Spaces:
Running on Zero
Running on Zero
Composer fallbacks: three distinct verified providers, per-provider timeouts, attempt logging
#7
by multimodalart HF Staff - opened
Why the overloaded error fires: the chain was baseten -> bare -> :fastest, but bare/:fastest can route to together, which 403s on this account — so one baseten hiccup killed all three attempts. Measured now: baseten 12-22s (8-way parallel, all valid JSON), deepinfra 43s, novita 80s (the old 60s timeout would kill it), together 403. New chain: baseten(45s) -> deepinfra(75s) -> novita(100s), each attempt's failure printed to the logs, and the error message no longer blames MiniMax-M3.
multimodalart changed pull request status to merged