verifiers 0.2.2.dev30__tar.gz → 0.2.2.dev32__tar.gz

{verifiers-0.2.2.dev30 → verifiers-0.2.2.dev32}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.2.2.dev30
+Version: 0.2.2.dev32
 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-0.2.2.dev30 → verifiers-0.2.2.dev32}/verifiers/v1/dialects/responses.py

@@ -1,9 +1,9 @@
 """The OpenAI Responses dialect (codex and friends).
 
 Request parsing walks the `input` items, folding each run of assistant-side items (reasoning /
-assistant message / function_call) into one typed assistant message; response parsing reads the
-`output` items. Relay-only: the eval client forwards the program's bytes to a `/responses`
-endpoint and this dialect parses a copy for the trace. Server-side statefulness
+assistant message / function or custom tool call) into one typed assistant message; response
+parsing reads the `output` items. Relay-only: the eval client forwards the program's bytes to a
+`/responses` endpoint and this dialect parses a copy for the trace. Server-side statefulness
 (`previous_response_id`) is not emulated — the endpoint owns it.
 """
 
@@ -97,8 +97,7 @@ def parse_content(content) -> str | list[ContentPart]:
 
 
 def fold_assistant(items: list[dict]) -> AssistantMessage:
-    """One run of assistant-side items (reasoning / message / function_call) -> one typed
-    assistant message."""
+    """One run of assistant-side items -> one typed assistant message."""
     content = ""
     reasoning: list[str] = []
     calls: list[ToolCall] = []
@@ -106,12 +105,12 @@ def fold_assistant(items: list[dict]) -> AssistantMessage:
         if item.get("type") == "reasoning":
             reasoning += [s.get("text", "") for s in item.get("summary") or []]
             reasoning += [c.get("text", "") for c in item.get("content") or []]
-        elif item.get("type") == "function_call":
+        elif item.get("type") in ("function_call", "custom_tool_call"):
             calls.append(
                 ToolCall(
                     id=item.get("call_id", ""),
                     name=item.get("name", ""),
-                    arguments=item.get("arguments", ""),
+                    arguments=item.get("arguments", item.get("input", "")),
                 )
             )
         else:  # an assistant message item
@@ -151,12 +150,12 @@ def response_from_wire(response: OpenAIResponse) -> Response:
         elif kind == "reasoning":
             reasoning += [s.get("text", "") for s in item.get("summary") or []]
             reasoning += [c.get("text", "") for c in item.get("content") or []]
-        elif kind == "function_call":
+        elif kind in ("function_call", "custom_tool_call"):
             calls.append(
                 ToolCall(
                     id=item.get("call_id", ""),
                     name=item.get("name", ""),
-                    arguments=item.get("arguments", ""),
+                    arguments=item.get("arguments", item.get("input", "")),
                 )
             )
     tool_calls = calls or None
@@ -279,7 +278,10 @@ class ResponsesDialect(Dialect[dict, OpenAIResponse]):
                 run = []
             if assistant:
                 run.append(item)
-            elif item.get("type") == "function_call_output":
+            elif item.get("type") in (
+                "function_call_output",
+                "custom_tool_call_output",
+            ):
                 output = item.get("output")
                 content = (
                     parse_content(output)

{verifiers-0.2.2.dev30 → verifiers-0.2.2.dev32}/verifiers/v1/graph.py

@@ -212,12 +212,17 @@ def _canonical_tool_arguments(arguments: str) -> str:
         return arguments
 
 
+# Provider-specific fields not represented by typed messages but required on replay.
+_PROVIDER_STATE_FIELDS = frozenset({"encrypted_content", "signature", "data", "phase"})
+
+
 def message_hash(message: Message) -> str:
     """Stable content hash on the fields that round-trip through a prompt — role, content
     (None and "" equal), assistant reasoning content when present, assistant tool calls,
-    tool call id. Two messages hash equal iff they're the same conversational message, so a
-    re-stated prefix message dedups to one node. The dedup key for sharing a prefix across
-    turns/branches; salt-free so it is identical across processes and after deserialization."""
+    opaque continuation state, tool call id. Two messages hash equal iff they're the same
+    conversational message, so a re-stated prefix message dedups to one node. The dedup key
+    for sharing a prefix across turns/branches; salt-free so it is identical across processes
+    and after deserialization."""
     digest = hashlib.blake2b(digest_size=16)
 
     def add(value: str) -> None:
@@ -241,10 +246,37 @@ def message_hash(message: Message) -> str:
         if message.reasoning_content is not None:
             add("reasoning_content")
             add(message.reasoning_content)
-        if message.provider_state:
-            # Signed/encrypted continuation state distinguishes otherwise equal turns.
+        for item in message.provider_state or []:
+            kind = item.get("type") or (
+                "message" if item.get("role") == "assistant" else ""
+            )
+            hashed_state = {
+                key: item[key]
+                for key in _PROVIDER_STATE_FIELDS
+                if item.get(key) is not None
+            }
+            if kind == "message" and isinstance(item.get("content"), list):
+                # Keep content parts the typed message does not expose, such as refusals.
+                unparsed_content = [
+                    part
+                    for part in item.get("content") or []
+                    if part.get("type") not in ("input_text", "output_text")
+                ]
+                if unparsed_content:
+                    hashed_state["content"] = unparsed_content
+            represented = kind in ("message", "reasoning") or (
+                kind in ("function_call", "custom_tool_call")
+                and any(
+                    call.id == item.get("call_id") for call in message.tool_calls or []
+                )
+            )
+            if represented and not hashed_state:
+                continue
+            # Unknown provider items still distinguish built-in calls and actions.
+            state = hashed_state if represented else item
             add("provider_state")
-            add(json.dumps(message.provider_state, sort_keys=True))
+            add(kind)
+            add(json.dumps(state, sort_keys=True))
         for tc in message.tool_calls or []:
             add("tool_call")
             add(tc.id)