freesolo-flash-dev 1.2.112__tar.gz → 1.2.113__tar.gz

{freesolo_flash_dev-1.2.112 → freesolo_flash_dev-1.2.113}/Dockerfile.worker

@@ -64,7 +64,7 @@ RUN pip install --no-cache-dir \
     "tilelang==0.1.9" \
     "apache-tvm-ffi==0.1.9" \
     "hf_transfer==0.1.9" \
-    "freesolo==0.4.1" \
+    "freesolo==0.4.2" \
     "runpod==1.12.0"
 # Gated-DeltaNet (Qwen3.5/3.6) runs on flash-linear-attention's Triton kernels. On consumer cards
 # (Ampere/Ada/Blackwell) the Triton path is correct. On Hopper (sm90) fla's GDN chunk_bwd is

{freesolo_flash_dev-1.2.112 → freesolo_flash_dev-1.2.113}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: freesolo-flash-dev
-Version: 1.2.112
+Version: 1.2.113
 Summary: Flash — managed LoRA post-training (SFT/GRPO/OPD) for Freesolo environments, driven by the `flash` CLI
 Project-URL: Homepage, https://github.com/freesolo-co/flash
 Project-URL: Repository, https://github.com/freesolo-co/flash
@@ -19,7 +19,7 @@ Requires-Python: <3.13,>=3.11
 Provides-Extra: dev
 Requires-Dist: datasets>=2.19; extra == 'dev'
 Requires-Dist: fastapi<1,>=0.136; extra == 'dev'
-Requires-Dist: freesolo>=0.4.1; extra == 'dev'
+Requires-Dist: freesolo>=0.4.2; extra == 'dev'
 Requires-Dist: httpx>=0.27; extra == 'dev'
 Requires-Dist: huggingface-hub>=1.2.0; extra == 'dev'
 Requires-Dist: jsonschema>=4.23; extra == 'dev'
@@ -39,7 +39,7 @@ Provides-Extra: gpu
 Requires-Dist: accelerate>=1.4; extra == 'gpu'
 Requires-Dist: bitsandbytes>=0.49; extra == 'gpu'
 Requires-Dist: datasets>=2.19; extra == 'gpu'
-Requires-Dist: freesolo>=0.4.1; extra == 'gpu'
+Requires-Dist: freesolo>=0.4.2; extra == 'gpu'
 Requires-Dist: huggingface-hub>=1.2.0; extra == 'gpu'
 Requires-Dist: peft>=0.19; extra == 'gpu'
 Requires-Dist: tiktoken>=0.12; extra == 'gpu'
@@ -59,7 +59,7 @@ Requires-Dist: vllm==0.23.0; extra == 'serve-runtime'
 Provides-Extra: server
 Requires-Dist: datasets>=2.19; extra == 'server'
 Requires-Dist: fastapi<1,>=0.136; extra == 'server'
-Requires-Dist: freesolo>=0.4.1; extra == 'server'
+Requires-Dist: freesolo>=0.4.2; extra == 'server'
 Requires-Dist: httpx>=0.27; extra == 'server'
 Requires-Dist: huggingface-hub>=1.2.0; extra == 'server'
 Requires-Dist: jsonschema>=4.23; extra == 'server'

{freesolo_flash_dev-1.2.112 → freesolo_flash_dev-1.2.113}/flash/envs/base.py

@@ -180,7 +180,7 @@ def with_system_prompt(messages: list[dict], contract_text: str) -> list[dict]:
     return [{"role": "system", "content": system_text}, *out]
 
 
-FREESOLO_WORKER_SPEC = "freesolo>=0.4.1"
+FREESOLO_WORKER_SPEC = "freesolo>=0.4.2"
 
 
 def worker_pip_for_env(env_id: str) -> list[str]:

{freesolo_flash_dev-1.2.112 → freesolo_flash_dev-1.2.113}/flash/providers/artifacts/weight_cache.py

@@ -24,9 +24,30 @@ from flash.providers.artifacts.hf import make_hf_text_reader
 
 # imported at the top rather than re-exported at the bottom: these are default argument
 # values below, and a default evaluates when the `def` runs, long before a bottom import.
-from flash.providers.artifacts.preload_runpod import (
+# the rest of the runpod preload surface is re-exported here for the same reason, one step
+# further along: `main()` runs from the `__main__` guard, which sits ABOVE where these used to be
+# imported, so every CLI invocation died on `NameError: catalog_model_ids` while the library path
+# stayed green (importing the module executes the whole file, bottom import included). tests that
+# patch `weight_cache.<name>` keep working because the names still land in this module's namespace.
+from flash.providers.artifacts.preload_runpod import (  # noqa: F401
+    _NO_CAPACITY_GRACE_S,
     _PRELOAD_GPU,
     _PRELOAD_TIMEOUT_S,
+    _QUEUED,
+    _THROTTLED_GRACE_S,
+    _UNHEALTHY_GRACE_S,
+    NoCapacityError,
+    _any_worker,
+    _has_worker,
+    _only_unhealthy_workers,
+    _poll_until_done,
+    _preload_one_dc,
+    _throttled_workers,
+    _worker_counts,
+    catalog_model_ids,
+    teardown_lambda_filesystems,
+    teardown_weight_cache,
+    warm_weight_cache,
 )
 from flash.providers.base import UnreconciledCreateError
 from flash.providers.runpod import api as runpod_api  # noqa: F401
@@ -839,25 +860,3 @@ def main(argv: list[str] | None = None) -> int:
 
 if __name__ == "__main__":
     raise SystemExit(main())
-
-
-# re-exported so the runpod preload surface stays reachable as `weight_cache.<name>`: the
-# preload tests patch and call all of these here, and `main` below dispatches to them.
-from flash.providers.artifacts.preload_runpod import (  # noqa: E402,F401
-    _NO_CAPACITY_GRACE_S,
-    _QUEUED,
-    _THROTTLED_GRACE_S,
-    _UNHEALTHY_GRACE_S,
-    NoCapacityError,
-    _any_worker,
-    _has_worker,
-    _only_unhealthy_workers,
-    _poll_until_done,
-    _preload_one_dc,
-    _throttled_workers,
-    _worker_counts,
-    catalog_model_ids,
-    teardown_lambda_filesystems,
-    teardown_weight_cache,
-    warm_weight_cache,
-)

{freesolo_flash_dev-1.2.112 → freesolo_flash_dev-1.2.113}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "freesolo-flash-dev"
-version = "1.2.112"
+version = "1.2.113"
 description ="Flash — managed LoRA post-training (SFT/GRPO/OPD) for Freesolo environments, driven by the `flash` CLI"
 readme = "README.md"
 # RunPod Flash supports Python 3.11-3.12 (not 3.13 yet).
@@ -40,7 +40,7 @@ gpu = [
     "uv>=0.5",
     # Worker runtime: Freesolo environment loading and HF checkpoint I/O.
     "datasets>=2.19",
-    "freesolo>=0.4.1",
+    "freesolo>=0.4.2",
     "huggingface-hub>=1.2.0",
 ]
 # Control-plane / operator deps. runpod-flash (RunPod provider SDK) + huggingface-hub (provider
@@ -68,7 +68,7 @@ server = [
     "regex>=2024.11.6",
     "huggingface-hub>=1.2.0",
     "datasets>=2.19",
-    "freesolo>=0.4.1",
+    "freesolo>=0.4.2",
     "transformers>=5.6,<5.13",
     # the control plane decodes images: flash/server/routes/runs.py imports
     # preflight_validate_image_opd at module load, and flash/content/multimodal.py opens every image
@@ -148,7 +148,7 @@ dev = [
     "regex>=2024.11.6",
     "huggingface-hub>=1.2.0",
     "datasets>=2.19",
-    "freesolo>=0.4.1",
+    "freesolo>=0.4.2",
     "transformers>=5.6,<5.13",
     "tokenizers>=0.22",
     "tiktoken>=0.12",
@@ -195,7 +195,7 @@ artifacts = ["flash/py.typed"]
 # auto-published by .github/workflows/publish-dev.yml: a push to `dev` whose version isn't yet on
 # PyPI publishes it; pushes that don't change it find it already published and no-op. So cutting a
 # dev-channel release = bump this (and [project].version) and merge to `dev`.
-version = "1.2.112"
+version = "1.2.113"
 
 [tool.pytest.ini_options]
 pythonpath = ["."]
@@ -235,7 +235,7 @@ dev = [
     # Offline worker/env tests import these without the heavy `gpu` stack; the CI test env
     # (`--extra server --dev`) gets httpx + huggingface-hub from the `server` extra.
     "datasets>=2.19",
-    "freesolo>=0.4.1",
+    "freesolo>=0.4.2",
     "tokenizers>=0.22",
     "tiktoken>=0.12",
     # flash/content/multimodal.py imports PIL directly to validate and decode every image source, so the

{freesolo_flash_dev-1.2.112 → freesolo_flash_dev-1.2.113}/tests/test_cli_commands.py

@@ -3000,7 +3000,7 @@ def test_submit_payload_carries_authored_pip_and_the_worker_appends_it(
     assert tuple(spec_payload(spec)["environment"]["pip"]) == ("pymongo>=4.6",)
     # and the submit paths install it after the worker requirement, not instead of it.
     assert worker_pip_with_extras(spec.environment.id, spec.environment.pip) == [
-        "freesolo>=0.4.1",
+        "freesolo>=0.4.2",
         "pymongo>=4.6",
     ]
 

{freesolo_flash_dev-1.2.112 → freesolo_flash_dev-1.2.113}/tests/test_provider_routing.py

@@ -2477,13 +2477,13 @@ def test_submit_supplies_the_worker_pip_when_the_author_declared_none() -> None:
         deadline_at=1_800_000_000.0,
     )
 
-    assert payload["extra_pip"] == ["freesolo>=0.4.1"]
+    assert payload["extra_pip"] == ["freesolo>=0.4.2"]
 
 
 def test_submit_appends_authored_pip_without_displacing_the_worker_spec() -> None:
     """The author's scorer deps are additional to the worker requirement, never a replacement.
 
-    Substituting instead of appending would drop ``freesolo>=0.4.1`` the moment anyone declared a
+    Substituting instead of appending would drop ``freesolo>=0.4.2`` the moment anyone declared a
     dependency, breaking the worker outright for exactly the users the knob exists to serve.
     """
     from dataclasses import replace
@@ -2502,19 +2502,19 @@ def test_submit_appends_authored_pip_without_displacing_the_worker_spec() -> Non
         deadline_at=1_800_000_000.0,
     )
 
-    assert payload["extra_pip"] == ["freesolo>=0.4.1", "pymongo>=4.6", "rapidfuzz"]
+    assert payload["extra_pip"] == ["freesolo>=0.4.2", "pymongo>=4.6", "rapidfuzz"]
 
 
 def test_worker_pip_with_extras_dedupes_and_ignores_blanks() -> None:
     """Restating the worker spec must not install it twice, and blanks must not reach pip."""
     from flash.envs.base import worker_pip_with_extras
 
-    assert worker_pip_with_extras("e", ["freesolo>=0.4.1", "pymongo"]) == [
-        "freesolo>=0.4.1",
+    assert worker_pip_with_extras("e", ["freesolo>=0.4.2", "pymongo"]) == [
+        "freesolo>=0.4.2",
         "pymongo",
     ]
-    assert worker_pip_with_extras("e", ["  pymongo  ", ""]) == ["freesolo>=0.4.1", "pymongo"]
-    assert worker_pip_with_extras("e", None) == ["freesolo>=0.4.1"]
+    assert worker_pip_with_extras("e", ["  pymongo  ", ""]) == ["freesolo>=0.4.2", "pymongo"]
+    assert worker_pip_with_extras("e", None) == ["freesolo>=0.4.2"]
 
 
 def _submit_failure(provider_obj):

{freesolo_flash_dev-1.2.112 → freesolo_flash_dev-1.2.113}/tests/test_verifiers.py

@@ -2122,7 +2122,7 @@ def test_worker_deps():
     import flash.envs.base as registry
 
     env_id = "github:owner/repo@main:env/environment.py"
-    assert registry.worker_pip_for_env(env_id) == ["freesolo>=0.4.1"]
+    assert registry.worker_pip_for_env(env_id) == ["freesolo>=0.4.2"]
 
 
 # ============================================================================================

{freesolo_flash_dev-1.2.112 → freesolo_flash_dev-1.2.113}/tests/test_weight_cache.py

@@ -60,6 +60,34 @@ def _ndc() -> int:
     return n
 
 
+def test_preload_cli_entrypoint_runs_as_a_subprocess():
+    """`python -m ...weight_cache --dry-run` must reach the planner, not die on a NameError.
+
+    MUST be a subprocess: the names `main()` needs were re-exported at the BOTTOM of the module,
+    below the `__main__` guard, so `python -m` ran `main()` before the import and every CLI
+    invocation died on `NameError: catalog_model_ids`. An in-process import cannot reproduce that --
+    importing executes the whole file, bottom import included, so the library path stayed green
+    while the only operator-facing entrypoint was dead. --dry-run provisions nothing.
+    """
+    import subprocess
+    import sys
+
+    proc = subprocess.run(
+        [sys.executable, "-m", "flash.providers.artifacts.weight_cache", "--dry-run"],
+        capture_output=True,
+        text=True,
+        timeout=300,
+    )
+    combined = proc.stdout + proc.stderr
+    assert "NameError" not in combined, f"CLI died on a NameError:\n{combined}"
+    assert proc.returncode == 0, f"--dry-run exited {proc.returncode}:\n{combined}"
+    # prove it reached the RUNPOD planner specifically. a bare "would warm" also matches the Lambda
+    # plan line, which prints from a different branch and would not exercise `catalog_model_ids`.
+    assert "datacenter(s):" in proc.stdout, f"runpod planner never ran:\n{combined}"
+    # the model list is what `catalog_model_ids` produces -- the name that used to NameError.
+    assert "model(s):" in proc.stdout, f"model catalog never resolved:\n{combined}"
+
+
 # ---------------------------------------------------------------------------
 # spec carrier round-trips (network_volume is platform-managed: it must survive every
 # to_internal_dict()->from_dict() hop, the control-plane/worker carrier. it is intentionally

{freesolo_flash_dev-1.2.112 → freesolo_flash_dev-1.2.113}/uv.lock

@@ -1466,20 +1466,21 @@ wheels = [
 
 [[package]]
 name = "freesolo"
-version = "0.4.1"
+version = "0.4.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
+    { name = "httpx" },
     { name = "openai" },
     { name = "typing-extensions" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/f4/59/4fbc3cb6813ae2ebec51e35cf7128c4d34b2a07e34b0ec7c85053ad850e7/freesolo-0.4.1.tar.gz", hash = "sha256:7e59492a5bf30b01ac7124b2164addde1b65f14b30f87f346da6c089930b6dff", size = 331064, upload-time = "2026-08-12T10:06:26.621Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/9b/f7/acb72339f2c5553248c7ac7f191a7a9e3a82999ab4937c3a9d2b87547ca9/freesolo-0.4.2.tar.gz", hash = "sha256:ce4f1f31fb65d1fd017ed76e1f859ede0fcb0b99d8b243b56b20314b438cfe40", size = 35625, upload-time = "2026-08-24T14:27:27.99Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/2a/eb/468f33119b74ebc249ecf528dcfc01e3a579362ef7cb58ca929567dc0982/freesolo-0.4.1-py3-none-any.whl", hash = "sha256:65db09befc335c8004ba4c542a7040fe643e1d1aaf765e2bb9c02b22fa38c893", size = 45388, upload-time = "2026-08-12T10:06:25.364Z" },
+    { url = "https://files.pythonhosted.org/packages/e2/05/164fcb1f8581e207d819273c3723db78d3f4c5e413fb24447c9b1e39783d/freesolo-0.4.2-py3-none-any.whl", hash = "sha256:fb2b9209cedfbf9c4f51418e74c2b5775fce5f0505cd9ad36ae30e97a9b274ee", size = 46971, upload-time = "2026-08-24T14:27:26.872Z" },
 ]
 
 [[package]]
 name = "freesolo-flash"
-version = "1.2.112"
+version = "1.2.113"
 source = { editable = "." }
 
 [package.optional-dependencies]
@@ -1581,9 +1582,9 @@ requires-dist = [
     { name = "fastapi", marker = "extra == 'serve-runtime'", specifier = ">=0.136,<0.137" },
     { name = "fastapi", marker = "extra == 'server'", specifier = ">=0.136,<1" },
     { name = "fastapi", marker = "extra == 'serving'", specifier = ">=0.136,<1" },
-    { name = "freesolo", marker = "extra == 'dev'", specifier = ">=0.4.1" },
-    { name = "freesolo", marker = "extra == 'gpu'", specifier = ">=0.4.1" },
-    { name = "freesolo", marker = "extra == 'server'", specifier = ">=0.4.1" },
+    { name = "freesolo", marker = "extra == 'dev'", specifier = ">=0.4.2" },
+    { name = "freesolo", marker = "extra == 'gpu'", specifier = ">=0.4.2" },
+    { name = "freesolo", marker = "extra == 'server'", specifier = ">=0.4.2" },
     { name = "httpx", marker = "extra == 'dev'", specifier = ">=0.27" },
     { name = "httpx", marker = "extra == 'server'", specifier = ">=0.27" },
     { name = "httpx", marker = "extra == 'serving'", specifier = ">=0.27.0" },
@@ -1636,7 +1637,7 @@ provides-extras = ["gpu", "server", "serving", "serve-runtime", "dev"]
 [package.metadata.requires-dev]
 dev = [
     { name = "datasets", specifier = ">=2.19" },
-    { name = "freesolo", specifier = ">=0.4.1" },
+    { name = "freesolo", specifier = ">=0.4.2" },
     { name = "mypy", specifier = ">=1.13.0" },
     { name = "packaging", specifier = ">=24,<27" },
     { name = "pillow", specifier = ">=11" },