freesolo 0.2.45__tar.gz → 0.2.47__tar.gz

{freesolo-0.2.45 → freesolo-0.2.47}/.github/workflows/publish-packages.yml

@@ -264,7 +264,12 @@ jobs:
             echo "::error::NPM_TOKEN is not configured; refusing to skip publish."
             exit 1
           fi
-          bun publish --access public
+          # bun publish does not pick up NODE_AUTH_TOKEN or ~/.npmrc auth, so
+          # publish the bun-built package with npm and a project npmrc.
+          umask 077
+          printf '//registry.npmjs.org/:_authToken=%s\n' "$NODE_AUTH_TOKEN" > .npmrc
+          npm publish --access public
+          rm -f .npmrc
 
       - name: No npm package changes
         if: github.event_name == 'push' && steps.changes.outputs.npm_changed == 'false'

{freesolo-0.2.45 → freesolo-0.2.47}/.github/workflows/python-checks.yml

@@ -26,16 +26,16 @@ jobs:
         run: python3 -m pip install --upgrade uv
 
       - name: Install dependencies
-        run: uv sync --locked --extra dev
+        run: uv sync --locked --group dev
 
       - name: Python compile check
         run: python3 -m py_compile $(find pypi tests -name '*.py' -print)
 
       - name: Ruff check
-        run: uv run --extra dev python -m ruff check .
+        run: uv run python -m ruff check .
 
       - name: Ruff format check
-        run: uv run --extra dev python -m ruff format --check .
+        run: uv run python -m ruff format --check .
 
       - name: Tests
-        run: uv run --extra dev python -m pytest tests
+        run: uv run python -m pytest tests

{freesolo-0.2.45 → freesolo-0.2.47}/AGENTS.md

@@ -21,3 +21,15 @@ This is a Python SDK (`freesolo`) for tracing, evaluating, and training LLM appl
 - When running examples outside of tests, set `PYTHONPATH="$PWD/pypi"` so the local source is used.
 - The `--local` flag on examples runs scorers locally without requiring `FREESOLO_API_KEY`.
 - Dev dependencies (`pytest`, `ruff`) are in the `[project.optional-dependencies] dev` group; use `uv sync --dev --extra dev` to install them.
+
+### Deployment / auto-pull
+
+- The freesolo agent-worker on the deploy VM does **not** install this SDK from
+  PyPI — it bind-mounts a host checkout of this repo and imports from source
+  (`PYTHONPATH=/freesolo-sdk/pypi`; see `freesolo`'s `docker-compose.yml`).
+- A push to `main` is auto-pulled: the deploy host polls `origin/main` every
+  ~2 min, fast-forwards `~/freesolo-sdk`, and restarts the agent-worker so it
+  re-imports the new code (`freesolo/scripts/sync-sdk.sh`, driven by
+  `freesolo-sdk-sync.timer`). So merging to `main` here updates the running
+  worker without a manual deploy — but it can interrupt an in-progress job
+  (requeue-stale re-enqueues it). Keep `main` deployable.

freesolo-0.2.47/PKG-INFO

@@ -0,0 +1,56 @@
+Metadata-Version: 2.4
+Name: freesolo
+Version: 0.2.47
+Summary: Environment and dataset helpers for Freesolo-generated repos.
+Requires-Python: >=3.10
+Requires-Dist: typing-extensions>=4.8.0
+Provides-Extra: bson
+Requires-Dist: pymongo>=4.0.0; extra == 'bson'
+Description-Content-Type: text/markdown
+
+# freesolo
+
+`freesolo` is the published Python SDK surface for generated repos: environments and
+datasets used for evaluation and task definition.
+
+It is intentionally narrow:
+
+- `freesolo.environments`
+- `freesolo.datasets`
+
+Everything else (evaluation, tracing, and internal helpers) is kept in the
+repository for internal workflows but is not part of the public SDK contract.
+
+## Install
+
+```bash
+pip install freesolo
+```
+
+From source:
+
+```bash
+cd freesolo-sdk
+export PYTHONPATH="$PWD/pypi"
+```
+
+## Example
+
+```python
+from freesolo.datasets import load_dataset
+from freesolo.environments import load_environment
+
+dataset = load_dataset("support.jsonl")
+environment = load_environment("freesolo/environment.py:load_environment")
+
+print(len(dataset.records))
+print(type(environment).__name__)
+```
+
+## API Guidance
+
+Use `freesolo.datasets` for task examples and `freesolo.environments` for environment
+loading/scoring interfaces.
+
+- No command-line help surface is published as part of the SDK contract.
+- Hidden modules remain available in source history for internal tooling only.

freesolo-0.2.47/README.md

@@ -0,0 +1,46 @@
+# freesolo
+
+`freesolo` is the published Python SDK surface for generated repos: environments and
+datasets used for evaluation and task definition.
+
+It is intentionally narrow:
+
+- `freesolo.environments`
+- `freesolo.datasets`
+
+Everything else (evaluation, tracing, and internal helpers) is kept in the
+repository for internal workflows but is not part of the public SDK contract.
+
+## Install
+
+```bash
+pip install freesolo
+```
+
+From source:
+
+```bash
+cd freesolo-sdk
+export PYTHONPATH="$PWD/pypi"
+```
+
+## Example
+
+```python
+from freesolo.datasets import load_dataset
+from freesolo.environments import load_environment
+
+dataset = load_dataset("support.jsonl")
+environment = load_environment("freesolo/environment.py:load_environment")
+
+print(len(dataset.records))
+print(type(environment).__name__)
+```
+
+## API Guidance
+
+Use `freesolo.datasets` for task examples and `freesolo.environments` for environment
+loading/scoring interfaces.
+
+- No command-line help surface is published as part of the SDK contract.
+- Hidden modules remain available in source history for internal tooling only.

{freesolo-0.2.45 → freesolo-0.2.47}/npm/README.md

@@ -8,7 +8,7 @@ This npm package intentionally contains only:
 - tracing helpers for exporting OpenTelemetry spans to Freesolo
 - evaluation primitives and `EvaluationClient`
 
-It does not include Freesolo training, datasets, GEPA, Tinker, or generated
+It does not include Freesolo training, datasets, GEPA, AutoSLM, or generated
 Python training-repo helpers.
 
 ## Tracing

{freesolo-0.2.45 → freesolo-0.2.47}/npm/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@freesolo/sdk",
-  "version": "0.2.45",
+  "version": "0.2.46",
   "description": "Tracing and evaluation utilities for TypeScript LLM applications.",
   "type": "module",
   "main": "./dist/index.js",

{freesolo-0.2.45 → freesolo-0.2.47}/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@freesolo/sdk",
-  "version": "0.2.45",
+  "version": "0.2.47",
   "description": "Tracing and evaluation utilities for LLM applications.",
   "type": "module",
   "main": "./npm/dist/index.js",

freesolo-0.2.47/pypi/freesolo/README.md

@@ -0,0 +1,47 @@
+# Freesolo SDK Package Map
+
+This package exposes the public surface for generated repos:
+
+- `freesolo.environments` for environment loading, scoring helpers, and episode metadata.
+- `freesolo.datasets` for task-record loading and prompt construction.
+
+Everything else in `pypi/freesolo/*` (evaluation, tracing, and utilities) is
+not part of the published SDK public contract.
+
+## Public Imports
+
+Use these package areas:
+
+- `freesolo.environments`: environment interface and environment utilities.
+- `freesolo.datasets`: dataset utilities and task examples.
+
+## Install
+
+```bash
+pip install freesolo
+```
+
+## From a repo checkout
+
+```bash
+cd freesolo-sdk
+export PYTHONPATH="$PWD/pypi"
+```
+
+```python
+from freesolo.datasets import load_dataset
+from freesolo.environments import load_environment
+
+dataset = load_dataset("support.jsonl")
+environment = load_environment("freesolo/environment.py:load_environment")
+
+print(len(dataset.records))
+print(type(environment).__name__)
+```
+
+## Public API
+
+The root module exports nothing directly. Import from:
+
+- `freesolo.environments`
+- `freesolo.datasets`

{freesolo-0.2.45 → freesolo-0.2.47}/pypi/freesolo/datasets/README.md

@@ -1,7 +1,7 @@
 # Datasets
 
-Dataset helpers turn raw records into `TaskExample` objects and SFT
-conversations.
+Dataset helpers turn raw records into `TaskExample` objects and generated
+conversation prompts.
 
 ## Public Imports
 
@@ -28,7 +28,7 @@ Each record must include one task field:
 - `query`
 - `input`
 
-Labeled/SFT records should also include one target field:
+Labeled records should also include one target field:
 
 - `ground_truth`
 - `expected_output`

freesolo-0.2.47/pypi/freesolo/datasets/_compat.py

@@ -0,0 +1,53 @@
+from __future__ import annotations
+
+import json
+from collections.abc import Mapping
+from typing import TYPE_CHECKING, TypedDict
+
+if TYPE_CHECKING:
+    JsonValue = (
+        str | int | float | bool | None | list["JsonValue"] | dict[str, "JsonValue"]
+    )
+    JsonObject = dict[str, JsonValue]
+    MetadataDict = dict[str, JsonValue]
+else:
+    JsonValue = object
+    JsonObject = dict[str, object]
+    MetadataDict = dict[str, object]
+
+
+class ChatMessage(TypedDict):
+    role: str
+    content: str
+
+
+def json_safe_value(value: object) -> JsonValue:
+    if value is None or isinstance(value, (str, int, float, bool)):
+        return value
+    if isinstance(value, Mapping):
+        return {str(key): json_safe_value(item) for key, item in value.items()}
+    if isinstance(value, (list, tuple, set)):
+        return [json_safe_value(item) for item in value]
+    try:
+        json.dumps(value)
+    except (TypeError, ValueError):
+        return str(value)
+    return value
+
+
+def serialize_value(value: object, *, pretty: bool = True) -> str:
+    if isinstance(value, str):
+        return value.strip()
+    payload = json_safe_value(value)
+    if pretty:
+        return json.dumps(payload, indent=2, sort_keys=True, ensure_ascii=True)
+    return json.dumps(payload, sort_keys=True, separators=(",", ":"), ensure_ascii=True)
+
+
+__all__ = [
+    "ChatMessage",
+    "JsonObject",
+    "JsonValue",
+    "MetadataDict",
+    "serialize_value",
+]

{freesolo-0.2.45 → freesolo-0.2.47}/pypi/freesolo/datasets/core.py

@@ -3,9 +3,7 @@ from __future__ import annotations
 from dataclasses import dataclass
 from typing import Protocol
 
-from freesolo.contracts.markdown import ChatMessage
-from freesolo.utils.core import serialize_value
-
+from ._compat import ChatMessage, serialize_value
 from .records import load_task_examples
 from .types import DatasetSource, TaskExample
 

{freesolo-0.2.45 → freesolo-0.2.47}/pypi/freesolo/datasets/records.py

@@ -5,8 +5,7 @@ import json
 from pathlib import Path
 from typing import Any, TypedDict
 
-from freesolo.utils.core import serialize_value
-
+from ._compat import serialize_value
 from .types import DatasetRecord, DatasetSource, TaskExample
 
 try:

{freesolo-0.2.45 → freesolo-0.2.47}/pypi/freesolo/datasets/types.py

@@ -5,7 +5,7 @@ from dataclasses import dataclass, field
 from pathlib import Path
 from typing import TypeAlias
 
-from freesolo.utils.core import JsonObject, MetadataDict
+from ._compat import JsonObject, MetadataDict
 
 DatasetRecord = JsonObject
 DatasetMetadata = MetadataDict

freesolo-0.2.47/pypi/freesolo/environments/README.md

@@ -0,0 +1,46 @@
+# Freesolo Environments
+
+Generated repos should use one canonical environment module:
+
+```text
+freesolo/environment.py
+```
+
+That file must expose:
+
+```python
+def load_environment(
+    *,
+    contract_path: str | None = None,
+    dataset_path: str | None = None,
+    reward_command: str | None = None,
+    mode: str = "eval",
+    **_: object,
+) -> Environment:
+    return RepoEnvironment(
+        contract_path=contract_path,
+        dataset_path=dataset_path,
+        reward_command=reward_command,
+        mode=mode,
+    )
+```
+
+For authoring, prefer the explicit branch base:
+
+- `EnvironmentSingleTurn` for one prompt → one assistant response tasks.
+- `EnvironmentMultiTurn` for bounded transcript-based tasks.
+
+Implement one concrete environment and keep environment loading in `load_environment()`.
+
+## Required environment API
+
+- `load_environment()` must return `EnvironmentSingleTurn` or `EnvironmentMultiTurn`.
+- `score_response(example, response_text)` must return `RewardResult`.
+- Optionally override `build_prompt_messages`, `extract_response_text`,
+  `normalize_response_text`, and episode handlers.
+
+## Optional helper
+
+`Environment.get_grpo_config()` exists for legacy compatibility. In published
+SDK builds, this helper may return a lightweight object when non-public training
+helpers are unavailable.

{freesolo-0.2.45 → freesolo-0.2.47}/pypi/freesolo/environments/__init__.py

@@ -4,11 +4,9 @@ from .base import (
     EnvironmentSingleTurn,
     load_environment,
 )
-from .evaluation import (
-    EnvironmentGeneration,
-)
 from .types import (
     EnvironmentEpisode,
+    EnvironmentGeneration,
     EnvironmentStepResult,
     EnvironmentTurn,
     RewardMetric,

freesolo-0.2.45/pypi/freesolo/contracts/markdown.py → freesolo-0.2.47/pypi/freesolo/environments/_compat.py

@@ -3,9 +3,24 @@ from __future__ import annotations
 import json
 import re
 from pathlib import Path
-from typing import cast
+from typing import TYPE_CHECKING, TypedDict, cast
+
+if TYPE_CHECKING:
+    from typing import TypeAlias
+
+    JsonValue: TypeAlias = (
+        str | int | float | bool | None | list["JsonValue"] | dict[str, "JsonValue"]
+    )
+    MetadataDict = dict[str, JsonValue]
+else:
+    JsonValue = object
+    MetadataDict = dict[str, object]
+
+
+class ChatMessage(TypedDict):
+    role: str
+    content: str
 
-from .types import ChatMessage, ContractMessageSpec, ContractSpec
 
 _FREESOLO_CONTRACT_BLOCK = re.compile(
     r"```(?:json\s+)?freesolo-contract\s*(.*?)```",
@@ -13,6 +28,21 @@ _FREESOLO_CONTRACT_BLOCK = re.compile(
 )
 
 
+class ContractMessageSpec(TypedDict, total=False):
+    role: str
+    content: str
+
+
+class PromptConfig(TypedDict, total=False):
+    system: str
+    user: str
+    messages: list[ContractMessageSpec]
+
+
+class ContractSpec(TypedDict, total=False):
+    prompt: PromptConfig
+
+
 def load_contract_text(path: str | Path) -> str:
     return Path(path).read_text(encoding="utf-8").strip()
 
@@ -34,10 +64,6 @@ def extract_contract_spec(contract_text: str) -> ContractSpec | None:
     return cast(ContractSpec, parsed)
 
 
-def load_contract_spec(path: str | Path) -> ContractSpec | None:
-    return extract_contract_spec(load_contract_text(path))
-
-
 def build_oracle_messages(
     task_text: str,
     contract_text: str,
@@ -105,3 +131,14 @@ def _render_prompt_content(content: str, task_text: str) -> str:
     for placeholder in ("{input}", "{task}", "{query}"):
         rendered = rendered.replace(placeholder, task_text)
     return rendered
+
+
+__all__ = [
+    "ChatMessage",
+    "ContractSpec",
+    "JsonValue",
+    "MetadataDict",
+    "build_oracle_messages",
+    "extract_contract_spec",
+    "load_contract_text",
+]

{freesolo-0.2.45 → freesolo-0.2.47}/pypi/freesolo/environments/base.py

@@ -13,13 +13,8 @@ from types import ModuleType
 from typing import TypeVar
 
 from freesolo.datasets import TaskExample
-from freesolo.training import GrpoConfig
 
-from ..contracts.markdown import (
-    ChatMessage,
-    ContractSpec,
-    build_oracle_messages,
-)
+from ._compat import ChatMessage, ContractSpec, build_oracle_messages
 from .types import (
     EnvironmentEpisode,
     EnvironmentStepResult,
@@ -34,7 +29,7 @@ _ScoreItem = TypeVar("_ScoreItem")
 
 
 class Environment(ABC):
-    """Task behavior adapter used by GRPO/RL, GEPA, and oracle generation.
+    """Task behavior adapter used by generated environments and oracle flows.
 
     Single-turn and multi-turn execution are peer branches: the default episode
     hooks implement the direct prompt/response branch, and interactive
@@ -107,8 +102,8 @@ class Environment(ABC):
             contract_spec=contract_spec,
         )
 
-    def get_grpo_config(self) -> GrpoConfig:
-        return GrpoConfig()
+    def get_grpo_config(self) -> object:
+        return object()
 
     def extract_response_text(self, parsed_message: object) -> str:
         if isinstance(parsed_message, dict):
@@ -281,13 +276,14 @@ def _validate_environment(
             "freesolo/environment.py with load_environment(...) returning a "
             "subclass of EnvironmentSingleTurn or EnvironmentMultiTurn."
         )
-    grpo_config = environment.get_grpo_config()
-    if not isinstance(grpo_config, GrpoConfig):
+    try:
+        environment.get_grpo_config()
+    except Exception as exc:
         source = f" from {reference!r}" if reference else ""
         raise TypeError(
-            f"Environment{source} get_grpo_config() must return "
-            "freesolo.training.GrpoConfig"
-        )
+            f"Environment{source} get_grpo_config() must be callable and return"
+            " a configuration object"
+        ) from exc
     return environment
 
 

{freesolo-0.2.45 → freesolo-0.2.47}/pypi/freesolo/environments/types.py

@@ -1,10 +1,9 @@
 from __future__ import annotations
 
 from dataclasses import dataclass, field
-from typing import Literal
+from typing import Literal, TypeAlias
 
-from ..contracts.markdown import ChatMessage
-from ..utils.core import JsonValue, MetadataDict
+from ._compat import ChatMessage, JsonValue, MetadataDict
 
 EnvironmentMetadata = MetadataDict
 RewardReturnType = Literal["binary", "numeric"]
@@ -34,6 +33,19 @@ class EnvironmentTurn:
         return payload
 
 
+@dataclass(slots=True)
+class EnvironmentGeneration:
+    """Model output returned by an environment callback used during local eval."""
+
+    response_text: str
+    latency_ms: int | None = None
+    total_tokens: int | None = None
+    metadata: EnvironmentMetadata = field(default_factory=dict)
+
+
+GenerationValue: TypeAlias = str | EnvironmentGeneration
+
+
 @dataclass(frozen=True)
 class EnvironmentStepResult:
     """Environment response after one assistant action in an episode."""
@@ -46,7 +58,7 @@ class EnvironmentStepResult:
 
 @dataclass(frozen=True)
 class EnvironmentEpisode:
-    """Completed trajectory used by evals, GEPA, tracing, and GRPO.
+    """Completed trajectory used by evals and environment-driven workflows.
 
     A single-turn episode usually has one assistant response. A multi-turn
     episode includes the full observable transcript as ``turns``.
@@ -130,8 +142,10 @@ class RewardResult:
 
 __all__ = [
     "EnvironmentEpisode",
+    "EnvironmentGeneration",
     "EnvironmentStepResult",
     "EnvironmentTurn",
+    "GenerationValue",
     "RewardMetric",
     "RewardResult",
 ]

freesolo-0.2.47/pyproject.toml

@@ -0,0 +1,65 @@
+[build-system]
+requires = ["hatchling>=1.25.0"]
+build-backend = "hatchling.build"
+
+[project]
+name = "freesolo"
+version = "0.2.47"
+description = "Environment and dataset helpers for Freesolo-generated repos."
+readme = "README.md"
+requires-python = ">=3.10"
+dependencies = [
+  "typing-extensions>=4.8.0",
+]
+
+[project.optional-dependencies]
+bson = [
+  "pymongo>=4.0.0",
+]
+
+[dependency-groups]
+dev = [
+  "gepa>=0.1.1; python_version >= '3.10'",
+  "httpx>=0.27.0",
+  "jsonschema>=4.0.0",
+  "numpy>=1.26.0; python_version >= '3.11'",
+  "opentelemetry-api>=1.28.0",
+  "opentelemetry-exporter-otlp-proto-http>=1.28.0",
+  "opentelemetry-sdk>=1.28.0",
+  "mypy>=1.13.0",
+  "pymongo>=4.0.0",
+  "pytest>=8.0.0",
+  "python-dotenv>=1.0.0",
+  "ruff>=0.11.0",
+  "openai>=1.0.0",
+  "verifiers>=0.1.14",
+  "wandb>=0.17.0; python_version >= '3.10'",
+]
+
+[tool.hatch.build.targets.wheel]
+packages = ["pypi/freesolo"]
+exclude = [
+  "pypi/freesolo/evaluation/**",
+  "pypi/freesolo/contracts/**",
+  "pypi/freesolo/gepa/**",
+  "pypi/freesolo/environments/evaluation.py",
+  "pypi/freesolo/tracing/**",
+  "pypi/freesolo/training/**",
+  "pypi/freesolo/utils/**",
+]
+
+[tool.hatch.build.targets.sdist]
+exclude = [
+  "pypi/freesolo/evaluation/**",
+  "pypi/freesolo/contracts/**",
+  "pypi/freesolo/gepa/**",
+  "pypi/freesolo/environments/evaluation.py",
+  "pypi/freesolo/tracing/**",
+  "pypi/freesolo/training/**",
+  "pypi/freesolo/utils/**",
+]
+
+[tool.mypy]
+python_version = "3.10"
+strict = false
+warn_unused_configs = true