Adds initial CI suport

.github/workflows/unit_test.yaml

@@ -0,0 +1,36 @@
+name: Unit Test
+
+on:
+  pull_request:
+
+
+jobs:
+  unit_tests:
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    strategy:
+      matrix:
+        python-version: ['3.10', '3.11', '3.12']
+    steps:
+      - name: Check out repo
+        uses: actions/checkout@v4
+      - name: Setup conda env
+        uses: conda-incubator/setup-miniconda@v2
+        with:
+          auto-update-conda: true
+          miniconda-version: "latest"
+          activate-environment: test
+          python-version: ${{ matrix.python-version }}
+      - name: Update pip
+        run: python -m pip install --upgrade pip
+      - name: Install dependencies
+        run: |
+          python -m pip install -e ".[dev]"
+          # Will have to pin until monarch wheel is reasonable
+          pip install torch==2.9.0.dev20250815+cpu --index-url https://download.pytorch.org/whl/nightly/cpu
+          python -m pip install --no-build-isolation --verbose assets/wheels/monarch_no_torch-0.1.0.dev20250815-py3-none-any.whl
+      - name: Run unit tests with coverage
+        # TODO add all tests
+        run: pytest tests/unit_tests --cov=. --cov-report=xml --durations=20 -vv
+      - name: Upload Coverage to Codecov
+        uses: codecov/codecov-action@v3

.gitignore

@@ -32,7 +32,6 @@ lib64/
 parts/
 sdist/
 var/
-wheels/
 share/python-wheels/
 *.egg-info/
 .installed.cfg

.pre-commit-config.yaml

@@ -12,8 +12,6 @@ repos:
     -   id: check-merge-conflict
     -   id: no-commit-to-branch
         args: ['--branch=main']
-    -   id: check-added-large-files
-        args: ['--maxkb=1000']
     -   id: end-of-file-fixer
         exclude: '^(.*\.svg)$'
 

apps/toy_rl/main.py

@@ -16,7 +16,6 @@
 import torch
 from forge.actors.collector import Collector
 
-from forge.controller.stack import stack
 from forge.data.replay_buffer import ReplayBuffer
 from forge.interfaces import Environment, Policy
 from forge.types import Action, Observation, State
@@ -150,7 +149,8 @@ async def main():
     # This policy just generates something between -2. and 2.
     policy = await policy_procs.spawn("policy", ToyPolicy, action_range=(-2.0, 2.0))
 
-    browser_collectors = await browser_procs.spawn(
+    # TODO - replace multiple collectors with a service.
+    collectors = await browser_procs.spawn(
         "browser",
         Collector,
         max_collector_steps=5,
@@ -162,29 +162,6 @@ async def main():
         # to do it this way.
         environment_creator=partial(ToyEnvironment, name="browser", max_steps=5),
     )
-    deep_research_collectors = await deep_research_procs.spawn(
-        "deep_research",
-        Collector,
-        max_collector_steps=5,
-        policy=policy,
-        replay_buffer=replay_buffer,
-        environment_creator=partial(ToyEnvironment, name="deep_research", max_steps=5),
-    )
-    coding_collectors = await coder_procs.spawn(
-        "coding",
-        Collector,
-        max_collector_steps=5,
-        policy=policy,
-        replay_buffer=replay_buffer,
-        environment_creator=partial(ToyEnvironment, name="coding", max_steps=5),
-    )
-
-    collectors = stack(
-        browser_collectors,
-        deep_research_collectors,
-        coding_collectors,
-        interface=Collector,
-    )
 
     # Create two async tasks
     async def episode_collector_task():
@@ -193,16 +170,7 @@ async def episode_collector_task():
         while True:
             try:
                 print(f"🎮 Running episode {episode_count + 1}...")
-
-                # call() is essentially our "map" - every collector runs their own
-                # episode loop.
-                # What's pretty elegant here is if we wanted to control off policiness, we could
-                # easily counter on steps and call policy.update_weights.call() at our desired
-                # frequency.
-                results = collectors.run_episode.call()
-
-                # Temporary hack due to Monarch changes - ideally you could just await results
-                results = [await r for r in results]
+                results = await collectors.run_episode.call()
                 num_trajectories = sum([len(r._values) for r in results])
                 episode_count += 1
                 print(

assets/version.txt

@@ -0,0 +1 @@
+0.1.0

pyproject.toml

@@ -13,6 +13,7 @@ dependencies = [
     # PyTorch
     "torchdata>=0.8.0",
     "torchtitan",
+    "torchao",
     # vLLM
     # TODO: pin specific vllm version
     #"vllm==0.10.0",
@@ -36,6 +37,7 @@ dev = [
     "pytest-cov",
     "pytest-timeout",
     "tensorboard",
+    "expecttest",
     "tomli>=1.1.0",
     "anyio",
     "pytest-asyncio",

src/forge/actors/__init__.py

@@ -5,6 +5,5 @@
 # LICENSE file in the root directory of this source tree.
 
 from .collector import Collector
-from .policy import Policy, PolicyRouter
 
-__all__ = ["Collector", "Policy", "PolicyRouter"]
+__all__ = ["Collector"]

src/forge/controller/__init__.py

@@ -8,7 +8,6 @@
 from .recoverable_mesh import RecoverableProcMesh
 from .service import AutoscalingConfig, Service, ServiceConfig
 from .spawn import spawn_service
-from .stack import stack
 
 __all__ = [
     "AutoscalingConfig",
@@ -19,5 +18,4 @@
     "get_proc_mesh",
     "ForgeActor",
     "RecoverableProcMesh",
-    "stack",
 ]