freesolo-flash 0.2.0__tar.gz → 0.2.2__tar.gz

{freesolo_flash-0.2.0 → freesolo_flash-0.2.2}/.env.example

@@ -23,7 +23,7 @@ PRIME_API_KEY=
 # flash's adapter registration calls to the freesolo serving app (below).
 FREESOLO_INTERNAL_KEY=
 # Where to verify user keys. In compose this is the backend on the internal network;
-# defaults to https://api.freesolo.co when unset.
+# defaults to https://api-dev.freesolo.co when unset.
 FREESOLO_BASE_URL=http://backend:8000
 
 # --- FreeSolo serving (adapter deploy/chat) ---

{freesolo_flash-0.2.0 → freesolo_flash-0.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: freesolo-flash
-Version: 0.2.0
+Version: 0.2.2
 Summary: Flash — managed LoRA post-training (SFT/GRPO) for verifiers 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-0.2.0 → freesolo_flash-0.2.2}/flash/__init__.py

@@ -8,4 +8,4 @@ GPU (RunPod or Vast.ai) behind the scenes.
 
 __all__ = ["__version__"]
 
-__version__ = "0.2.0"
+__version__ = "0.2.2"

{freesolo_flash-0.2.0 → freesolo_flash-0.2.2}/flash/cli/main/__init__.py

@@ -79,7 +79,7 @@ def main(argv: list[str] | None = None) -> int:
     login.add_argument(
         "--freesolo-url",
         dest="freesolo_url",
-        help="freesolo backend base URL (default: FREESOLO_BASE_URL or https://api.freesolo.co)",
+        help="freesolo backend base URL (default: FREESOLO_BASE_URL or https://api-dev.freesolo.co)",
     )
     login.add_argument(
         "--api-url", help="flash control-plane URL for training calls (default: FLASH_API_URL)"

{freesolo_flash-0.2.0 → freesolo_flash-0.2.2}/flash/client/http.py

@@ -29,7 +29,7 @@ class ApiError(ClientError):
 # Login is handled by the freesolo backend (not the flash control plane): `flash login`
 # verifies the user's freesolo API key here. The same key authenticates the flash
 # control plane, which accepts freesolo-issued keys.
-DEFAULT_FREESOLO_BASE_URL = "https://api.freesolo.co"
+DEFAULT_FREESOLO_BASE_URL = "https://api-dev.freesolo.co"
 FREESOLO_AUTH_VERIFY_PATH = "/api/auth/verify"
 
 

{freesolo_flash-0.2.0 → freesolo_flash-0.2.2}/flash/providers/runpod/pricing.py

@@ -24,7 +24,7 @@ logger = get_logger(__name__)
 
 CACHE_TTL_S = 6 * 3600.0  # 6h price cache
 _CACHE_PATH = Path.home() / ".flash" / "gpu_rates.json"
-_MEM: dict = {"ts": 0.0, "rates": {}}
+_MEM: dict = {"ts": 0.0, "rates": {}, "failed_ts": 0.0, "warned": False}
 
 
 def _static_rates() -> dict[str, float]:
@@ -76,6 +76,12 @@ def live_rates(refresh: bool = False) -> dict[str, float]:
     if not refresh:
         if _MEM["rates"] and now - _MEM["ts"] < CACHE_TTL_S:
             return {**static, **_MEM["rates"]}
+        # A recent live fetch FAILED (e.g. the client has no `runpod` module, no network, or no
+        # RunPod key — none of which change mid-process). The static rates are correct, so don't
+        # re-attempt (and re-warn) on every lookup; without this a single `flash gpus`/`train`
+        # logs the warning once per GPU class. Retry only after the cache TTL.
+        if _MEM["failed_ts"] and now - _MEM["failed_ts"] < CACHE_TTL_S:
+            return static
         try:
             disk = json.loads(_CACHE_PATH.read_text())
             if now - float(disk.get("ts", 0)) < CACHE_TTL_S and disk.get("rates"):
@@ -87,7 +93,10 @@ def live_rates(refresh: bool = False) -> dict[str, float]:
     try:
         fetched = _fetch_live_rates()
     except Exception as exc:
-        logger.warning("live GPU pricing unavailable (%s); using static rates", exc)
+        _MEM["failed_ts"] = now  # remember the failure so we don't refetch/rewarn every call
+        if not _MEM["warned"]:  # warn once per process, not once per GPU class
+            logger.warning("live GPU pricing unavailable (%s); using static rates", exc)
+            _MEM["warned"] = True
         fetched = {}
     if fetched:
         _MEM.update(ts=now, rates=fetched)

{freesolo_flash-0.2.0 → freesolo_flash-0.2.2}/flash/server/auth.py

@@ -25,7 +25,7 @@ INTERNAL_KEY_ENV = "FREESOLO_INTERNAL_KEY"
 # the bearer to this control plane. Any non-internal token is verified against the freesolo
 # backend and (on success) resolved to a per-token identity.
 FREESOLO_BASE_URL_ENV = "FREESOLO_BASE_URL"
-DEFAULT_FREESOLO_BASE_URL = "https://api.freesolo.co"
+DEFAULT_FREESOLO_BASE_URL = "https://api-dev.freesolo.co"
 _VERIFY_TIMEOUT_S = 5.0
 _VERIFY_CACHE_TTL_S = 300.0  # short TTL so it isn't a backend round-trip per request
 # Negative verdicts get a much SHORTER TTL than positives. The freesolo verify endpoint

{freesolo_flash-0.2.0 → freesolo_flash-0.2.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "freesolo-flash"
-version = "0.2.0"
+version = "0.2.2"
 description = "Flash — managed LoRA post-training (SFT/GRPO) for verifiers environments, driven by the `flash` CLI"
 readme = "README.md"
 # RunPod Flash supports Python 3.10-3.12 (not 3.13 yet).