synth-ai 0.2.9.dev0__py3-none-any.whl → 0.2.23.dev3__py3-none-any.whl

examples/swe/task_app/grpo_swe_mini.py

@@ -0,0 +1,604 @@
+"""Task app configuration for the mini-SWE agent integration."""
+
+from __future__ import annotations
+
+import json
+import logging
+import os
+from collections.abc import Iterable, Sequence
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any
+
+from synth_ai.task.apps import ModalDeploymentConfig, TaskAppEntry, register_task_app
+from synth_ai.task.contracts import RolloutMetrics, RolloutRequest, RolloutResponse, TaskInfo
+from synth_ai.task.datasets import TaskDatasetRegistry, TaskDatasetSpec
+from synth_ai.task.rubrics import load_rubric
+from synth_ai.task.server import ProxyConfig, RubricBundle, TaskAppConfig
+from synth_ai.task.tracing_utils import (
+    build_tracer_factory,
+    resolve_sft_output_dir,
+    resolve_tracing_db_url,
+    tracing_env_enabled,
+)
+from synth_ai.tracing_v3.session_tracer import SessionTracer
+
+logger = logging.getLogger(__name__)
+
+
+try:
+    from examples.swe.task_app.hosted.branching import (  # type: ignore
+        router as branching_router,
+    )
+    from examples.swe.task_app.hosted.environment_routes import (  # type: ignore # noqa: E501
+        router as environment_router,
+    )
+    from examples.swe.task_app.hosted.policy_routes import (  # type: ignore
+        router as policy_router,
+    )
+    from examples.swe.task_app.hosted.rollout import (  # type: ignore
+        RolloutEnvSpec as LegacyRolloutEnvSpec,
+    )
+    from examples.swe.task_app.hosted.rollout import (
+        RolloutPolicySpec as LegacyRolloutPolicySpec,
+    )
+    from examples.swe.task_app.hosted.rollout import (
+        RolloutRecordConfig as LegacyRolloutRecordConfig,
+    )
+    from examples.swe.task_app.hosted.rollout import (
+        RolloutRequest as LegacyRolloutRequest,
+    )
+    from examples.swe.task_app.hosted.rollout import (
+        RolloutResponse as LegacyRolloutResponse,
+    )
+    from examples.swe.task_app.hosted.rollout import (
+        RolloutSafetyConfig as LegacyRolloutSafetyConfig,
+    )
+    from examples.swe.task_app.hosted.rollout import (
+        execute_rollout as legacy_execute_rollout,
+    )
+    HAS_HOSTED = True
+except Exception:
+    try:  # pragma: no cover - optional dependency path
+        from examples.task_apps.crafter.task_app.synth_envs_hosted.branching import (  # type: ignore
+            BranchingEnvironmentConfig,
+        )
+        from examples.task_apps.crafter.task_app.synth_envs_hosted.environment_routes import (  # type: ignore # noqa: E501
+            CrafterEnvironmentRoutes,
+        )
+        from examples.task_apps.crafter.task_app.synth_envs_hosted.policy_routes import (  # type: ignore
+            PolicyRoutes,
+        )
+        from examples.task_apps.crafter.task_app.synth_envs_hosted.rollout import (  # type: ignore
+            RolloutPayload,
+        )
+        from examples.task_apps.crafter.task_app.synth_envs_hosted.rollout import (
+            EnvironmentConfig,
+        )
+        from examples.task_apps.crafter.task_app.synth_envs_hosted.rollout import (
+            PolicyConfig,
+        )
+        from examples.task_apps.crafter.task_app.synth_envs_hosted.rollout import (
+            RolloutRequest,
+        )
+        from examples.task_apps.crafter.task_app.synth_envs_hosted.rollout import (
+            RolloutResponse,
+        )
+        from examples.task_apps.crafter.task_app.synth_envs_hosted.rollout import (
+            RunSpec,
+        )
+        from examples.task_apps.crafter.task_app.synth_envs_hosted.rollout import (
+            ToolUse,
+        )
+        from examples.task_apps.crafter.task_app.hosted.rollout import (  # type: ignore
+            RolloutEnvSpec as LegacyRolloutEnvSpec,
+        )
+        from examples.task_apps.crafter.task_app.hosted.rollout import (
+            RolloutPolicySpec as LegacyRolloutPolicySpec,
+        )
+        from examples.task_apps.crafter.task_app.hosted.rollout import (
+            RolloutRecordConfig as LegacyRolloutRecordConfig,
+        )
+        from examples.task_apps.crafter.task_app.hosted.rollout import (
+            RolloutRequest as LegacyRolloutRequest,
+        )
+        from examples.task_apps.crafter.task_app.hosted.rollout import (
+            RolloutResponse as LegacyRolloutResponse,
+        )
+        from examples.task_apps.crafter.task_app.hosted.rollout import (
+            RolloutSafetyConfig as LegacyRolloutSafetyConfig,
+        )
+        from examples.task_apps.crafter.task_app.hosted.rollout import (
+            execute_rollout as legacy_execute_rollout,
+        )
+        HAS_HOSTED = True
+    except Exception as exc:  # pragma: no cover - optional dependency path
+        logger.warning(
+            "mini-swe task app running without hosted environment support: %s", exc
+        )
+        HAS_HOSTED = False
+
+
+_HERE = Path(__file__).resolve()
+
+
+def _resolve_repo_root() -> Path:
+    candidates = [
+        Path(os.getenv("SYNTH_AI_REPO_ROOT", "")).expanduser(),
+        _HERE.parents[3],
+        Path(__file__).resolve().parents[2],
+        Path.cwd(),
+    ]
+    for candidate in candidates:
+        if not candidate:
+            continue
+        try:
+            resolved = candidate.resolve()
+        except Exception:
+            continue
+        if (resolved / "pyproject.toml").exists():
+            return resolved
+        if (resolved / "synth_ai").is_dir():
+            return resolved
+    return Path(__file__).resolve().parents[3]
+
+
+REPO_ROOT = _resolve_repo_root()
+def _load_hf_dataset(source: str) -> list[dict[str, Any]]:
+    try:
+        from datasets import load_dataset  # type: ignore
+    except Exception as exc:  # pragma: no cover - optional dependency
+        raise RuntimeError(
+            "datasets package is required for hf:// dataset sources. "
+            "pip install datasets or set SWE_MINI_DATASET=file://<path>."
+        ) from exc
+
+    # source looks like hf://namespace/dataset:split
+    cleaned = source[len("hf://") :]
+    if ":" in cleaned:
+        dataset_id, split = cleaned.split(":", 1)
+    else:
+        dataset_id, split = cleaned, "train"
+    logger.info("Loading HuggingFace dataset %s split=%s", dataset_id, split)
+    ds = load_dataset(dataset_id, split=split)
+    return [dict(record) for record in ds]
+
+
+def _parse_records(text: str) -> list[dict[str, Any]]:
+    text = text.strip()
+    if not text:
+        return []
+    if text.startswith("["):
+        parsed = json.loads(text)
+        if isinstance(parsed, list):
+            return [dict(item) for item in parsed]
+        raise ValueError("Expected list JSON for dataset file")
+    records = []
+    for line in text.splitlines():
+        line = line.strip()
+        if not line:
+            continue
+        records.append(json.loads(line))
+    return records
+
+
+def _normalize_instance(raw: dict[str, Any]) -> dict[str, Any]:
+    instance_id = raw.get("instance_id") or raw.get("id")
+    if not instance_id:
+        raise ValueError(f"Dataset entry missing instance_id: {raw}")
+    problem_statement = raw.get("problem_statement") or raw.get("prompt") or ""
+    instructions = raw.get("instructions") or raw.get("extra") or ""
+    metadata = raw.get("metadata") or {}
+    if not isinstance(metadata, dict):
+        metadata = {"raw_metadata": metadata}
+    for key, value in raw.items():
+        if key in {"metadata", "instructions", "problem_statement"}:
+            continue
+        metadata.setdefault(key, value)
+    metadata.setdefault("raw_instance", raw)
+    metadata.setdefault("instance_id", instance_id)
+    if "image_name" not in metadata:
+        iid = str(instance_id).replace("__", "_1776_")
+        metadata["image_name"] = f"docker.io/swebench/sweb.eval.x86_64.{iid}:latest".lower()
+    return {
+        "instance_id": instance_id,
+        "problem_statement": problem_statement,
+        "instructions": instructions,
+        "metadata": metadata,
+    }
+
+
+def _load_instances() -> list[dict[str, Any]]:
+    default_dataset = "hf://princeton-nlp/SWE-Bench_Verified:test"
+    source = os.getenv("SWE_MINI_DATASET", default_dataset).strip()
+    if not source:
+        source = default_dataset
+
+    if source.startswith("file://"):
+        path = Path(source[len("file://") :]).expanduser()
+        if not path.exists():
+            raise FileNotFoundError(f"SWE_MINI_DATASET file not found: {path}")
+        data = path.read_text(encoding="utf-8")
+        records = _parse_records(data)
+    elif source.startswith("hf://"):
+        records = _load_hf_dataset(source)
+    else:
+        path = Path(source).expanduser()
+        if path.exists():
+            data = path.read_text(encoding="utf-8")
+            records = _parse_records(data)
+        else:
+            raise ValueError(
+                f"Unsupported SWE_MINI_DATASET value '{source}'. "
+                "Use file://..., or hf://dataset:split."
+            )
+
+    normalised = []
+    for record in records:
+        try:
+            normalised.append(_normalize_instance(record))
+        except Exception as exc:
+            logger.warning("Skipping invalid dataset entry: %s", exc)
+    if not normalised:
+        raise RuntimeError("No valid mini-swe dataset entries found.")
+    return normalised
+
+
+@dataclass
+class MiniSweDataset:
+    instances: list[dict[str, Any]]
+
+    def __post_init__(self) -> None:
+        self.by_id = {item["instance_id"]: item for item in self.instances}
+
+    def ids(self) -> list[str]:
+        return [item["instance_id"] for item in self.instances]
+
+    def get(self, instance_id: str) -> dict[str, Any]:
+        if instance_id not in self.by_id:
+            raise KeyError(f"Unknown mini-swe instance_id: {instance_id}")
+        return self.by_id[instance_id]
+
+    def sample_by_index(self, index: int) -> dict[str, Any]:
+        if not self.instances:
+            raise RuntimeError("Mini-swe dataset is empty")
+        return self.instances[index % len(self.instances)]
+
+
+DATASET_SPEC = TaskDatasetSpec(
+    id="mini_swe_sample",
+    name="mini-SWE Tasks",
+    version="0.1.0",
+    splits=["train"],
+    default_split="train",
+    description="Interactive SWE tasks executed via mini-swe-agent environments.",
+)
+
+
+def build_dataset() -> tuple[TaskDatasetRegistry, MiniSweDataset]:
+    registry = TaskDatasetRegistry()
+    dataset = MiniSweDataset(_load_instances())
+    registry.register(DATASET_SPEC, lambda _spec: dataset, cache=True)
+    return registry, dataset
+
+
+def _base_task_info(dataset: MiniSweDataset) -> TaskInfo:
+    return TaskInfo(
+        task={"id": "swe_mini", "name": "mini-SWE Tasks", "version": "0.1.0"},
+        environment="swe-mini",
+        action_space={
+            "type": "tool",
+            "tools": ["run_command", "submit_patch"],
+            "description": "Issue bash commands or submit the final patch.",
+        },
+        observation={
+            "summary": "Step-wise command output and submission status.",
+            "keys": ["task", "history", "last", "submitted"],
+        },
+        dataset={
+            **DATASET_SPEC.model_dump(),
+            "instances": dataset.ids()[:50],
+        },
+        rubric={
+            "version": "1",
+            "criteria_count": 2,
+            "source": "inline",
+            "aggregation": "weighted_sum",
+        },
+        inference={
+            "supports_proxy": True,
+            "endpoints": {
+                "openai": "/proxy/v1/chat/completions",
+                "groq": "/proxy/groq/v1/chat/completions",
+            },
+            "tool": {"name": "run_command", "parallel_tool_calls": False},
+        },
+        limits={"max_ops": 2000, "max_time_s": 7200},
+    )
+
+
+OUTCOME_RUBRIC = load_rubric(
+    {
+        "version": "1",
+        "goal_text": "Complete the software engineering task and ensure tests pass.",
+        "aggregation": "weighted_sum",
+        "criteria": [
+            {
+                "id": "functional",
+                "description": "All acceptance tests and lint checks succeed.",
+                "weight": 1.0,
+            },
+            {
+                "id": "quality",
+                "description": "Code changes follow project conventions and include required updates.",
+                "weight": 1.0,
+            },
+        ],
+    }
+)
+
+EVENTS_RUBRIC = load_rubric(
+    {
+        "version": "1",
+        "goal_text": "Encourage deliberate, well-scoped shell interactions.",
+        "aggregation": "weighted_sum",
+        "criteria": [
+            {
+                "id": "productive_steps",
+                "description": "Commands meaningfully progress the task (editing files, running tests, inspecting context).",
+                "weight": 1.0,
+            }
+        ],
+    }
+)
+
+
+def describe_taskset(dataset: MiniSweDataset) -> dict[str, Any]:
+    return {
+        **DATASET_SPEC.model_dump(),
+        "instance_ids": dataset.ids(),
+    }
+
+
+def provide_task_instances(
+    dataset: MiniSweDataset, base_info: TaskInfo, seeds: Sequence[int]
+) -> Iterable[TaskInfo]:
+    infos: list[TaskInfo] = []
+    base_observation = getattr(base_info, "observation", None)
+    if hasattr(base_observation, "model_dump"):
+        base_observation_data = base_observation.model_dump()
+    elif isinstance(base_observation, dict):
+        base_observation_data = dict(base_observation)
+    else:
+        base_observation_data = {}
+
+    for seed in seeds:
+        instance = dataset.sample_by_index(int(seed))
+        infos.append(
+            TaskInfo(
+                task=base_info.task,
+                environment=base_info.environment,
+                action_space=base_info.action_space,
+                observation={
+                    **base_observation_data,
+                    "instance_id": instance["instance_id"],
+                },
+                dataset={
+                    **base_info.dataset.model_dump(),
+                    "instance_id": instance["instance_id"],
+                },
+                rubric=base_info.rubric,
+                inference=base_info.inference,
+                limits=base_info.limits,
+            )
+        )
+    return infos
+
+
+def _ensure_env_has_task(
+    dataset: MiniSweDataset, env_spec: LegacyRolloutEnvSpec
+) -> LegacyRolloutEnvSpec:
+    config = dict(env_spec.config or {})
+    if "task" not in config:
+        instance_id = config.get("instance_id")
+        if not instance_id:
+            raise ValueError("mini-swe rollout request requires env.config.instance_id")
+        config["task"] = dataset.get(instance_id)
+    env_cfg = dict(config.get("environment") or {})
+    if "environment_class" not in env_cfg and os.getenv("MORPH_API_KEY"):
+        env_cfg["environment_class"] = "morph"
+    config["environment"] = env_cfg
+    return env_spec.model_copy(update={"config": config})
+
+
+def build_config() -> TaskAppConfig:
+    registry, dataset = build_dataset()
+    base_info = _base_task_info(dataset)
+
+    tracing_enabled = tracing_env_enabled()
+    tracing_db_url = resolve_tracing_db_url()
+    tracer_factory = build_tracer_factory(SessionTracer, enabled=tracing_enabled, db_url=tracing_db_url)
+    sft_output_dir = resolve_sft_output_dir()
+
+    hosted_task_app = None
+    if HAS_HOSTED:
+        HostedTaskAppCls = None
+        try:
+            from examples.swe.task_app.hosted.hosted_app import (  # type: ignore
+                TaskApp as HostedTaskApp,
+            )
+            HostedTaskAppCls = HostedTaskApp
+        except Exception:
+            try:
+                from examples.task_apps.crafter.task_app.synth_envs_hosted.hosted_app import (  # type: ignore
+                    create_app,
+                )
+                HostedTaskAppCls = create_app
+            except Exception as exc:  # pragma: no cover - optional dependency path
+                logger.warning("Unable to import HostedTaskApp for swe-mini: %s", exc)
+        if HostedTaskAppCls is not None:
+            hosted_task_app = HostedTaskAppCls(
+                service_base_url=os.getenv("SWE_MINI_SERVICE_BASE_URL"),
+                vllm_base_url=os.getenv(
+                    "SWE_MINI_VLLM_BASE_URL",
+                    "http://localhost:8020/proxy",
+                ),
+                default_model=os.getenv("SWE_MINI_DEFAULT_MODEL"),
+            )
+
+    app_state: dict[str, Any] = {
+        "dataset": dataset,
+        "allowed_environments": ["swe-mini"],
+        "tracing_enabled": tracing_enabled,
+    }
+    if tracer_factory is not None:
+        app_state["session_tracer_factory"] = tracer_factory
+    if sft_output_dir:
+        app_state["sft_output_dir"] = sft_output_dir
+    if hosted_task_app is not None:
+        app_state["task_app"] = hosted_task_app
+
+    if tracing_enabled:
+        logger.info("[swe-mini:tracing] enabled (db=%s)", tracing_db_url or "default")
+    else:
+        logger.info("[swe-mini:tracing] disabled")
+    if sft_output_dir:
+        logger.info("[swe-mini:sft] writing JSONL to %s", sft_output_dir)
+
+    routers: tuple = (environment_router, policy_router, branching_router) if HAS_HOSTED else ()
+
+    async def rollout_executor(request: RolloutRequest, fastapi_request) -> RolloutResponse:
+        if not HAS_HOSTED:
+            return RolloutResponse(
+                run_id=request.run_id,
+                trajectories=[],
+                branches={},
+                metrics=RolloutMetrics(
+                    episode_returns=[],
+                    mean_return=0.0,
+                    num_steps=0,
+                    num_episodes=0,
+                ),
+                aborted=True,
+                ops_executed=0,
+                trace=None,
+            )
+
+        env_spec = _ensure_env_has_task(dataset, request.env)
+
+        legacy_request = LegacyRolloutRequest(
+            run_id=request.run_id,
+            mode=request.mode,  # Preserve mode for nested requests
+            env=LegacyRolloutEnvSpec(
+                env_id=request.env.env_id,
+                env_name=env_spec.env_name or "swe-mini",
+                config=env_spec.config,
+                seed=request.env.seed,
+            ),
+            policy=LegacyRolloutPolicySpec(
+                policy_id=request.policy.policy_id,
+                policy_name=request.policy.policy_name or "swe-mini-react",
+                config=request.policy.config,
+            ),
+            ops=request.ops,
+            record=LegacyRolloutRecordConfig(**request.record.model_dump()),
+            on_done=request.on_done,
+            branch=getattr(request, "branch", None),
+            safety=LegacyRolloutSafetyConfig(**request.safety.model_dump()),
+            training_session_id=request.training_session_id,
+            synth_base_url=request.synth_base_url,
+        )
+
+        legacy_response: LegacyRolloutResponse = await legacy_execute_rollout(
+            legacy_request, fastapi_request
+        )
+        data = legacy_response.model_dump()
+        metrics = data.get("metrics", {}) or {}
+        metrics.setdefault("details", {})
+        data["metrics"] = metrics
+        return RolloutResponse.model_validate(data)
+
+    async def _noop_rollout(request: RolloutRequest, fastapi_request) -> RolloutResponse:
+        return RolloutResponse(
+            run_id=request.run_id,
+            trajectories=[],
+            branches={},
+            metrics=RolloutMetrics(
+                episode_returns=[],
+                mean_return=0.0,
+                num_steps=0,
+                num_episodes=0,
+            ),
+            aborted=True,
+            ops_executed=0,
+            trace=None,
+        )
+
+    rollout_callable = rollout_executor if HAS_HOSTED else _noop_rollout
+
+    config = TaskAppConfig(
+        app_id="swe-mini",
+        name="mini-SWE Task App",
+        description="Interactive SWE tasks executed via mini-swe-agent environments.",
+        base_task_info=base_info,
+        describe_taskset=lambda: describe_taskset(dataset),
+        provide_task_instances=lambda seeds: provide_task_instances(dataset, base_info, seeds),
+        rollout=rollout_callable,
+        dataset_registry=registry,
+        rubrics=RubricBundle(outcome=OUTCOME_RUBRIC, events=EVENTS_RUBRIC),
+        proxy=ProxyConfig(enable_openai=True, enable_groq=True),
+        routers=routers,
+        app_state=app_state,
+        cors_origins=["*"],
+    )
+    return config
+
+
+register_task_app(
+    entry=TaskAppEntry(
+        app_id="swe-mini",
+        description="mini-swe-agent task app with rollout + proxy endpoints",
+        config_factory=build_config,
+        aliases=("mini-swe", "swe-mini-task"),
+        modal=ModalDeploymentConfig(
+            app_name="swe-mini-task-app",
+            python_version="3.11",
+            pip_packages=(
+                "fastapi>=0.109.0",
+                "uvicorn>=0.23.0",
+                "pydantic>=2.7.0",
+                "numpy>=1.24.0",
+                "aiohttp>=3.8.0",
+                "httpx>=0.24.0",
+                "python-dotenv>=1.0.1",
+                "sqlalchemy>=2.0.42",
+                "aiosqlite>=0.21.0",
+                "greenlet>=3.0.3",
+                "modal>=0.63.0",
+                "tenacity>=8.2.3",
+                "swebench[modal]>=1.1.0",
+                "swe-rex[modal]>=1.4.0",
+                "mini-swe-agent>=1.14.2",
+                "datasets>=2.18.0",
+                "litellm>=1.75.5",
+                "rich>=13.7.0",
+                "jinja2>=3.1.3",
+            ),
+            extra_local_dirs=(
+                (str(REPO_ROOT / "synth_ai"), "/opt/synth_ai_repo/synth_ai"),
+                (
+                    str(REPO_ROOT / "examples" / "swe" / "task_app" / "hosted"),
+                    "/opt/synth_ai_repo/examples/swe/task_app/hosted",
+                ),
+                (
+                    str(_HERE.parent),
+                    "/opt/synth_ai_repo/examples/swe/task_app",
+                ),
+            ),
+            secret_names=("swe-mini-environment", "groq-api-key", "openai-api-key"),
+            memory=32768,
+            cpu=6.0,
+            max_containers=10,
+        ),
+    )
+)