verifiers 0.1.15.dev11__tar.gz → 0.1.15.dev13__tar.gz

{verifiers-0.1.15.dev11 → verifiers-0.1.15.dev13}/.gitignore

@@ -43,8 +43,6 @@ logs/
 .chroma_db/
 
 # development
-packages/harnesses
-packages/tasksets
 scratch/
 .idea/
 .vscode/

{verifiers-0.1.15.dev11 → verifiers-0.1.15.dev13}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.1.15.dev11
+Version: 0.1.15.dev13
 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
@@ -53,8 +53,18 @@ Provides-Extra: browser
 Requires-Dist: aiohttp>=3.9.0; extra == 'browser'
 Requires-Dist: python-dotenv>=1.0.0; extra == 'browser'
 Requires-Dist: stagehand>=3.0.0; extra == 'browser'
+Provides-Extra: harnesses
+Requires-Dist: harnesses>=0.1.1; extra == 'harnesses'
+Provides-Extra: nemogym
+Requires-Dist: harnesses[nemogym]>=0.1.1; extra == 'nemogym'
+Requires-Dist: tasksets[nemogym]>=0.1.1; extra == 'nemogym'
 Provides-Extra: openenv
-Requires-Dist: openenv-core>=0.3.0; extra == 'openenv'
+Requires-Dist: tasksets[openenv]>=0.1.1; extra == 'openenv'
+Provides-Extra: openreward
+Requires-Dist: tasksets[openreward]>=0.1.1; extra == 'openreward'
+Provides-Extra: packages
+Requires-Dist: harnesses>=0.1.1; extra == 'packages'
+Requires-Dist: tasksets[openenv,openreward,ta]>=0.1.1; extra == 'packages'
 Provides-Extra: renderers
 Requires-Dist: renderers>=0.1.8.dev28; extra == 'renderers'
 Provides-Extra: rg
@@ -73,6 +83,8 @@ Requires-Dist: wandb; extra == 'rl'
 Provides-Extra: ta
 Requires-Dist: nltk; extra == 'ta'
 Requires-Dist: textarena; extra == 'ta'
+Provides-Extra: tasksets
+Requires-Dist: tasksets[openenv,openreward,ta]>=0.1.1; extra == 'tasksets'
 Description-Content-Type: text/markdown
 
 <p align="center">
@@ -226,7 +238,7 @@ class MyTasksetConfig(vf.TasksetConfig):
 
 
 class MyTaskset(vf.Taskset[MyTasksetConfig]):
-    def load_tasks(self) -> vf.Tasks:
+    def load_tasks(self, split: vf.TaskSplit = "train") -> vf.Tasks:
         rows = [
             {
                 "prompt": [{"role": "user", "content": "Reverse abc."}],
@@ -251,16 +263,17 @@ def load_environment(config: vf.EnvConfig) -> vf.Env:
 ```
 If no harness is passed, `vf.Env` uses the base endpoint-backed harness. See
 **[BYO Harness](docs/byo-harness.md)** for the advanced v1 taskset/harness API.
-Reusable taskset and harness packages live under `verifiers.v1.packages`. For
-example, Harbor task directories can run through the bundled OpenCode CLI
-harness with:
+Reusable taskset and harness packages live in `tasksets` and `harnesses`.
+Install them with `uv add "verifiers[packages]"`, or with the narrower
+`verifiers[tasksets]` and `verifiers[harnesses]` extras. For example, Harbor
+task directories can run through the bundled OpenCode CLI harness with:
 
 ```python
-from verifiers.v1.packages.harnesses import OpenCode, OpenCodeConfig
-from verifiers.v1.packages.tasksets import HarborTaskset, HarborTasksetConfig
+from harnesses import OpenCode, OpenCodeConfig
+from tasksets import HarborTaskset, HarborTasksetConfig
 
 env = vf.Env(
-    taskset=HarborTaskset(config=HarborTasksetConfig()),
+    taskset=HarborTaskset(config=HarborTasksetConfig(bundle_package=__name__)),
     harness=OpenCode(config=OpenCodeConfig()),
 )
 ```

{verifiers-0.1.15.dev11 → verifiers-0.1.15.dev13}/README.md

@@ -149,7 +149,7 @@ class MyTasksetConfig(vf.TasksetConfig):
 
 
 class MyTaskset(vf.Taskset[MyTasksetConfig]):
-    def load_tasks(self) -> vf.Tasks:
+    def load_tasks(self, split: vf.TaskSplit = "train") -> vf.Tasks:
         rows = [
             {
                 "prompt": [{"role": "user", "content": "Reverse abc."}],
@@ -174,16 +174,17 @@ def load_environment(config: vf.EnvConfig) -> vf.Env:
 ```
 If no harness is passed, `vf.Env` uses the base endpoint-backed harness. See
 **[BYO Harness](docs/byo-harness.md)** for the advanced v1 taskset/harness API.
-Reusable taskset and harness packages live under `verifiers.v1.packages`. For
-example, Harbor task directories can run through the bundled OpenCode CLI
-harness with:
+Reusable taskset and harness packages live in `tasksets` and `harnesses`.
+Install them with `uv add "verifiers[packages]"`, or with the narrower
+`verifiers[tasksets]` and `verifiers[harnesses]` extras. For example, Harbor
+task directories can run through the bundled OpenCode CLI harness with:
 
 ```python
-from verifiers.v1.packages.harnesses import OpenCode, OpenCodeConfig
-from verifiers.v1.packages.tasksets import HarborTaskset, HarborTasksetConfig
+from harnesses import OpenCode, OpenCodeConfig
+from tasksets import HarborTaskset, HarborTasksetConfig
 
 env = vf.Env(
-    taskset=HarborTaskset(config=HarborTasksetConfig()),
+    taskset=HarborTaskset(config=HarborTasksetConfig(bundle_package=__name__)),
     harness=OpenCode(config=OpenCodeConfig()),
 )
 ```

{verifiers-0.1.15.dev11 → verifiers-0.1.15.dev13}/pyproject.toml

@@ -69,12 +69,12 @@ dev = [
     "ipykernel",
     "ipywidgets",
     "reasoning-gym",
-    "textarena",
     "stagehand>=3.0.0",
     "aiohttp>=3.9.0",
     "python-dotenv>=1.0.0",
-    "nltk",
+    "harnesses",
     "renderers>=0.1.8.dev28",
+    "tasksets[openenv,openreward,ta]",
 ]
 policy = [
     "semgrep>=1.150.0",
@@ -85,8 +85,18 @@ rg = [
     "reasoning-gym",
 ]
 ta = [
-    "textarena",
     "nltk",
+    "textarena",
+]
+tasksets = [
+    "tasksets[openenv,openreward,ta]>=0.1.1",
+]
+harnesses = [
+    "harnesses>=0.1.1",
+]
+packages = [
+    "harnesses>=0.1.1",
+    "tasksets[openenv,openreward,ta]>=0.1.1",
 ]
 browser = [
     "stagehand>=3.0.0",
@@ -94,7 +104,14 @@ browser = [
     "python-dotenv>=1.0.0",
 ]
 openenv = [
-    "openenv-core>=0.3.0",
+    "tasksets[openenv]>=0.1.1",
+]
+openreward = [
+    "tasksets[openreward]>=0.1.1",
+]
+nemogym = [
+    "harnesses[nemogym]>=0.1.1",
+    "tasksets[nemogym]>=0.1.1",
 ]
 renderers = [
     "renderers>=0.1.8.dev28",
@@ -121,12 +138,53 @@ conflicts = [
         { extra = "openenv" },
         { group = "policy" },
     ],
+    [
+        { extra = "nemogym" },
+        { extra = "openenv" },
+    ],
+    [
+        { extra = "nemogym" },
+        { group = "dev" },
+    ],
+    [
+        { extra = "nemogym" },
+        { extra = "tasksets" },
+    ],
+    [
+        { extra = "nemogym" },
+        { extra = "packages" },
+    ],
+    [
+        { extra = "ta" },
+        { group = "policy" },
+    ],
+    [
+        { extra = "tasksets" },
+        { group = "policy" },
+    ],
+    [
+        { extra = "packages" },
+        { group = "policy" },
+    ],
+    [
+        { extra = "openreward" },
+        { group = "policy" },
+    ],
+    [
+        { group = "dev" },
+        { group = "policy" },
+    ],
 ]
+
 [[tool.uv.index]]
 name = "pypi"
 url = "https://pypi.org/simple"
 default = true
 
+[tool.uv.sources]
+harnesses = { path = "packages/harnesses", editable = true }
+tasksets = { path = "packages/tasksets", editable = true }
+
 [tool.uv.exclude-newer-package]
 # PrimeIntellect-published on PyPI (trusted publisher)
 prime-tunnel = false
@@ -134,6 +192,8 @@ prime-sandboxes = false
 prime-pydantic-config = false
 renderers = false
 openenv-core = false
+harnesses = false
+tasksets = false
 
 [tool.uv.extra-build-dependencies]
 flash-attn = [{ requirement = "torch", match-runtime = true }]

{verifiers-0.1.15.dev11 → verifiers-0.1.15.dev13}/tests/test_endpoint_registry.py

@@ -58,8 +58,8 @@ def test_load_endpoints_toml_groups_variants_by_endpoint_id(tmp_path: Path):
 
     assert set(endpoints.keys()) == {"gpt-5-mini"}
     assert len(endpoints["gpt-5-mini"]) == 2
-    assert endpoints["gpt-5-mini"][0]["url"] == "https://api.pinference.ai/api/v1"
-    assert endpoints["gpt-5-mini"][1]["url"] == "https://api.openai.com/v1"
+    assert endpoints["gpt-5-mini"][0].base_url == "https://api.pinference.ai/api/v1"
+    assert endpoints["gpt-5-mini"][1].base_url == "https://api.openai.com/v1"
 
 
 def test_load_endpoints_toml_accepts_long_field_names(tmp_path: Path):
@@ -75,8 +75,8 @@ def test_load_endpoints_toml_accepts_long_field_names(tmp_path: Path):
 
     endpoints = load_endpoints(str(registry_path))
 
-    assert endpoints["gpt-5-mini"][0]["url"] == "https://api.pinference.ai/api/v1"
-    assert endpoints["gpt-5-mini"][0]["key"] == "PRIME_API_KEY"
+    assert endpoints["gpt-5-mini"][0].base_url == "https://api.pinference.ai/api/v1"
+    assert endpoints["gpt-5-mini"][0].api_key_var == "PRIME_API_KEY"
 
 
 def test_load_endpoints_toml_accepts_matching_short_and_long_fields(tmp_path: Path):
@@ -94,8 +94,8 @@ def test_load_endpoints_toml_accepts_matching_short_and_long_fields(tmp_path: Pa
 
     endpoints = load_endpoints(str(registry_path))
 
-    assert endpoints["gpt-5-mini"][0]["url"] == "https://api.pinference.ai/api/v1"
-    assert endpoints["gpt-5-mini"][0]["key"] == "PRIME_API_KEY"
+    assert endpoints["gpt-5-mini"][0].base_url == "https://api.pinference.ai/api/v1"
+    assert endpoints["gpt-5-mini"][0].api_key_var == "PRIME_API_KEY"
 
 
 def test_load_endpoints_toml_rejects_conflicting_url_fields(tmp_path: Path):
@@ -165,14 +165,10 @@ def test_load_endpoints_directory_uses_toml_and_warns_on_ignored_python(
 def test_qwen3_vl_endpoint_ids_map_to_vl_models():
     endpoints = load_endpoints("./configs/endpoints.toml")
 
-    assert endpoints["qwen3-vl-30b-i"][0]["model"] == "qwen/qwen3-vl-30b-a3b-instruct"
-    assert endpoints["qwen3-vl-30b-t"][0]["model"] == "qwen/qwen3-vl-30b-a3b-thinking"
-    assert (
-        endpoints["qwen3-vl-235b-i"][0]["model"] == "qwen/qwen3-vl-235b-a22b-instruct"
-    )
-    assert (
-        endpoints["qwen3-vl-235b-t"][0]["model"] == "qwen/qwen3-vl-235b-a22b-thinking"
-    )
+    assert endpoints["qwen3-vl-30b-i"][0].model == "qwen/qwen3-vl-30b-a3b-instruct"
+    assert endpoints["qwen3-vl-30b-t"][0].model == "qwen/qwen3-vl-30b-a3b-thinking"
+    assert endpoints["qwen3-vl-235b-i"][0].model == "qwen/qwen3-vl-235b-a22b-instruct"
+    assert endpoints["qwen3-vl-235b-t"][0].model == "qwen/qwen3-vl-235b-a22b-thinking"
 
 
 def test_load_endpoints_toml_accepts_type_shorthand(tmp_path: Path):
@@ -189,7 +185,7 @@ def test_load_endpoints_toml_accepts_type_shorthand(tmp_path: Path):
 
     endpoints = load_endpoints(str(registry_path))
 
-    assert endpoints["haiku"][0]["api_client_type"] == "anthropic_messages"
+    assert endpoints["haiku"][0].api_client_type == "anthropic_messages"
 
 
 def test_load_endpoints_toml_accepts_openai_responses_type(tmp_path: Path):
@@ -206,7 +202,7 @@ def test_load_endpoints_toml_accepts_openai_responses_type(tmp_path: Path):
 
     endpoints = load_endpoints(str(registry_path))
 
-    assert endpoints["gpt-responses"][0]["api_client_type"] == "openai_responses"
+    assert endpoints["gpt-responses"][0].api_client_type == "openai_responses"
 
 
 def test_load_endpoints_toml_accepts_headers_table(tmp_path: Path):
@@ -223,7 +219,7 @@ def test_load_endpoints_toml_accepts_headers_table(tmp_path: Path):
 
     endpoints = load_endpoints(str(registry_path))
 
-    assert endpoints["proxy"][0]["extra_headers"] == {"X-Custom": "v1"}
+    assert endpoints["proxy"][0].extra_headers == {"X-Custom": "v1"}
 
 
 def test_load_endpoints_toml_accepts_extra_headers_alias(tmp_path: Path):
@@ -240,7 +236,7 @@ def test_load_endpoints_toml_accepts_extra_headers_alias(tmp_path: Path):
 
     endpoints = load_endpoints(str(registry_path))
 
-    assert endpoints["proxy"][0]["extra_headers"] == {"X-A": "a"}
+    assert endpoints["proxy"][0].extra_headers == {"X-A": "a"}
 
 
 def test_load_endpoints_toml_rejects_headers_and_extra_headers_together(

{verifiers-0.1.15.dev11 → verifiers-0.1.15.dev13}/tests/test_envs.py

@@ -158,6 +158,10 @@ def test_env(env_dir: Path, tmp_path_factory: pytest.TempPathFactory):
         f"{repo_root.as_posix()} && "
         "uv pip install "
         "--exclude-newer-package prime-pydantic-config=2026-05-20T00:00:00Z "
+        f"{(repo_root / 'packages' / 'tasksets').as_posix()} "
+        f"{(repo_root / 'packages' / 'harnesses').as_posix()} && "
+        "uv pip install "
+        "--exclude-newer-package prime-pydantic-config=2026-05-20T00:00:00Z "
         f"{env_dir.absolute().as_posix()}"
     )
     try: