diff --git a/python/sglang/benchmark/datasets/common.py b/python/sglang/benchmark/datasets/common.py
index 648b6140d..d6eaad4ef 100644
--- a/python/sglang/benchmark/datasets/common.py
+++ b/python/sglang/benchmark/datasets/common.py
@@ -66,8 +66,12 @@ def compute_random_lens(full_len: int, range_ratio: float, num: int) -> List[int
 
 @lru_cache(maxsize=1)
 def get_available_tokens(tokenizer):
-    """Get all available token ids from the tokenizer vocabulary."""
-    return list(tokenizer.get_vocab().values())
+    """Get valid token ids from the tokenizer vocabulary."""
+    return [
+        token_id
+        for token_id in tokenizer.get_vocab().values()
+        if isinstance(token_id, int)
+    ]
 
 
 def gen_prompt(tokenizer, token_num):
diff --git a/python/sglang/srt/observability/req_time_stats.py b/python/sglang/srt/observability/req_time_stats.py
index 22c5d06e9..151205db8 100644
--- a/python/sglang/srt/observability/req_time_stats.py
+++ b/python/sglang/srt/observability/req_time_stats.py
@@ -936,6 +936,16 @@ class SchedulerReqTimeStats(ReqTimeStatsBase):
         )
         self.trace_slice(stage, self.decode_prealloc_queue_entry_time, ts)
 
+        if self.enable_metrics and self.bootstrap_done_time > 0:
+            bootstrap_ms = (
+                self.bootstrap_done_time - self.decode_prealloc_queue_entry_time
+            ) * 1000
+            alloc_ms = (ts - self.bootstrap_done_time) * 1000
+            self.metrics_collector.observe_kv_transfer_bootstrap(
+                bootstrap_ms=bootstrap_ms,
+                alloc_ms=alloc_ms,
+            )
+
     def set_bootstrap_done_time(self, ts=None):
         ts = ts or time.perf_counter()
         if self.bootstrap_done_time == 0.0:
diff --git a/scripts/playground/replay_request_dump.py b/scripts/playground/replay_request_dump.py
index 5e42e80d6..7a5f6ef0d 100644
--- a/scripts/playground/replay_request_dump.py
+++ b/scripts/playground/replay_request_dump.py
@@ -1,10 +1,10 @@
 """
 Usage:
 # replay from a folder
-python3 replay_request_dump.py --file-number 100 --parallel 512 --input-folder /data/lianmin/sglang_request_dump/grok-mini-0220-engine-5756f8f94-28bm6/
+python3 replay_request_dump.py --file-number 100 --parallel 512 --input-folder /data/lianmin/sglang_request_dump/engine-34xd1/
 
 # replay from a single file
-python3 replay_request_dump.py --parallel 512 --input-file /data/sglang_crash_dump/memx-cti-34-sr1.xpop.twttr.net/crash_dump_2025-06-04_20-13-18.pkl
+python3 replay_request_dump.py --parallel 512 --input-file /data/sglang_crash_dump/crash_dump_2025-06-04_20-13-18.pkl
 """
 
 import argparse