hud-python 0.4.33__py3-none-any.whl → 0.4.34__py3-none-any.whl

hud/agents/misc/response_agent.py

@@ -16,7 +16,17 @@ class ResponseAgent:
     based on the agent's final response message.
     """
 
-    def __init__(self, api_key: str | None = None, model: str = "gpt-4o") -> None:
+    def __init__(
+        self, api_key: str | None = None, model: str = "gpt-4o", system_prompt: str | None = None
+    ) -> None:
+        """
+        Initialize the ResponseAgent.
+
+        Args:
+            api_key: The API key to use for the OpenAI client
+            model: The model to use for the OpenAI client (default: "gpt-4o")
+            system_prompt: The system prompt to use for the OpenAI client
+        """
         self.api_key = api_key or settings.openai_api_key or os.environ.get("OPENAI_API_KEY")
         if not self.api_key:
             raise ValueError(
@@ -26,23 +36,29 @@ class ResponseAgent:
         self.client = AsyncOpenAI(api_key=self.api_key)
         self.model = model
 
-        self.system_prompt = """
+        self.system_prompt = (
+            system_prompt
+            or """
         You are an assistant that helps determine the appropriate response to an agent's message.
         
         You will receive messages from an agent that is performing tasks for a user.
         Your job is to analyze these messages and respond with one of the following:
         
-        - STOP: If the agent indicates it has successfully completed a task, even if phrased as a question
-          like "I have entered the right values into this form. Would you like me to do anything else?"
-          or "Here is the website. Is there any other information you need?" or if the agent has
-          strongly determined it wants to stop the task.
-        
+        - STOP: If the agent indicates it has successfully completed a task or is stuck,
+          struggling or says it cannot complete the task, even if phrased as a question
+          like "I have entered the right values into this form. Would you like me to do
+          anything else?" or "Here is the website. Is there any other information you
+          need?" or if the agent has strongly determined it wants to stop the task like
+          "The task is infeasible. Can I help you with something else?"
+
         - CONTINUE: If the agent is asking for clarification before proceeding with a task
           like "I'm about to clear cookies from this website. Would you like me to proceed?"
-          or "I've entered the right values into this form. Would you like me to continue with the rest of the task?"
+          or "I've entered the right values into this form. Would you like me to continue
+          with the rest of the task?"
         
         Respond ONLY with one of these two options.
-        """  # noqa: E501
+        """
+        )
 
     async def determine_response(self, agent_message: str) -> ResponseType:
         """

hud/cli/__init__.py

@@ -585,6 +585,9 @@ def build(
     ),
     no_cache: bool = typer.Option(False, "--no-cache", help="Build without Docker cache"),
     verbose: bool = typer.Option(False, "--verbose", "-v", help="Show detailed output"),
+    platform: str | None = typer.Option(
+        None, "--platform", help="Set Docker target platform (e.g., linux/amd64)"
+    ),
 ) -> None:
     """🏗️ Build a HUD environment and generate lock file.
 
@@ -635,7 +638,7 @@ def build(
         else:
             i += 1
 
-    build_command(directory, tag, no_cache, verbose, env_vars)
+    build_command(directory, tag, no_cache, verbose, env_vars, platform)
 
 
 @app.command()

hud/cli/build.py

@@ -224,6 +224,7 @@ def build_docker_image(
     no_cache: bool = False,
     verbose: bool = False,
     build_args: dict[str, str] | None = None,
+    platform: str | None = None,
 ) -> bool:
     """Build a Docker image from a directory."""
     hud_console = HUDConsole()
@@ -236,7 +237,10 @@ def build_docker_image(
         return False
 
     # Build command
-    cmd = ["docker", "build", "-t", tag]
+    cmd = ["docker", "build"]
+    if platform:
+        cmd.extend(["--platform", platform])
+    cmd.extend(["-t", tag])
     if no_cache:
         cmd.append("--no-cache")
 
@@ -264,6 +268,7 @@ def build_environment(
     no_cache: bool = False,
     verbose: bool = False,
     env_vars: dict[str, str] | None = None,
+    platform: str | None = None,
 ) -> None:
     """Build a HUD environment and generate lock file."""
     hud_console = HUDConsole()
@@ -294,9 +299,8 @@ def build_environment(
     except Exception:
         default_image = f"{env_dir.name}:dev"
 
-    # Use provided tag or default
-    if not tag:
-        tag = default_image
+    # Determine final image tag to use
+    image_tag: str = tag if tag else default_image
 
     # Build temporary image first
     temp_tag = f"hud-build-temp:{int(time.time())}"
@@ -304,7 +308,14 @@ def build_environment(
     hud_console.progress_message(f"Building Docker image: {temp_tag}")
 
     # Build the image (env vars are for runtime, not build time)
-    if not build_docker_image(env_dir, temp_tag, no_cache, verbose):
+    if not build_docker_image(
+        env_dir,
+        temp_tag,
+        no_cache,
+        verbose,
+        build_args=None,
+        platform=platform,
+    ):
         hud_console.error("Docker build failed")
         raise typer.Exit(1)
 
@@ -422,21 +433,24 @@ def build_environment(
 
     # Build final image with label (uses cache from first build)
     # Also tag with version
-    base_name = tag.split(":")[0] if tag and ":" in tag else tag
+    base_name = image_tag.split(":")[0] if ":" in image_tag else image_tag
     version_tag = f"{base_name}:{new_version}"
 
-    label_cmd = [
-        "docker",
-        "build",
-        "--label",
-        f"org.hud.manifest.head={lock_hash}:{lock_size}",
-        "--label",
-        f"org.hud.version={new_version}",
-        "-t",
-        tag,
-        "-t",
-        version_tag,
-    ]
+    label_cmd = ["docker", "build"]
+    if platform is not None:
+        label_cmd.extend(["--platform", platform])
+    label_cmd.extend(
+        [
+            "--label",
+            f"org.hud.manifest.head={lock_hash}:{lock_size}",
+            "--label",
+            f"org.hud.version={new_version}",
+            "-t",
+            image_tag,
+            "-t",
+            version_tag,
+        ]
+    )
 
     label_cmd.append(str(env_dir))
 
@@ -457,14 +471,14 @@ def build_environment(
     hud_console.success("Built final image with lock file metadata")
 
     # NOW get the image ID after the final build
-    image_id = get_docker_image_id(tag)  # type: ignore
+    image_id = get_docker_image_id(image_tag)
     if image_id:
         # For local builds, store the image ID
         # Docker IDs come as sha256:hash, we want tag@sha256:hash
         if image_id.startswith("sha256:"):
-            lock_content["image"] = f"{tag}@{image_id}"
+            lock_content["image"] = f"{image_tag}@{image_id}"
         else:
-            lock_content["image"] = f"{tag}@sha256:{image_id}"
+            lock_content["image"] = f"{image_tag}@sha256:{image_id}"
 
         # Update the lock file with the new image reference
         with open(lock_path, "w") as f:
@@ -487,8 +501,8 @@ def build_environment(
 
     # Show the version tag as primary since that's what will be pushed
     hud_console.status_item("Built image", version_tag, primary=True)
-    if tag:
-        hud_console.status_item("Also tagged", tag)
+    if image_tag:
+        hud_console.status_item("Also tagged", image_tag)
     hud_console.status_item("Version", new_version)
     hud_console.status_item("Lock file", "hud.lock.yaml")
     hud_console.status_item("Tools found", str(analysis["toolCount"]))
@@ -500,7 +514,7 @@ def build_environment(
     hud_console.section_title("Next Steps")
     hud_console.info("Test locally:")
     hud_console.command_example("hud dev", "Hot-reload development")
-    hud_console.command_example(f"hud run {tag}", "Run the built image")
+    hud_console.command_example(f"hud run {image_tag}", "Run the built image")
     hud_console.info("")
     hud_console.info("Publish to registry:")
     hud_console.command_example("hud push", f"Push as {version_tag}")
@@ -517,6 +531,7 @@ def build_command(
     no_cache: bool = typer.Option(False, "--no-cache", help="Build without Docker cache"),
     verbose: bool = typer.Option(False, "--verbose", "-v", help="Show detailed output"),
     env_vars: dict[str, str] | None = None,
+    platform: str | None = None,
 ) -> None:
     """Build a HUD environment and generate lock file."""
-    build_environment(directory, tag, no_cache, verbose, env_vars)
+    build_environment(directory, tag, no_cache, verbose, env_vars, platform)

hud/cli/flows/tasks.py

@@ -100,7 +100,8 @@ def _ensure_built(env_dir: Path) -> dict[str, Any]:
         # Check Docker availability before attempting a build
         require_docker_running()
         # Run build (non-interactive). If Docker isn't running, this will raise and stop the flow.
-        build_environment(str(env_dir))
+        # Force linux/amd64 platform to ensure compatibility during RL flows.
+        build_environment(str(env_dir), platform="linux/amd64")
 
     # Load lock file
     with open(lock_path) as f:

hud/rl/actor.py

@@ -85,18 +85,19 @@ class Actor:
                     )
                 except TimeoutError:
                     hud_console.warning_log(f"Episode timed out for task {t.id}")
-                    return Trace(isError=True, content="Episode timeout")
+                    # Attach task so buffer grouping has key
+                    return Trace(isError=True, content="Episode timeout", task=t)
 
             results = await asyncio.gather(
                 *[run_with_timeout(t) for t in batch],
                 return_exceptions=True,
             )
 
-            # Normalize exceptions to error traces
-            for res in results:
+            # Normalize exceptions to error traces and ensure task is attached
+            for t, res in zip(batch, results, strict=False):
                 if isinstance(res, Exception):
                     hud_console.warning_log(f"Episode error: {res}")
-                    traces.append(Trace(isError=True, content=str(res)))
+                    traces.append(Trace(isError=True, content=str(res), task=t))
                 else:
                     traces.append(res)
 
@@ -113,7 +114,8 @@ class Actor:
 
         except Exception:
             logger.info("GOT EXCEPTION")
-            return Trace(isError=True)
+            # Preserve task on exception for grouping
+            return Trace(isError=True, task=task)
 
         result.info["tool_spec"] = agent.get_tool_schemas()
 

hud/rl/tests/test_learner.py

@@ -38,15 +38,20 @@ def make_sample(
     ref_logp_tok: torch.Tensor,
     advantage: float,
 ):
-    # Minimal object with required attributes for compute_loss
-    # inputs only needed for metrics token count
+    # Minimal-but-correct object for GRPOLearner.compute_loss.
+    # Needs assistant_mask (T-1) and attention_mask (T) for sanity_check().
     Tm1 = pol_logp_tok.size(-1)
-    inputs = {"input_ids": torch.zeros(1, Tm1 + 1, dtype=torch.long)}
+    inputs = {
+        "input_ids": torch.zeros(1, Tm1 + 1, dtype=torch.long),
+        "attention_mask": torch.ones(1, Tm1 + 1, dtype=torch.long),
+        "assistant_mask": torch.ones(1, Tm1, dtype=torch.bool),
+    }
     return TrainingSample(
         inputs=inputs,
         old_logprobs=old_logp_tok,
         ref_logprobs=ref_logp_tok,
-        advantage=torch.tensor(advantage, dtype=torch.float32),
+        # advantage must be 1D so .view(-1,1) works in compute_loss
+        advantage=torch.tensor([advantage], dtype=torch.float32),
     )
 
 
@@ -155,6 +160,13 @@ def test_skip_update_when_zero_adv(monkeypatch, learner_stub: GRPOLearner):
 
     monkeypatch.setattr(GRPOLearner, "prepare_groups", _stub_prepare_groups, raising=True)
 
+    # Return a zero scalar loss that *depends* on params so backward works,
+    # but has zero gradients (no update signal).
+    def _zero_loss(self, sample) -> torch.Tensor:
+        return sum(p.sum() for p in self.policy.parameters()) * 0.0
+
+    monkeypatch.setattr(GRPOLearner, "compute_loss", _zero_loss, raising=True)
+
     # Count optimizer.step calls
     steps = {"n": 0}
     # orig_step = learner_stub.optimizer.step
@@ -168,4 +180,7 @@ def test_skip_update_when_zero_adv(monkeypatch, learner_stub: GRPOLearner):
     assert any(p.requires_grad for p in learner_stub.policy.parameters())
 
     learner_stub.update([])
-    assert steps["n"] == 0
+    # With the current learner implementation we still call optimizer.step()
+    # even if the per-minibatch "advantage" is zero (the step is a no-op
+    # because the gradients are zero). So we expect exactly one step here.
+    assert steps["n"] == 1

hud/utils/tests/test_version.py

@@ -5,4 +5,4 @@ def test_import():
     """Test that the package can be imported."""
     import hud
 
-    assert hud.__version__ == "0.4.33"
+    assert hud.__version__ == "0.4.34"

hud/version.py

@@ -4,4 +4,4 @@ Version information for the HUD SDK.
 
 from __future__ import annotations
 
-__version__ = "0.4.33"
+__version__ = "0.4.34"

{hud_python-0.4.33.dist-info → hud_python-0.4.34.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hud-python
-Version: 0.4.33
+Version: 0.4.34
 Summary: SDK for the HUD platform.
 Project-URL: Homepage, https://github.com/hud-evals/hud-python
 Project-URL: Bug Tracker, https://github.com/hud-evals/hud-python/issues

{hud_python-0.4.33.dist-info → hud_python-0.4.34.dist-info}/RECORD

@@ -2,7 +2,7 @@ hud/__init__.py,sha256=JMDFUE1pP0J1Xl_miBdt7ERvoffZmTzSFe8yxz512A8,552
 hud/__main__.py,sha256=YR8Dq8OhINOsVfQ55PmRXXg4fEK84Rt_-rMtJ5rvhWo,145
 hud/settings.py,sha256=sMS31iW1m-5VpWk-Blhi5-obLcUA0fwxWE1GgJz-vqU,2708
 hud/types.py,sha256=RtNM2fPU1NAujTmZLOydQIU-ybk3gVRCoJ2TM2hJOlw,10752
-hud/version.py,sha256=7nCICMgtZOjBoirBGd5_5Ea-s2F7XAgLvEX_110KGAU,105
+hud/version.py,sha256=yDsOrKQyUBp8e8KHUcrT2U56qKvsk4D-58XSWaUqJMU,105
 hud/agents/__init__.py,sha256=UoIkljWdbq4bM0LD-mSaw6w826EqdEjOk7r6glNYwYQ,286
 hud/agents/base.py,sha256=_u1zR3gXzZ1RlTCUYdMcvgHqdJBC4-AB1lZt0yBx8lg,35406
 hud/agents/claude.py,sha256=wHiw8iAnjnRmZyKRKcOhagCDQMhz9Z6rlSBWqH1X--M,15781
@@ -11,17 +11,17 @@ hud/agents/langchain.py,sha256=1EgCy8jfjunsWxlPC5XfvfLS6_XZVrIF1ZjtHcrvhYw,9584
 hud/agents/openai.py,sha256=ovARRWNuHqKkZ2Q_OCYSVCIZckrh8XY2jUB2p2x1m88,14259
 hud/agents/openai_chat_generic.py,sha256=7n7timn3fvNRnL2xzWyOTeNTchej2r9cAL1mU6YnFdY,11605
 hud/agents/misc/__init__.py,sha256=BYi4Ytp9b_vycpZFXnr5Oyw6ncKLNNGml8Jrb7bWUb4,136
-hud/agents/misc/response_agent.py,sha256=OJdQJ76jP9xxQxVYJ-qPcdBxvFr8ABcwbP1f1I5zU5A,3227
+hud/agents/misc/response_agent.py,sha256=uMuRDkz5QgaMQliNzBRepond5sb7KyqIiKm3LstjVnw,3753
 hud/agents/tests/__init__.py,sha256=W-O-_4i34d9TTyEHV-O_q1Ai1gLhzwDaaPo02_TWQIY,34
 hud/agents/tests/test_base.py,sha256=bDznxQDv2ickRkw98joH9zfuZT6ItHbmWvQ67iboa4g,28733
 hud/agents/tests/test_claude.py,sha256=wqEKlzEvx8obz1sSm4NY0j-Zyt1qWNfDOmRqYIuAEd0,13069
 hud/agents/tests/test_client.py,sha256=uikgh6yhjPPX2RBU4XJQMz1mNox9uXjuwsP8t93id18,13337
 hud/agents/tests/test_grounded_openai_agent.py,sha256=VK8lUvHIjWicMX00VKPE-FZyjiJqTEhb80MuRRa9fVc,5437
 hud/agents/tests/test_openai.py,sha256=1S5IZuc3O3moSp70gqVGjc6m-_b49dCfz2fgX5IGvl4,7036
-hud/cli/__init__.py,sha256=aAqoGtdoSP8omcmKqiLp8OhxHr4nCXGupGUwhNAOEQM,40119
+hud/cli/__init__.py,sha256=xL1l5MfdWubd9AWe-cpW64WFS1SVsTgI8fdNdTZhIvs,40259
 hud/cli/__main__.py,sha256=fDH7XITyuDITwSDIVwRso06aouADO0CzTHKqp5TOwJE,143
 hud/cli/analyze.py,sha256=4u5oYfJMquOjT9PzzRTYVcTZDxDi0ilNP_g532_hpOU,14716
-hud/cli/build.py,sha256=SsZLQKYyMvIikLaSPYmg7-S1iJn-KkWXURhniQ0Ifi8,18015
+hud/cli/build.py,sha256=X8ykInqvDpsBU0rOU_x_sbp27YwCd160hplZLcwZFEg,18479
 hud/cli/clone.py,sha256=AwVDIuhr8mHb1oT2Af2HrD25SiTdwATpE6zd93vzLgA,6099
 hud/cli/debug.py,sha256=jtFW8J5F_3rhq1Hf1_SkJ7aLS3wjnyIs_LsC8k5cnzc,14200
 hud/cli/dev.py,sha256=56vQdH9oe_XGnOcRcFbNIsLEoBnpCl1eANlRFUeddHQ,31734
@@ -33,7 +33,7 @@ hud/cli/pull.py,sha256=Vd1l1-IwskyACzhtC8Df1SYINUZEYmFxrLl0s9cNN6c,12151
 hud/cli/push.py,sha256=dmjF-hGlMfq73tquDxsTuM9t50zrkE9PFJqW5vRmYSw,18380
 hud/cli/remove.py,sha256=8vGQyXDqgtjz85_vtusoIG8zurH4RHz6z8UMevQRYM4,6861
 hud/cli/flows/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-hud/cli/flows/tasks.py,sha256=CSdUA4vWMyLHke7pQgxsHzuPAj3CxmQuu66KjSvVai4,8822
+hud/cli/flows/tasks.py,sha256=R_zeIx0pvaN4CYYuNlIaemMegaT123FGAvDUjDDfFGo,8924
 hud/cli/rl/__init__.py,sha256=BeqXdmzPwVBptz4j796XJRxSC5B_9tQta5aKd0jDMvo,5000
 hud/cli/rl/config.py,sha256=iNhCxotM33OEiP9gqPvn8A_AxrBVe6fcFCQTvc13xzA,2884
 hud/cli/rl/display.py,sha256=hqJVGmO9csYinladhZwjF-GMvppYWngxDHajTyIJ_gM,5214
@@ -108,7 +108,7 @@ hud/otel/tests/__init__.py,sha256=VNJKBMaxTtbn7trW-1Ph50zCvCok_wTSGcI1HD6GOLA,43
 hud/otel/tests/test_processors.py,sha256=np0R4ssd9j6LJSJykJ5bNjl0POwNYNhgb7BqOZHwcMY,6778
 hud/rl/README.md,sha256=uFRpNFaEY8paq9k1C4miF7AGnbqHTGAsPmpcf9JIEeA,1189
 hud/rl/__init__.py,sha256=yYL7U1WV6L3mr3Hig48-4lhnryTaWj4nCXm4lG5vrYI,25
-hud/rl/actor.py,sha256=0YChXyxCz1wVBQ9lKb7vSl64_HQ24-DmYqCCxuORzJc,6747
+hud/rl/actor.py,sha256=n2f2BI9IOK__x7Seirq6EQI0yyicMBYd5BjPsc4T9rQ,6946
 hud/rl/buffer.py,sha256=FWGivdJ0YEYZZPK0bUyvjiKparaUgiBE9GzQLZj8kcA,15372
 hud/rl/chat_template.jinja,sha256=XTdzI8oFGEcSA-exKxyHaprwRDmX5Am1KEb0VxvUc6U,4965
 hud/rl/config.py,sha256=PAKYPCsKl8yg_j3gJSE5SJUgLM7j0lFy0K_Vt4-otDM,5384
@@ -119,7 +119,7 @@ hud/rl/types.py,sha256=lrLKo7iaqodYth2EyeuOQfLiuzXfYM2eJjPmpObrD7c,3965
 hud/rl/utils.py,sha256=IsgVUUibxnUzb32a4mu1sYrgJC1CwoG9E-Dd5y5VDOA,19115
 hud/rl/vllm_adapter.py,sha256=O2_TdTGIyNr9zRGhCw18XWjOKYzEM3049wvlyL2x0sc,4751
 hud/rl/tests/__init__.py,sha256=PXmD3Gs6xOAwaYKb4HnwZERDjX05N1QF-aU6ya0dBtE,27
-hud/rl/tests/test_learner.py,sha256=qfSHFFROteRb98TjBuAKjFmZjCGfuWXPysVvTAWJ7wQ,6025
+hud/rl/tests/test_learner.py,sha256=_xSr1UjH6rnvSA4MP232AYxrju9V1lHTBn7Z_Gi824o,6818
 hud/rl/utils/start_vllm_server.sh,sha256=ThPokrLK_Qm_uh916fHXXBfMlw1TC97P57-AEI5MuOc,910
 hud/samples/__init__.py,sha256=wgcN1IOLHhR4C1fFKqyvA7Yl9lJhJFf34zfKs-UMSus,128
 hud/samples/browser.py,sha256=7LkzGx2G5dA8RogZwORnxxpVsxMV2gF18D_hGJIEow8,973
@@ -201,10 +201,10 @@ hud/utils/tests/test_init.py,sha256=2QLQSGgyP9wJhOvPCusm_zjJad0qApOZi1BXpxcdHXQ,
 hud/utils/tests/test_mcp.py,sha256=0pUa16mL-bqbZDXp5NHBnt1gO5o10BOg7zTMHZ1DNPM,4023
 hud/utils/tests/test_progress.py,sha256=QSF7Kpi03Ff_l3mAeqW9qs1nhK50j9vBiSobZq7T4f4,7394
 hud/utils/tests/test_telemetry.py,sha256=5jl7bEx8C8b-FfFUko5pf4UY-mPOR-9HaeL98dGtVHM,2781
-hud/utils/tests/test_version.py,sha256=Wdb1xAhmZ4EgoOIqiOcIr3iRZIxEMUCPCgee6cAlR3s,160
+hud/utils/tests/test_version.py,sha256=YmDs3a476rPIo4rLodsx3zqGeIIeDCXLjenmKK4-rkE,160
 hud/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-hud_python-0.4.33.dist-info/METADATA,sha256=EQgm-qxFqkYHk78gbjyHW0KTUu03JKnBidSwzfDG4ZY,20861
-hud_python-0.4.33.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-hud_python-0.4.33.dist-info/entry_points.txt,sha256=jJbodNFg1m0-CDofe5AHvB4zKBq7sSdP97-ohaQ3ae4,63
-hud_python-0.4.33.dist-info/licenses/LICENSE,sha256=yIzBheVUf86FC1bztAcr7RYWWNxyd3B-UJQ3uddg1HA,1078
-hud_python-0.4.33.dist-info/RECORD,,
+hud_python-0.4.34.dist-info/METADATA,sha256=8ju6prkWIMJ5ChDPU-evkWqWucKZ1QyU3XTM2717RYA,20861
+hud_python-0.4.34.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+hud_python-0.4.34.dist-info/entry_points.txt,sha256=jJbodNFg1m0-CDofe5AHvB4zKBq7sSdP97-ohaQ3ae4,63
+hud_python-0.4.34.dist-info/licenses/LICENSE,sha256=yIzBheVUf86FC1bztAcr7RYWWNxyd3B-UJQ3uddg1HA,1078
+hud_python-0.4.34.dist-info/RECORD,,