gpu-dev 0.5.13__tar.gz → 0.5.14__tar.gz

{gpu_dev-0.5.13 → gpu_dev-0.5.14}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gpu-dev
-Version: 0.5.13
+Version: 0.5.14
 Summary: CLI tool for PyTorch GPU developer server reservations
 Author: PyTorch Team
 Requires-Python: >=3.10

{gpu_dev-0.5.13 → gpu_dev-0.5.14}/cli-tools/gpu-dev-cli/gpu_dev.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gpu-dev
-Version: 0.5.13
+Version: 0.5.14
 Summary: CLI tool for PyTorch GPU developer server reservations
 Author: PyTorch Team
 Requires-Python: >=3.10

{gpu_dev-0.5.13 → gpu_dev-0.5.14}/cli-tools/gpu-dev-cli/gpu_dev_cli/cli.py

@@ -688,6 +688,7 @@ def reserve(
             # and total wall-clock time drops from sum to max(each).
             from concurrent.futures import ThreadPoolExecutor
             config = load_config()
+            reservation_mgr = ReservationManager(config)
 
             with Live(
                 Spinner("dots", text="🚀 Loading…"), console=console
@@ -704,9 +705,7 @@ def reserve(
                     else:
                         f_ssh = ex.submit(validate_ssh_key_matches_github_user, config, None)
                         ssh_result = None
-                    f_avail = ex.submit(
-                        lambda: ReservationManager(config).get_gpu_availability_by_type()
-                    )
+                    f_avail = ex.submit(reservation_mgr.get_gpu_availability_by_type)
 
                     # Surface auth failure first (most actionable).
                     try:
@@ -2496,10 +2495,10 @@ def _show_availability() -> None:
             table = Table(
                 title="GPU Availability by Type (numbers are GPUs, not nodes)")
             table.add_column("GPU Type", style="cyan")
-            table.add_column("Available", style="green")
-            table.add_column("Max Reservable", style="bright_green")
+            table.add_column("Avail", style="green")
+            table.add_column("Max\nReservable", style="bright_green")
             table.add_column("Total", style="blue")
-            table.add_column("Queue Length", style="yellow")
+            table.add_column("Queue\nLength", style="yellow")
             table.add_column("Architecture", style="dim")
             table.add_column("Est. Wait Time", style="magenta")
 
@@ -2657,10 +2656,10 @@ def _show_availability_watch(interval: int) -> None:
                         table = Table(
                             title="GPU Availability by Type (numbers are GPUs, not nodes)")
                         table.add_column("GPU Type", style="cyan")
-                        table.add_column("Available", style="green")
-                        table.add_column("Max Reservable", style="blue")
+                        table.add_column("Avail", style="green")
+                        table.add_column("Max\nReservable", style="blue")
                         table.add_column("Total", style="blue")
-                        table.add_column("Queue Length", style="yellow")
+                        table.add_column("Queue\nLength", style="yellow")
                         table.add_column("Architecture", style="dim")
                         table.add_column("Est. Wait Time", style="magenta")
 

{gpu_dev-0.5.13 → gpu_dev-0.5.14}/cli-tools/gpu-dev-cli/gpu_dev_cli/interactive.py

@@ -88,9 +88,9 @@ def select_gpu_type_interactive(
     console.print("\n[cyan]🖥️  GPU Availability:[/cyan]")
     table = Table()
     table.add_column("GPU Type", style="cyan")
-    table.add_column("Available", style="green")
+    table.add_column("Avail", style="green")
     table.add_column("Total", style="blue")
-    table.add_column("Queue Length", style="yellow")
+    table.add_column("Queue\nLength", style="yellow")
     table.add_column("Est. Wait Time", style="magenta")
 
     choices = []

{gpu_dev-0.5.13 → gpu_dev-0.5.14}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "gpu-dev"
-version = "0.5.13"
+version = "0.5.14"
 description = "CLI tool for PyTorch GPU developer server reservations"
 authors = [{name = "PyTorch Team"}]
 readme = "cli-tools/gpu-dev-cli/README.md"

{gpu_dev-0.5.13 → gpu_dev-0.5.14}/terraform-gpu-devservers/mig-config.tf

@@ -25,23 +25,40 @@ resource "kubernetes_config_map" "gpu_dev_mig_parted_config" {
   depends_on = [helm_release.nvidia_gpu_operator]
 }
 
-# Optional declarative B200 MIG node label. Set b200_mig_node_name in tfvars (or override the
-# variable's default below) to dedicate a specific B200 node to the mixed profile. Empty string
-# means "no node currently labelled" — the existing all-disabled stays in effect.
+# Declarative B200 MIG node label. Set b200_mig_node_name (per workspace via the locals lookup
+# below, or override via tfvars / -var) to dedicate a specific B200 node to the mixed profile.
+# Empty string means "no node labelled" — every B200 stays full.
+#
+# Future cleanup: when we split a B200 CR into two ASGs (one with mig_profile, one without),
+# the user_data path will set this label at boot for any instance in the MIG-dedicated ASG —
+# matching the H100 cr3 pattern. Until then, this declarative label pins the role to a hostname.
+locals {
+  # Workspace-scoped defaults so the resource is a no-op in non-prod and no apply ever tries to
+  # label a node that doesn't exist.
+  default_b200_mig_node_by_workspace = {
+    prod = "ip-10-0-67-125.us-east-2.compute.internal"
+  }
+  b200_mig_node_effective = (
+    var.b200_mig_node_name != ""
+    ? var.b200_mig_node_name
+    : lookup(local.default_b200_mig_node_by_workspace, terraform.workspace, "")
+  )
+}
+
 variable "b200_mig_node_name" {
-  description = "Hostname of the B200 node to label with nvidia.com/mig.config=b200-6full-2mig-balanced. Leave empty to skip."
+  description = "Hostname of the B200 node to label with nvidia.com/mig.config=b200-6full-2mig-balanced. Leave empty to use the per-workspace default in mig-config.tf."
   type        = string
   default     = ""
 }
 
 resource "kubernetes_labels" "b200_mig_node" {
-  count = var.b200_mig_node_name == "" ? 0 : 1
+  count = local.b200_mig_node_effective == "" ? 0 : 1
 
   api_version = "v1"
   kind        = "Node"
 
   metadata {
-    name = var.b200_mig_node_name
+    name = local.b200_mig_node_effective
   }
 
   labels = {