verifiers 0.3.2.dev130__py3-none-any.whl → 0.3.2.dev132__py3-none-any.whl

verifiers/v1/harnesses/utils/core.py

@@ -240,7 +240,9 @@ async def chat(
     *,
     tool_choice: str | None = None,
 ):
-    kwargs = {"model": model, "messages": messages, "tools": tools or None}
+    kwargs = {"model": model, "messages": messages}
+    if tools:
+        kwargs["tools"] = tools
     if tools and tool_choice is not None:
         kwargs["tool_choice"] = tool_choice
     raw_stream = await client.chat.completions.create(

verifiers/v1/tasksets/harbor/env.py

@@ -7,7 +7,9 @@ box it worked in. A separate-verifier task is graded by `finalize` instead: the
 solver's declared artifacts travel (collected after its task `finalize` while its
 box is alive), a fresh box is provisioned from the task's verifier declaration,
 `tests/` is staged there, and the verifier's rewards land on the solver's trace.
-No second agent is involved — the verifier is the task's own `tests/test.sh`.
+No second agent is involved — the verifier is the task's own `tests/test.sh`,
+staged by the task's own class, so taskset subclasses customize grading through
+their task hooks rather than a custom env.
 """
 
 import verifiers.v1 as vf
@@ -56,14 +58,16 @@ class HarborEnv(IsolatedVerifierEnv, vf.Env[HarborEnvConfig]):
         Provision a fresh box from the task's verifier declaration, restore the
         solver's collected artifacts, stage `tests/`, run the verifier, and record
         its rewards (and any extra reward.json keys as metrics) on the solver's
-        trace. Setup, restoration, staging, and scoring failures retry per
-        `verifier.retries`; the last one fails the episode."""
+        trace. The grader is the task's own class, so a `HarborTask` subclass's
+        `setup` and `stage_verifier` run in the verifier box too. Setup,
+        restoration, staging, and scoring failures retry per `verifier.retries`;
+        the last one fails the episode."""
         if not isinstance(task, HarborTask) or task.data.verifier is None:
             return
         solution = episode.traces[0]
         if not solution.ok:
             return
-        grader = HarborTask(verifier_box_data(task.data))
+        grader = type(task)(verifier_box_data(task.data), task.config)
         scores, solution = await self.grade(
             self.verifier_config(task),
             grader,

{verifiers-0.3.2.dev130.dist-info → verifiers-0.3.2.dev132.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifiers
-Version: 0.3.2.dev130
+Version: 0.3.2.dev132
 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.dev130.dist-info → verifiers-0.3.2.dev132.dist-info}/RECORD

@@ -118,7 +118,7 @@ verifiers/v1/harnesses/terminus_2/harness.py,sha256=cKQJluW01mkxEccTLFwSWsVMavyi
 verifiers/v1/harnesses/terminus_2/program.py,sha256=X8MZdyItU3gbXGXnGOlJ9LWcUTnqIDaxlhkZvoRiqKY,2703
 verifiers/v1/harnesses/utils/__init__.py,sha256=eTS-Ft2HwoLiYn5u1t_KueLmaZB_W12cdYmcVsE-G5A,364
 verifiers/v1/harnesses/utils/compaction.py,sha256=DeY6CYLEsv7-vJm26sT2mRaPHu3W2ywiZhvFr_pi6Hg,10495
-verifiers/v1/harnesses/utils/core.py,sha256=r7kYmIr27jGXIgGHzYwaqmEkTZc7r310mxRH2tc_slc,18498
+verifiers/v1/harnesses/utils/core.py,sha256=emAeNqtha__3cR0u7pvoQGnl2sE57wadE0wlKGtO1Yk,18520
 verifiers/v1/harnesses/utils/install.py,sha256=BDkkUVZ1brSzCrbQOvMO5mVIE2DsV2pE7rTC5yROYUw,2992
 verifiers/v1/harnesses/utils/launch.py,sha256=RqPuUY-kT2XVuLSGtXW7kRQEaP-o2KzkiJ7vcoo2neM,2883
 verifiers/v1/harnesses/utils/mcp.py,sha256=oNhKfLjAW-wScJcDIsW_HfuvKY-JkaGBBZrRLqgRK1g,7355
@@ -158,7 +158,7 @@ verifiers/v1/serve/server.py,sha256=i6XrY_PMAapkrP54tYX8-K6iGmOmrR7BbQ-3Qhn3w4s,
 verifiers/v1/serve/types.py,sha256=Qw5IjiJhsZz4giSXUBi9dfInBnQZ5_xiUmgH2qp8jPI,1741
 verifiers/v1/tasksets/__init__.py,sha256=2ijE3qIlLoUA02g6iysOwWQylyBBQL1GhSpjv8_bp0w,521
 verifiers/v1/tasksets/harbor/__init__.py,sha256=JxThzHFYEkuSfQbVf3pEnjordgdfe4MU-QOsSSbyovU,326
-verifiers/v1/tasksets/harbor/env.py,sha256=ATMkfwmrE_2jpOOffbjLQ91feG7TOtefBGlpwWbRq8c,3604
+verifiers/v1/tasksets/harbor/env.py,sha256=0Q23PtPuH72q2QmG-iSsV4r7utIXstQgAaf32IA_FYo,3879
 verifiers/v1/tasksets/harbor/taskset.py,sha256=4TwjiH9QhZ05rFHiDFGiUFNFMfQfBrBhdse7wrPOH1w,34204
 verifiers/v1/tasksets/harbor/toolset.py,sha256=_7HXu-uuvqzhmJ1EX6c3KWLg0g0eRrLjCq1kBsZhIwo,3159
 verifiers/v1/tasksets/nemo_gym/__init__.py,sha256=654gAvG3Wd_Z_nh6mypeKn4s_IIQ0Ge6vcZvxTHC2_E,306
@@ -190,8 +190,8 @@ verifiers/v1/utils/retries.py,sha256=Y2ZgrAjn-qNkRKZP_RVNL_7EK0iaRcZeCa404lpGYi4
 verifiers/v1/utils/scope.py,sha256=bzUEiWDOVdbj7dXOwGOlzsQ_-Xu0j9NR5TalQB-8TS8,838
 verifiers/v1/utils/score.py,sha256=493yJVMw8teCu9JxapxMFPFzI0hNUqdo0Y2nGW4kckk,6200
 verifiers/v1/utils/version.py,sha256=-obEo_-l9-D8FLef4hYxncOe-uJpxrM1g2Hig_37Sgs,1607
-verifiers-0.3.2.dev130.dist-info/METADATA,sha256=ObL6yNfqVUgdlvm3Q_7x_Tw6032FzVhZsyJIc2fTbDA,4161
-verifiers-0.3.2.dev130.dist-info/WHEEL,sha256=zOwg4jB6zX2kU910N-cMawjivD6tO8NEWvE12je1bVk,87
-verifiers-0.3.2.dev130.dist-info/entry_points.txt,sha256=iugElcdWPKbQM7uFF0lZ8iUpHsNr17-BwEAAjJWxV3U,259
-verifiers-0.3.2.dev130.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
-verifiers-0.3.2.dev130.dist-info/RECORD,,
+verifiers-0.3.2.dev132.dist-info/METADATA,sha256=NMlzb9XAkDB3ZiYpzuOjF3kv84Dn9HmXn37XCq0XyF8,4161
+verifiers-0.3.2.dev132.dist-info/WHEEL,sha256=zOwg4jB6zX2kU910N-cMawjivD6tO8NEWvE12je1bVk,87
+verifiers-0.3.2.dev132.dist-info/entry_points.txt,sha256=iugElcdWPKbQM7uFF0lZ8iUpHsNr17-BwEAAjJWxV3U,259
+verifiers-0.3.2.dev132.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
+verifiers-0.3.2.dev132.dist-info/RECORD,,