.1.dev30 → verifiers-0.4.1.dev31}/verifiers/v1/interception/server.py

@@ -288,7 +288,9 @@ async def _buffered_stream(
                 await stream.write(
                     replay.body
                     if response.status < 400
-                    else dialect.stream_error(from_json(replay.body))
+                    else dialect.stream_error(
+                        from_json(replay.body), status_code=response.status
+                    )
                 )
                 await stream.write_eof()
             except ConnectionResetError: