Fix prefill OOM due to wrong token calculation when page > 1 (#7397)

python/sglang/srt/managers/schedule_policy.py

@@ -55,6 +55,9 @@ IN_BATCH_PREFIX_CACHING_DEPRIORITIZE_THRESHOLD = int(
 )
 
 
+IGNORE_EOS_RESERVE_TOKENS = 1
+
+
 class CacheAwarePolicy(Enum):
     """Scheduling policies that are aware of the tree cache."""
 
@@ -293,6 +296,7 @@ class PrefillAdder:
         self.can_run_list = []
         self.new_chunked_req = None
         self.log_hit_tokens = 0
+        # TODO(lsyin): report the real input tokens excluding page alignment
         self.log_input_tokens = 0
 
         if running_batch is not None:
@@ -323,6 +327,9 @@ class PrefillAdder:
             - self.cur_rem_token_offset
         )
 
+    def ceil_paged_tokens(self, tokens: int) -> int:
+        return -(-tokens // self.page_size) * self.page_size
+
     def budget_state(self):
         if self.rem_total_tokens <= 0 or self.cur_rem_tokens <= 0:
             return AddReqResult.NO_TOKEN
@@ -334,9 +341,12 @@ class PrefillAdder:
 
         return AddReqResult.CONTINUE
 
-    def _prefill_one_req(
+    def _update_prefill_budget(
         self, prefix_len: int, extend_input_len: int, max_new_tokens: int
     ):
+        # TODO(lsyin): check this workaround logic, which only ensures the prefill will not out of memory, and may be too conservative
+        extend_input_len = self.ceil_paged_tokens(extend_input_len)
+
         self.rem_total_token_offset += extend_input_len + max_new_tokens
         self.cur_rem_token_offset += extend_input_len
         self.rem_input_tokens -= extend_input_len
@@ -351,7 +361,7 @@ class PrefillAdder:
         req.extend_input_len = min(req.extend_input_len, self.rem_chunk_tokens)
         req.fill_ids = req.fill_ids[: len(req.prefix_indices) + req.extend_input_len]
         self.can_run_list.append(req)
-        self._prefill_one_req(
+        self._update_prefill_budget(
             0,
             req.extend_input_len,
             (
@@ -373,6 +383,12 @@ class PrefillAdder:
             self.tree_cache.dec_lock_ref(last_node)
 
     def add_one_req_ignore_eos(self, req: Req, has_chunked_req: bool):
+        # Early exit if no enough tokens for the input tokens
+        if self.ceil_paged_tokens(req.extend_input_len) > min(
+            self.cur_rem_tokens, self.rem_total_tokens
+        ):
+            return AddReqResult.NO_TOKEN
+
         def add_req_state(r, insert_sort=False):
             new_token_ratio = (
                 1.0 if r.sampling_params.ignore_eos else self.new_token_ratio
@@ -382,7 +398,9 @@ class PrefillAdder:
             )
             tokens_occupied = len(r.origin_input_ids) + len(r.output_ids)
 
-            if tokens_left > 0:
+            if tokens_left <= 0:
+                return
+
             if not insert_sort:
                 self.req_states.append((tokens_left, tokens_occupied))
             else:
@@ -407,13 +425,11 @@ class PrefillAdder:
         cur_rem_tokens = self.cur_rem_tokens - len(req.origin_input_ids)
         tokens_freed = 0
         for i, (tokens_left, tokens_occupied) in enumerate(self.req_states):
-            decode_steps = (
-                self.req_states[i + 1][0]
-                if i + 1 < len(self.req_states)
-                else tokens_left
-            )
+            # tokens_left gives a reservative calculation as the last token is not stored
             bs = len(self.req_states) - i
-            if cur_rem_tokens + tokens_freed - decode_steps * bs <= 0:
+            min_free_tokens = cur_rem_tokens + tokens_freed - tokens_left * bs
+            # reserve tokens for corner cases
+            if min_free_tokens <= IGNORE_EOS_RESERVE_TOKENS * bs:
                 return AddReqResult.NO_TOKEN
             tokens_freed += tokens_occupied
 
@@ -423,7 +439,7 @@ class PrefillAdder:
         ):
             # Non-chunked prefill
             self.can_run_list.append(req)
-            self._prefill_one_req(
+            self._update_prefill_budget(
                 0,
                 req.extend_input_len,
                 min(req.sampling_params.max_new_tokens, CLIP_MAX_NEW_TOKENS_ESTIMATION),
@@ -439,7 +455,7 @@ class PrefillAdder:
             req.fill_ids = req.fill_ids[:trunc_len]
             self.can_run_list.append(req)
             self.new_chunked_req = req
-            self._prefill_one_req(0, trunc_len, 0)
+            self._update_prefill_budget(0, trunc_len, 0)
 
         return self.budget_state()
 
@@ -453,7 +469,7 @@ class PrefillAdder:
 
         # adjusting the input_tokens based on host_hit_length and page_size
         real_input_tokens = req.extend_input_len - req.host_hit_length
-        real_input_tokens = -(-real_input_tokens // self.page_size) * self.page_size
+        real_input_tokens = self.ceil_paged_tokens(real_input_tokens)
         prefix_len = len(req.prefix_indices)
 
         if total_tokens >= self.rem_total_tokens:
@@ -475,7 +491,7 @@ class PrefillAdder:
                 req.extend_input_len = len(req.fill_ids) - len(req.prefix_indices)
                 prefix_len = len(req.prefix_indices)
 
-            input_tokens = -(-req.extend_input_len // self.page_size) * self.page_size
+            input_tokens = self.ceil_paged_tokens(req.extend_input_len)
 
             if input_tokens >= self.rem_input_tokens and len(self.can_run_list) != 0:
                 return AddReqResult.OTHER
@@ -484,7 +500,7 @@ class PrefillAdder:
                 # Non-chunked prefill
                 self.can_run_list.append(req)
                 self.tree_cache.inc_lock_ref(req.last_node)
-                self._prefill_one_req(
+                self._update_prefill_budget(
                     prefix_len,
                     input_tokens,
                     min(
@@ -505,6 +521,6 @@ class PrefillAdder:
                 self.can_run_list.append(req)
                 self.new_chunked_req = req
                 self.tree_cache.inc_lock_ref(req.last_node)
-                self._prefill_one_req(prefix_len, trunc_len, 0)
+                self._update_prefill_budget(prefix_len, trunc_len, 0)
 
         return self.budget_state()