diff --git a/src/h3_blackwell_runtime/checkpoint.py b/src/h3_blackwell_runtime/checkpoint.py
index 4ddd308..11b9d5c 100644
--- a/src/h3_blackwell_runtime/checkpoint.py
+++ b/src/h3_blackwell_runtime/checkpoint.py
@@ -32,7 +32,7 @@ class H3Checkpoint:
             with fastsafe_open(filenames=[self.path], nogds=True, device=self._fast_device()) as checkpoint:
                 self._no_mmap_tensors = {
                     name: checkpoint.get_tensor(name).clone().detach()
-                    for name in checkpoint.get_keys()
+                    for name in checkpoint.keys()
                 }
         return self._no_mmap_tensors
 
diff --git a/src/h3_blackwell_runtime/vae_decoder.py b/src/h3_blackwell_runtime/vae_decoder.py
index 39dba9c..67763e3 100644
--- a/src/h3_blackwell_runtime/vae_decoder.py
+++ b/src/h3_blackwell_runtime/vae_decoder.py
@@ -172,7 +172,7 @@ class MiniMaxH3VideoVAE(nn.Module):
             with fastsafe_open(filenames=[str(path)], nogds=True, device=fast_device) as checkpoint:
                 available_weights = {
                     name: checkpoint.get_tensor(name).clone().detach()
-                    for name in checkpoint.get_keys()
+                    for name in checkpoint.keys()
                 }
             available = set(available_weights)
             missing = sorted(set(expected) - available)
diff --git a/tools/direct_t2v_preview.py b/tools/direct_t2v_preview.py
index e62b0ca..af60f7c 100644
--- a/tools/direct_t2v_preview.py
+++ b/tools/direct_t2v_preview.py
@@ -4,6 +4,8 @@ import argparse
 import os
 from pathlib import Path
 import subprocess
+import time
+from datetime import datetime, timezone
 import warnings
 
 warnings.filterwarnings("ignore", message="Found GPU0 NVIDIA GB10 which is of cuda capability 12.1.*", category=UserWarning)
@@ -32,11 +34,15 @@ parser.add_argument("--attention", choices=("sage2", "sdpa", "sage3"), default="
 parser.add_argument("--model-timesteps-capture", type=Path, help="Directory containing captured input_XX.pt H3 timesteps for strict parity checks.")
 parser.add_argument("--profile-memory", action="store_true")
 args = parser.parse_args()
+started = time.perf_counter()
+last_report = started
 
 
 def report_memory(stage: str) -> None:
+    global last_report
     if not args.profile_memory:
         return
+    now = time.perf_counter()
     rss_kb = 0
     try:
         with open("/proc/self/status", encoding="utf-8") as file:
@@ -48,7 +54,8 @@ def report_memory(stage: str) -> None:
         pass
     cuda_alloc = torch.cuda.memory_allocated() / 1024**3 if torch.cuda.is_available() else 0.0
     cuda_reserved = torch.cuda.memory_reserved() / 1024**3 if torch.cuda.is_available() else 0.0
-    print({"stage": stage, "rss_gb": round(rss_kb / 1024**2, 3), "cuda_alloc_gb": round(cuda_alloc, 3), "cuda_reserved_gb": round(cuda_reserved, 3), "fast_safetensors": os.getenv("H3_FAST_SAFETENSORS", ""), "disable_mmap": os.getenv("H3_DISABLE_MMAP", "")}, flush=True)
+    print({"stage": stage, "ts": datetime.now(timezone.utc).isoformat(), "epoch_s": round(time.time(), 3), "elapsed_s": round(now - started, 3), "delta_s": round(now - last_report, 3), "rss_gb": round(rss_kb / 1024**2, 3), "cuda_alloc_gb": round(cuda_alloc, 3), "cuda_reserved_gb": round(cuda_reserved, 3), "fast_safetensors": os.getenv("H3_FAST_SAFETENSORS", ""), "disable_mmap": os.getenv("H3_DISABLE_MMAP", "")}, flush=True)
+    last_report = now
 
 report_memory("start")
 checkpoint = H3Checkpoint("/models/minimax_h3_fl2va_pruned_nvfp4.safetensors")