[WIP] Before fix bench_offload.py.

nanovllm/engine/model_runner.py

@@ -37,7 +37,7 @@ class ModelRunner:
         self.sampler = GreedySampler()
         
         #> Disable warmup for debugging
-        # self.warmup_model()
+        self.warmup_model()
         
         self.allocate_kv_cache()
         if not self.enforce_eager:
@@ -62,7 +62,7 @@ class ModelRunner:
                 self.shm.unlink()
         if not self.enforce_eager:
             del self.graphs, self.graph_pool
-        torch.cuda.synchronize()
+        # torch.cuda.synchronize()
         dist.destroy_process_group()
 
     def loop(self):

nanovllm/engine/scheduler.py

@@ -35,7 +35,29 @@ class Scheduler:
             if Observer.ttft_start == 0:
                 Observer.ttft_start = perf_counter_ns()
             seq = self.waiting[0]
-            if num_batched_tokens + len(seq) > self.max_num_batched_tokens or not self.kvcache_manager.can_allocate(seq):
+
+            # Check if sequence is too large
+            if not self.running and num_seqs == 0:
+                # First sequence, give clear error if it can't be scheduled
+                if len(seq) > self.max_num_batched_tokens:
+                    raise RuntimeError(
+                        f"Sequence too long: {len(seq)} tokens exceeds "
+                        f"max_num_batched_tokens={self.max_num_batched_tokens}. "
+                        f"Increase max_num_batched_tokens (set equal to max_model_len for long sequences)."
+                    )
+                if not self.kvcache_manager.can_allocate(seq):
+                    blocks_needed = seq.num_blocks
+                    blocks_available = self.kvcache_manager.num_free_blocks
+                    raise RuntimeError(
+                        f"Cannot allocate KV cache for sequence: "
+                        f"need {blocks_needed} blocks ({len(seq)} tokens), "
+                        f"but only {blocks_available} blocks available. "
+                        f"Increase max_model_len to allocate more blocks."
+                    )
+
+            if num_batched_tokens + len(seq) > self.max_num_batched_tokens:
+                break
+            if not self.kvcache_manager.can_allocate(seq):
                 break
             num_seqs += 1
             self.kvcache_manager.allocate(seq)
@@ -60,7 +82,7 @@ class Scheduler:
                 num_seqs += 1
                 self.kvcache_manager.may_append(seq)
                 scheduled_seqs.append(seq)
-        assert scheduled_seqs
+        assert scheduled_seqs, "No sequences scheduled - this should not happen"
         self.running.extendleft(reversed(scheduled_seqs))
         return scheduled_seqs, False
 

nanovllm/kvcache/offload_engine.py

@@ -201,7 +201,7 @@ class OffloadEngine:
         # This prevents undefined behavior on first load_to_slot_layer call
         for slot_idx in range(self.num_ring_slots):
             self.ring_slot_compute_done[slot_idx].record()
-        torch.cuda.synchronize()  # Ensure all events are recorded
+        # torch.cuda.synchronize()  # Ensure all events are recorded
 
         # ========== Event tracking for async transfers ==========
         self.pending_events: Dict[Tuple[int, int], torch.cuda.Event] = {}

nanovllm/layers/attention.py

@@ -32,6 +32,10 @@ def store_kvcache(
     """
     # Filter out invalid slots (slot == -1)
     valid_mask = slot_mapping >= 0
+
+    is_capturing = torch.cuda.is_current_stream_capturing()
+
+    if not is_capturing:
         if not valid_mask.any():
             return
 
@@ -51,6 +55,7 @@ def store_kvcache(
     valid_values_flat = valid_values.reshape(-1, D)
 
     # In-place scatter using index_copy_
+    # 即使 valid_slots 为空张量，index_copy_ 也是安全的（不会修改数据）。
     k_cache_flat.index_copy_(0, valid_slots.long(), valid_keys_flat)
     v_cache_flat.index_copy_(0, valid_slots.long(), valid_values_flat)
 
@@ -86,7 +91,7 @@ class Attention(nn.Module):
         )
         
         #! Ensure synchronization before accessing k_cache/v_cache
-        torch.cuda.synchronize()
+        # torch.cuda.synchronize()
         #! =======================================================
 
         if is_chunked_offload:

tests/test_needle.py

@@ -123,7 +123,7 @@ if __name__ == "__main__":
     parser.add_argument(
         "--max-model-len",
         type=int,
-        default=36 * 1024,
+        default=128 * 1024,
         help="Maximum model context length"
     )
     parser.add_argument(

tests/test_sequential.py

@@ -148,7 +148,7 @@ if __name__ == "__main__":
     parser.add_argument(
         "--model", "-m",
         type=str,
-        default=os.path.expanduser("~/models/Qwen3-4B-Instruct-2507/"),
+        default=os.path.expanduser("~/models/Qwen3-0.6B/"),
         help="Path to model"
     )
     parser.add_argument(

tests/utils.py

@@ -96,46 +96,41 @@ def generate_needle_prompt(
         idx = int(len(parts) * needle_position)
         return max(0, min(idx, len(parts)))
 
-    # Phase 1: Build haystack with full paragraphs until we exceed target
+    # Pre-compute tokens per paragraph for efficiency (avoid O(n²) tokenization)
+    para_tokens = []
+    for para in HAYSTACK_PARAGRAPHS:
+        para_tokens.append(len(tokenizer.encode(para, add_special_tokens=False)))
+    avg_tokens_per_para = sum(para_tokens) / len(para_tokens)
+
+    # Estimate overhead (needle + question + chat template)
+    overhead_prompt = build_prompt([HAYSTACK_PARAGRAPHS[0]], 0)
+    overhead_tokens = count_tokens(overhead_prompt) - para_tokens[0]
+
+    # Phase 1: Estimate number of paragraphs needed
+    estimated_paras = int((target_length - overhead_tokens) / avg_tokens_per_para) + 1
+
+    # Build haystack with estimated paragraphs
     haystack_parts = []
-    para_idx = 0
+    for i in range(estimated_paras):
+        haystack_parts.append(HAYSTACK_PARAGRAPHS[i % len(HAYSTACK_PARAGRAPHS)])
 
-    while True:
+    # Phase 2: Adjust to get closer to target
+    prompt = build_prompt(haystack_parts, get_needle_idx(haystack_parts))
+    current_tokens = count_tokens(prompt)
+
+    # Add more if under target
+    para_idx = len(haystack_parts)
+    while current_tokens < target_length and para_idx < 100000:
         para = HAYSTACK_PARAGRAPHS[para_idx % len(HAYSTACK_PARAGRAPHS)]
-        test_parts = haystack_parts + [para]
-        prompt = build_prompt(test_parts, get_needle_idx(test_parts))
-
-        if count_tokens(prompt) > target_length:
-            break
-
         haystack_parts.append(para)
+        current_tokens += para_tokens[para_idx % len(HAYSTACK_PARAGRAPHS)]
         para_idx += 1
 
-        if para_idx > 10000:  # Safety limit
-            break
-
-    # Phase 2: Fine-tune by adding words from next paragraph
-    next_para = HAYSTACK_PARAGRAPHS[para_idx % len(HAYSTACK_PARAGRAPHS)]
-    words = next_para.split()
-
-    best_parts = haystack_parts.copy()
-    best_diff = abs(target_length - count_tokens(build_prompt(haystack_parts, get_needle_idx(haystack_parts))))
-
-    for i in range(1, len(words) + 1):
-        partial = " ".join(words[:i]) + " "
-        test_parts = haystack_parts + [partial]
-        prompt = build_prompt(test_parts, get_needle_idx(test_parts))
-        token_count = count_tokens(prompt)
-        diff = abs(target_length - token_count)
-
-        if diff < best_diff:
-            best_diff = diff
-            best_parts = test_parts.copy()
-
-        if token_count >= target_length:
-            break
-
-    haystack_parts = best_parts
+    # Remove if over target
+    while current_tokens > target_length + 100 and len(haystack_parts) > 1:
+        removed_para_idx = (len(haystack_parts) - 1) % len(HAYSTACK_PARAGRAPHS)
+        haystack_parts.pop()
+        current_tokens -= para_tokens[removed_para_idx]
 
     # Final build
     needle_idx = get_needle_idx(haystack_parts)