freesolo-flash-dev 1.2.131__tar.gz → 1.2.132__tar.gz

{freesolo_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/.github/workflows/auto-rebake.yml

@@ -232,7 +232,7 @@ jobs:
           sudo docker image prune -af || true
           df -h /
 
-      - uses: docker/setup-buildx-action@37fe631027851001ddb9b187196cc803df7f5f0e # v4.3.0
+      - uses: docker/setup-buildx-action@f87e5991a6d7451dcb8d9637bfbc97413f497069 # v4.4.1
 
       - name: Log in to GHCR
         uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f # v4.6.0
@@ -255,7 +255,7 @@ jobs:
           fi
 
       - name: Re-layer cache onto the fresh base (no GPU)
-        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7.3.0
+        uses: docker/build-push-action@c3c9e263c25d99ce0380d002d59b67737d91b0dc # v7.4.0
         with:
           context: .
           file: docker/Dockerfile.kernelcache.relayer

{freesolo_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/.github/workflows/bake-kernel-cache.yml

@@ -195,7 +195,7 @@ jobs:
 
       - name: Install uv
         if: steps.gate.outputs.run == 'true'
-        uses: astral-sh/setup-uv@20cfd1bf945f4377ade1205e4dbc17946fc9a30d # v10.0.1
+        uses: astral-sh/setup-uv@1c37ad07a6a961277cf70c0d37d6f313000f5884 # v10.3.0
         with:
           # pin the installed binary as well as the action implementation.
           version: "0.12.5"
@@ -237,7 +237,7 @@ jobs:
           sudo docker image prune -af || true
           df -h /
 
-      - uses: docker/setup-buildx-action@37fe631027851001ddb9b187196cc803df7f5f0e # v4.3.0
+      - uses: docker/setup-buildx-action@f87e5991a6d7451dcb8d9637bfbc97413f497069 # v4.4.1
         if: steps.gate.outputs.run == 'true'
 
       # thin layer FROM the SAME base the warmup ran inside -> the cache toolchain always matches the
@@ -245,7 +245,7 @@ jobs:
       # published base and bake a cache that silently won't load). GHCR auth from the login step above.
       - name: Build + push baked per-sm image (thin layer on the warmed base)
         if: steps.gate.outputs.run == 'true'
-        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7.3.0
+        uses: docker/build-push-action@c3c9e263c25d99ce0380d002d59b67737d91b0dc # v7.4.0
         with:
           context: .
           file: docker/Dockerfile.kernelcache

{freesolo_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/.github/workflows/ci.yml

@@ -33,7 +33,7 @@ jobs:
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Install uv
-        uses: astral-sh/setup-uv@20cfd1bf945f4377ade1205e4dbc17946fc9a30d # v10.0.1
+        uses: astral-sh/setup-uv@1c37ad07a6a961277cf70c0d37d6f313000f5884 # v10.3.0
         with:
           # pin the installed binary as well as the action implementation.
           version: "0.12.5"
@@ -163,7 +163,7 @@ jobs:
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Install uv
-        uses: astral-sh/setup-uv@20cfd1bf945f4377ade1205e4dbc17946fc9a30d # v10.0.1
+        uses: astral-sh/setup-uv@1c37ad07a6a961277cf70c0d37d6f313000f5884 # v10.3.0
         with:
           # pin the installed binary as well as the action implementation.
           version: "0.12.5"

{freesolo_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/.github/workflows/deploy-modal-dev.yml

@@ -98,7 +98,7 @@ jobs:
         with:
           python-version: "3.12"
 
-      - uses: astral-sh/setup-uv@20cfd1bf945f4377ade1205e4dbc17946fc9a30d # v10.0.1
+      - uses: astral-sh/setup-uv@1c37ad07a6a961277cf70c0d37d6f313000f5884 # v10.3.0
         if: steps.gate.outputs.deploy == 'true'
         with:
           # pin the installed binary as well as the action implementation.

{freesolo_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/.github/workflows/publish-dev.yml

@@ -136,7 +136,7 @@ jobs:
 
       - name: Install uv
         if: steps.decide.outputs.publish == 'true'
-        uses: astral-sh/setup-uv@20cfd1bf945f4377ade1205e4dbc17946fc9a30d # v10.0.1
+        uses: astral-sh/setup-uv@1c37ad07a6a961277cf70c0d37d6f313000f5884 # v10.3.0
         with:
           # pin the installed binary as well as the action implementation.
           version: "0.12.5"

{freesolo_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/.github/workflows/publish-image.yml

@@ -43,7 +43,7 @@ jobs:
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@37fe631027851001ddb9b187196cc803df7f5f0e # v4.3.0
+        uses: docker/setup-buildx-action@f87e5991a6d7451dcb8d9637bfbc97413f497069 # v4.4.1
 
       - name: Log in to GHCR
         uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f # v4.6.0
@@ -98,7 +98,7 @@ jobs:
       # published here still runs on a plain --env-file. The Dockerfile defaults the argument
       # to false, so `docker build .` from a checkout stays vendor-free.
       - name: Build and push
-        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7.3.0
+        uses: docker/build-push-action@c3c9e263c25d99ce0380d002d59b67737d91b0dc # v7.4.0
         with:
           context: .
           file: ./Dockerfile

{freesolo_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/.github/workflows/publish-serving-image.yml

@@ -150,7 +150,7 @@ jobs:
 
       - name: Set up Docker Buildx
         if: steps.gate.outputs.publish == 'true'
-        uses: docker/setup-buildx-action@37fe631027851001ddb9b187196cc803df7f5f0e # v4.3.0
+        uses: docker/setup-buildx-action@f87e5991a6d7451dcb8d9637bfbc97413f497069 # v4.4.1
 
       - name: Log in to GHCR
         if: steps.gate.outputs.publish == 'true'
@@ -184,7 +184,7 @@ jobs:
       - name: Build and push
         if: steps.gate.outputs.publish == 'true'
         id: build
-        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7.3.0
+        uses: docker/build-push-action@c3c9e263c25d99ce0380d002d59b67737d91b0dc # v7.4.0
         with:
           context: .
           file: ./Dockerfile.serve

{freesolo_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/.github/workflows/publish.yml

@@ -202,7 +202,7 @@ jobs:
 
       - name: Install uv
         if: github.event_name == 'workflow_dispatch' || steps.changes.outputs.package_changed == 'true'
-        uses: astral-sh/setup-uv@20cfd1bf945f4377ade1205e4dbc17946fc9a30d # v10.0.1
+        uses: astral-sh/setup-uv@1c37ad07a6a961277cf70c0d37d6f313000f5884 # v10.3.0
         with:
           # pin the installed binary as well as the action implementation.
           version: "0.12.5"

{freesolo_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/.github/workflows/worker-image.yml

@@ -129,7 +129,7 @@ jobs:
           sudo docker image prune -af || true
           df -h /
 
-      - uses: docker/setup-buildx-action@37fe631027851001ddb9b187196cc803df7f5f0e # v4.3.0
+      - uses: docker/setup-buildx-action@f87e5991a6d7451dcb8d9637bfbc97413f497069 # v4.4.1
 
       - name: Log in to GHCR
         uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f # v4.6.0
@@ -140,7 +140,7 @@ jobs:
 
       - name: Build + push worker image
         id: image
-        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7.3.0
+        uses: docker/build-push-action@c3c9e263c25d99ce0380d002d59b67737d91b0dc # v7.4.0
         with:
           context: .
           file: Dockerfile.worker

{freesolo_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: freesolo-flash-dev
-Version: 1.2.131
+Version: 1.2.132
 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
@@ -29,7 +29,7 @@ Requires-Dist: pillow>=11; extra == 'dev'
 Requires-Dist: pytest>=9.0.3; extra == 'dev'
 Requires-Dist: referencing>=0.28.4; extra == 'dev'
 Requires-Dist: regex>=2024.11.6; extra == 'dev'
-Requires-Dist: ruff<0.17,>=0.6; extra == 'dev'
+Requires-Dist: ruff<0.18,>=0.6; extra == 'dev'
 Requires-Dist: runpod-flash<2,>=1.19; extra == 'dev'
 Requires-Dist: tiktoken>=0.12; extra == 'dev'
 Requires-Dist: tokenizers>=0.22; extra == 'dev'
@@ -63,7 +63,7 @@ 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'
-Requires-Dist: modal==1.5.4; extra == 'server'
+Requires-Dist: modal==1.6.1; extra == 'server'
 Requires-Dist: pillow>=11; extra == 'server'
 Requires-Dist: referencing>=0.28.4; extra == 'server'
 Requires-Dist: regex>=2024.11.6; extra == 'server'
@@ -73,7 +73,7 @@ Requires-Dist: uvicorn<1,>=0.52; extra == 'server'
 Provides-Extra: serving
 Requires-Dist: fastapi<1,>=0.136; extra == 'serving'
 Requires-Dist: httpx>=0.27.0; extra == 'serving'
-Requires-Dist: modal==1.5.4; extra == 'serving'
+Requires-Dist: modal==1.6.1; extra == 'serving'
 Requires-Dist: orjson>=3.10.0; extra == 'serving'
 Requires-Dist: pillow>=11; extra == 'serving'
 Requires-Dist: pydantic-settings>=2.6.0; extra == 'serving'

{freesolo_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/flash/cli/ui/render.py

@@ -484,9 +484,9 @@ def _humanize_ts(value) -> str | None:
     """Format an epoch seconds value as a compact UTC timestamp, leaving non-numbers alone."""
     if not isinstance(value, (int, float)) or value <= 0:
         return None
-    import datetime
+    import datetime as dt
 
-    return datetime.datetime.fromtimestamp(value, datetime.UTC).strftime("%Y-%m-%d %H:%M UTC")
+    return dt.datetime.fromtimestamp(value, dt.UTC).strftime("%Y-%m-%d %H:%M UTC")
 
 
 def gpu_label(spec: dict, remote: dict) -> str:

{freesolo_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/flash/serve/runtime/sampling.py

@@ -121,6 +121,9 @@ def _normalize_token_position(
             selected = record
     if selected is None:
         raise RuntimeNotReadyError("vllm logprobs omitted the selected token")
+    # vllm may insert the sampled token before higher-probability alternatives.
+    # rank before truncating; token id makes equal-probability ties deterministic.
+    normalized.sort(key=lambda item: (-item[1]["logprob"], item[0]))
     return {
         **selected,
         "top_logprobs": [record for _, record in normalized[:top_logprobs]],

{freesolo_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "freesolo-flash-dev"
-version = "1.2.131"
+version = "1.2.132"
 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).
@@ -60,7 +60,7 @@ server = [
     "fastapi>=0.136,<1",
     "uvicorn>=0.52,<1",
     # customer-owned modal lifecycle mutations execute in the control plane.
-    "modal==1.5.4",
+    "modal==1.6.1",
     "runpod-flash>=1.19,<2",
     "httpx>=0.27",
     "jsonschema>=4.23",
@@ -98,7 +98,7 @@ serving = [
     # this one is what the serving image installs ALONE -- so a range here lets the container
     # resolve a different modal than the control plane was validated against, on a rebuild that
     # changed nothing else. one policy per package.
-    "modal==1.5.4",
+    "modal==1.6.1",
     "orjson>=3.10.0",
     # `src/multimodal.py` decodes and normalizes inbound images at module scope. the `gpu` extra
     # gets Pillow transitively via vllm, but this extra deliberately excludes vllm, so it has to
@@ -134,11 +134,11 @@ dev = [
     "pytest>=9.0.3",
     # dependency contract tests parse exact pins without relying on a transitive install.
     "packaging>=24,<27",
-    # Upper bound tracks uv.lock (ruff 0.16.x), which is what CI lints with. Without it, an
+    # Upper bound tracks uv.lock (ruff 0.17.x), which is what CI lints with. Without it, an
     # install that ignores the lockfile (`uv pip install -e '.[dev]'`, plain pip) resolves whatever
     # ruff is newest, whose added rules fail on a tree CI calls clean. Raise both sites together
     # with the lockfile, never ahead of it.
-    "ruff>=0.6,<0.17",
+    "ruff>=0.6,<0.18",
     # mirror server's 0.136 floor so dev remains compatible with the deployed serving extra pair.
     "fastapi>=0.136,<1",
     "uvicorn>=0.52,<1",
@@ -196,7 +196,7 @@ 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 = "1.2.131"
+version = "1.2.132"
 
 [tool.pytest.ini_options]
 pythonpath = ["."]
@@ -232,7 +232,7 @@ dev = [
     # dependency contract tests parse exact pins without relying on a transitive install.
     "packaging>=24,<27",
     # Keep in lockstep with the `dev` extra above and with uv.lock.
-    "ruff>=0.6,<0.17",
+    "ruff>=0.6,<0.18",
     # 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_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/tests/test_mirror_sampling_parity.py

@@ -320,12 +320,12 @@ def test_choice_normalization_rejects_malformed_complete_sets(outputs):
 @pytest.mark.parametrize(
     ("top_logprobs", "candidates", "expected_tokens"),
     [
-        (2, [(12, -1.2, "å"), (10, -0.1, "a"), (13, -2.0, "c")], ["å", "a"]),
-        (2, [(12, -1.2, "å"), (13, -2.0, "c"), (10, -0.1, "a")], ["å", "c"]),
+        (2, [(12, -1.2, "å"), (10, -0.1, "a"), (13, -2.0, "c")], ["a", "å"]),
+        (2, [(12, -1.2, "å"), (13, -2.0, "c"), (10, -0.1, "a")], ["a", "å"]),
         (0, [(10, -0.1, "a"), (12, -1.2, "å")], []),
     ],
 )
-def test_vllm_logprobs_match_openai_candidate_slicing(top_logprobs, candidates, expected_tokens):
+def test_vllm_logprobs_rank_candidates_before_truncating(top_logprobs, candidates, expected_tokens):
     def candidate(value, token):
         return SimpleNamespace(logprob=value, decoded_token=token)
 
@@ -675,3 +675,27 @@ def test_text_only_stream_rejects_wrong_control_types_before_transport(
         deploy.chat_stream(
             "run/final", [{"role": "user", "content": "hi"}], org_id="org-1", **kwargs
         )
+
+
+@pytest.mark.parametrize("top_count", [0, 1, 2, 3, 20])
+def test_logprob_ranking_is_independent_of_candidate_insertion_order(top_count):
+    from itertools import permutations
+
+    candidates = [(99, -9.0, "sampled"), (1, -0.1, "best"), (2, -0.1, "tied")]
+    for ordering in permutations(candidates):
+        result = normalize_token_logprobs(
+            [99],
+            [
+                {
+                    token_id: SimpleNamespace(logprob=value, decoded_token=token)
+                    for token_id, value, token in ordering
+                }
+            ],
+            top_logprobs=top_count,
+        )
+        assert result is not None
+        assert result[0]["token"] == "sampled"
+        assert result[0]["logprob"] == -9.0
+        assert [item["token"] for item in result[0]["top_logprobs"]] == ["best", "tied", "sampled"][
+            :top_count
+        ]

{freesolo_flash_dev-1.2.131 → freesolo_flash_dev-1.2.132}/tests/test_serve_runtime_engine.py

@@ -759,7 +759,7 @@ def test_packaged_engine_threads_top_logprobs_through_buffered_and_streaming() -
         runtime.generate(GenerationRequest(prompt="hello", logprobs=True, top_logprobs=1))
     )
     assert result.choices[0].logprobs is not None
-    assert [record["token"] for record in result.choices[0].logprobs[0]["top_logprobs"]] == ["b"]
+    assert [record["token"] for record in result.choices[0].logprobs[0]["top_logprobs"]] == ["a"]
 
     engine.responses.append([_output("a", [1], logprobs=[candidates])])