feat: monitor g7e.48xlarge capacity

README.md

@@ -27,10 +27,10 @@ AWS_PROFILE=<profile> uv run gpu-capacity
 # then open http://127.0.0.1:8300
 ```
 
-One representative (smallest) size per GPU family is swept automatically when
-`AUTO_REFRESH_MINUTES` is set (the deployed Space uses 15); any other size can
-be probed on demand from the dashboard input or the API. Results are cached in
-`state.json` and shown with their age; they fade once older than
+One representative (smallest) size per GPU family, plus `g7e.48xlarge`, is swept
+automatically when `AUTO_REFRESH_MINUTES` is set (the deployed Space uses 15);
+any other size can be probed on demand from the dashboard input or the API.
+Results are cached in `state.json` and shown with their age; they fade once older than
 `STALE_AFTER_SECONDS` (default 900).
 
 On the Space, `STATE_FILE` lives in the mounted HF bucket

gpu_capacity/config.py

@@ -1,9 +1,9 @@
 import os
 
-# One representative (smallest) size per GPU family, covered by the background
-# sweep. Any other size can be probed on demand via the API or dashboard.
-# Kept to small sizes on purpose: a successful probe holds the reservation
-# ~2 s billed at the instance's on-demand rate.
+# One representative (smallest) size per GPU family, plus g7e.48xlarge, covered
+# by the background sweep. Any other size can be probed on demand via the API
+# or dashboard. Most sizes are kept small: a successful probe holds the
+# reservation ~2 s billed at the instance's on-demand rate.
 INSTANCE_TYPES = [
     "g4ad.xlarge",
     "g4dn.xlarge",
@@ -12,6 +12,7 @@ INSTANCE_TYPES = [
     "g6.xlarge",
     "g6e.xlarge",
     "g7e.2xlarge",
+    "g7e.48xlarge",
     "gr6.4xlarge",
     "p3.2xlarge",
     "p4d.24xlarge",