verifiers 0.3.2.dev113__tar.gz → 0.3.2.dev115__tar.gz

{verifiers-0.3.2.dev113 → verifiers-0.3.2.dev115}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifiers
-Version: 0.3.2.dev113
+Version: 0.3.2.dev115
 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-0.3.2.dev113 → verifiers-0.3.2.dev115}/verifiers/v1/harnesses/utils/core.py

@@ -310,8 +310,12 @@ async def run_chat_loop(
         try:
             completion, messages = await compactor.complete(messages)
         except APIStatusError as error:
-            # Null cannot compact, so context exhaustion ends it with the transcript so far.
-            if args.bash or not is_context_overflow(error):
+            # Without compaction (off, or on with no discoverable window to compact
+            # against), context exhaustion is a budget limit, not a crash: end the run
+            # with the transcript so far. When the compactor could act, it already
+            # tried, so an overflow reaching here is a real failure.
+            compacting = compactor.enabled and compactor.threshold is not None
+            if compacting or not is_context_overflow(error):
                 raise
             return
         message = completion.choices[0].message

{verifiers-0.3.2.dev113 → verifiers-0.3.2.dev115}/verifiers/v1/trace.py

@@ -513,8 +513,10 @@ class Trace(BaseModel, Generic[DataT, StateT, AgentConfigT]):
 
     @property
     def num_total_tokens(self) -> int:
-        """Final sequence lengths (last prompt + completion) summed across branches."""
-        return sum(branch.num_total_tokens for branch in self.branches)
+        """New input plus generated tokens, counted once per call across branches.
+        Input is a lower bound when the engine drops tokens between calls."""
+        usage = self.usage
+        return self.num_input_tokens + (usage.completion_tokens if usage else 0)
 
     @property
     def usage(self) -> Usage | None: