freesolo-flash-dev 0.2.32__tar.gz → 0.2.33__tar.gz

{freesolo_flash_dev-0.2.32 → freesolo_flash_dev-0.2.33}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: freesolo-flash-dev
-Version: 0.2.32
+Version: 0.2.33
 Summary: Flash — managed LoRA post-training (SFT/GRPO) 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

freesolo_flash_dev-0.2.33/flash/serve/pricing.py

@@ -0,0 +1,94 @@
+"""Serving token prices for Flash deployments (per 1M tokens)."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+from flash.catalog import MODELS
+
+SERVING_MARKUP = 1.20
+
+
+@dataclass(frozen=True)
+class ServingPrice:
+    model_id: str
+    typical_input_usd_per_mtok: float
+    typical_output_usd_per_mtok: float
+    typical_cached_input_usd_per_mtok: float
+
+    @property
+    def billed_input_usd_per_mtok(self) -> float:
+        return self.typical_input_usd_per_mtok * SERVING_MARKUP
+
+    @property
+    def billed_output_usd_per_mtok(self) -> float:
+        return self.typical_output_usd_per_mtok * SERVING_MARKUP
+
+    @property
+    def billed_cached_input_usd_per_mtok(self) -> float:
+        return self.typical_cached_input_usd_per_mtok * SERVING_MARKUP
+
+
+SERVING_PRICES: dict[str, ServingPrice] = {
+    "openbmb/MiniCPM5-1B": ServingPrice(
+        model_id="openbmb/MiniCPM5-1B",
+        typical_input_usd_per_mtok=0.01,
+        typical_output_usd_per_mtok=0.05,
+        typical_cached_input_usd_per_mtok=0.002,
+    ),
+    "Qwen/Qwen3.5-0.8B": ServingPrice(
+        model_id="Qwen/Qwen3.5-0.8B",
+        typical_input_usd_per_mtok=0.01,
+        typical_output_usd_per_mtok=0.05,
+        typical_cached_input_usd_per_mtok=0.002,
+    ),
+    "Qwen/Qwen3.5-2B": ServingPrice(
+        model_id="Qwen/Qwen3.5-2B",
+        typical_input_usd_per_mtok=0.02,
+        typical_output_usd_per_mtok=0.10,
+        typical_cached_input_usd_per_mtok=0.004,
+    ),
+    "Qwen/Qwen3.5-4B": ServingPrice(
+        model_id="Qwen/Qwen3.5-4B",
+        typical_input_usd_per_mtok=0.03,
+        typical_output_usd_per_mtok=0.15,
+        typical_cached_input_usd_per_mtok=0.006,
+    ),
+    "Qwen/Qwen3.5-9B": ServingPrice(
+        model_id="Qwen/Qwen3.5-9B",
+        typical_input_usd_per_mtok=0.10,
+        typical_output_usd_per_mtok=0.15,
+        typical_cached_input_usd_per_mtok=0.020,
+    ),
+    "Qwen/Qwen3.6-35B-A3B": ServingPrice(
+        model_id="Qwen/Qwen3.6-35B-A3B",
+        typical_input_usd_per_mtok=0.15,
+        typical_output_usd_per_mtok=1.00,
+        typical_cached_input_usd_per_mtok=0.050,
+    ),
+}
+
+
+def serving_price(model_id: str) -> ServingPrice:
+    try:
+        return SERVING_PRICES[model_id]
+    except KeyError as exc:
+        raise ValueError(f"unknown serving model {model_id!r}") from exc
+
+
+def serving_price_rows() -> list[dict[str, float | str]]:
+    rows: list[dict[str, float | str]] = []
+    for model_id in sorted(MODELS):
+        price = serving_price(model_id)
+        rows.append(
+            {
+                "model_id": model_id,
+                "typical_input_usd_per_mtok": price.typical_input_usd_per_mtok,
+                "typical_output_usd_per_mtok": price.typical_output_usd_per_mtok,
+                "typical_cached_input_usd_per_mtok": price.typical_cached_input_usd_per_mtok,
+                "billed_input_usd_per_mtok": price.billed_input_usd_per_mtok,
+                "billed_output_usd_per_mtok": price.billed_output_usd_per_mtok,
+                "billed_cached_input_usd_per_mtok": price.billed_cached_input_usd_per_mtok,
+            }
+        )
+    return rows

{freesolo_flash_dev-0.2.32 → freesolo_flash_dev-0.2.33}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "freesolo-flash-dev"
-version = "0.2.32"
+version = "0.2.33"
 description = "Flash — managed LoRA post-training (SFT/GRPO) for Freesolo environments, driven by the `flash` CLI"
 readme = "README.md"
 # RunPod Flash supports Python 3.11-3.12 (not 3.13 yet).
@@ -111,7 +111,7 @@ 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 = "0.2.32"
+version = "0.2.33"
 
 [tool.pytest.ini_options]
 pythonpath = ["."]

{freesolo_flash_dev-0.2.32 → freesolo_flash_dev-0.2.33}/tests/test_serving_contract.py

@@ -141,7 +141,7 @@ def test_deployment_roundtrip_dict():
     assert "mode" not in data
 
 
-def test_serving_prices_cover_catalog_and_apply_markup():
+def test_serving_prices_cover_catalog():
     from flash.catalog import MODELS
     from flash.serve.pricing import SERVING_MARKUP, SERVING_PRICES, serving_price_rows
 
@@ -150,11 +150,44 @@ def test_serving_prices_cover_catalog_and_apply_markup():
     rows = serving_price_rows()
     assert len(rows) == len(MODELS)
     for row in rows:
+        assert row["typical_input_usd_per_mtok"] > 0
+        assert row["typical_output_usd_per_mtok"] > 0
+        assert row["typical_cached_input_usd_per_mtok"] > 0
+        assert row["billed_input_usd_per_mtok"] > 0
+        assert row["billed_output_usd_per_mtok"] > 0
+        assert row["billed_cached_input_usd_per_mtok"] > 0
+        assert row["billed_cached_input_usd_per_mtok"] < row["billed_input_usd_per_mtok"]
         assert row["billed_input_usd_per_mtok"] == pytest.approx(
-            row["base_input_usd_per_mtok"] * SERVING_MARKUP
+            row["typical_input_usd_per_mtok"] * SERVING_MARKUP
         )
         assert row["billed_output_usd_per_mtok"] == pytest.approx(
-            row["base_output_usd_per_mtok"] * SERVING_MARKUP
+            row["typical_output_usd_per_mtok"] * SERVING_MARKUP
+        )
+        assert row["billed_cached_input_usd_per_mtok"] == pytest.approx(
+            row["typical_cached_input_usd_per_mtok"] * SERVING_MARKUP
+        )
+
+
+def test_serving_prices_pin_public_rates_plus_markup():
+    from flash.serve.pricing import SERVING_MARKUP, SERVING_PRICES
+
+    typical = {
+        "openbmb/MiniCPM5-1B": (0.01, 0.05, 0.002),
+        "Qwen/Qwen3.5-0.8B": (0.01, 0.05, 0.002),
+        "Qwen/Qwen3.5-2B": (0.02, 0.10, 0.004),
+        "Qwen/Qwen3.5-4B": (0.03, 0.15, 0.006),
+        "Qwen/Qwen3.5-9B": (0.10, 0.15, 0.020),
+        "Qwen/Qwen3.6-35B-A3B": (0.15, 1.00, 0.050),
+    }
+    for model_id, (input_rate, output_rate, cached_rate) in typical.items():
+        price = SERVING_PRICES[model_id]
+        assert price.typical_input_usd_per_mtok == pytest.approx(input_rate)
+        assert price.typical_output_usd_per_mtok == pytest.approx(output_rate)
+        assert price.typical_cached_input_usd_per_mtok == pytest.approx(cached_rate)
+        assert price.billed_input_usd_per_mtok == pytest.approx(input_rate * SERVING_MARKUP)
+        assert price.billed_output_usd_per_mtok == pytest.approx(output_rate * SERVING_MARKUP)
+        assert price.billed_cached_input_usd_per_mtok == pytest.approx(
+            cached_rate * SERVING_MARKUP
         )
 
 

{freesolo_flash_dev-0.2.32 → freesolo_flash_dev-0.2.33}/uv.lock

@@ -1024,7 +1024,7 @@ wheels = [
 
 [[package]]
 name = "freesolo-flash"
-version = "0.2.32"
+version = "0.2.33"
 source = { editable = "." }
 
 [package.optional-dependencies]

freesolo_flash_dev-0.2.32/flash/serve/pricing.py

@@ -1,58 +0,0 @@
-"""Serving token prices for Flash deployments (per 1M tokens)."""
-
-from __future__ import annotations
-
-from dataclasses import dataclass
-
-from flash.catalog import MODELS
-
-SERVING_MARKUP = 1.20
-
-
-@dataclass(frozen=True)
-class ServingPrice:
-    model_id: str
-    base_input_usd_per_mtok: float
-    base_output_usd_per_mtok: float
-
-    @property
-    def billed_input_usd_per_mtok(self) -> float:
-        return self.base_input_usd_per_mtok * SERVING_MARKUP
-
-    @property
-    def billed_output_usd_per_mtok(self) -> float:
-        return self.base_output_usd_per_mtok * SERVING_MARKUP
-
-
-SERVING_PRICES: dict[str, ServingPrice] = {
-    "openbmb/MiniCPM5-1B": ServingPrice("openbmb/MiniCPM5-1B", 0.05, 0.10),
-    "Qwen/Qwen3.5-0.8B": ServingPrice("Qwen/Qwen3.5-0.8B", 0.05, 0.10),
-    "Qwen/Qwen3.5-2B": ServingPrice("Qwen/Qwen3.5-2B", 0.10, 0.20),
-    "Qwen/Qwen3.5-4B": ServingPrice("Qwen/Qwen3.5-4B", 0.20, 0.40),
-    "Qwen/Qwen3.5-9B": ServingPrice("Qwen/Qwen3.5-9B", 0.50, 1.00),
-    # MoE but serves on B200 → priced ~4x the 9B.
-    "Qwen/Qwen3.6-35B-A3B": ServingPrice("Qwen/Qwen3.6-35B-A3B", 2.00, 4.00),
-}
-
-
-def serving_price(model_id: str) -> ServingPrice:
-    try:
-        return SERVING_PRICES[model_id]
-    except KeyError as exc:
-        raise ValueError(f"unknown serving model {model_id!r}") from exc
-
-
-def serving_price_rows() -> list[dict[str, float | str]]:
-    rows: list[dict[str, float | str]] = []
-    for model_id in sorted(MODELS):
-        price = serving_price(model_id)
-        rows.append(
-            {
-                "model_id": model_id,
-                "base_input_usd_per_mtok": price.base_input_usd_per_mtok,
-                "base_output_usd_per_mtok": price.base_output_usd_per_mtok,
-                "billed_input_usd_per_mtok": price.billed_input_usd_per_mtok,
-                "billed_output_usd_per_mtok": price.billed_output_usd_per_mtok,
-            }
-        )
-    return rows