From f457906c29e9e63e68914761285da211469d1b37 Mon Sep 17 00:00:00 2001
From: project6-dev <dev@project6.ai>
Date: Thu, 13 Aug 2026 01:02:53 +0000
Subject: [PATCH] fix: copy_blocks + max_completion_tokens (no new .so)

---
 .gitignore                      | 2 ++
 qwen3_6_scripts/paged_attn.py   | 3 ++-
 qwen3_6_scripts/protocol.py     | 3 +++
 qwen3_6_scripts/serving_chat.py | 3 +++
 4 files changed, 10 insertions(+), 1 deletion(-)
 create mode 100644 .gitignore

diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..7a60b85
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,2 @@
+__pycache__/
+*.pyc
diff --git a/qwen3_6_scripts/paged_attn.py b/qwen3_6_scripts/paged_attn.py
index e7daa2f..e8a8f2c 100644
--- a/qwen3_6_scripts/paged_attn.py
+++ b/qwen3_6_scripts/paged_attn.py
@@ -2256,4 +2256,5 @@ class PagedAttention:
     ) -> None:
         key_caches = [kv_cache[0] for kv_cache in kv_caches]
         value_caches = [kv_cache[1] for kv_cache in kv_caches]
-        ops.copy_blocks(key_caches, value_caches, src_to_dists)
+        import ixformer.functions as ixf_F
+        ixf_F.vllm_copy_cache(key_caches, value_caches, src_to_dists)
diff --git a/qwen3_6_scripts/protocol.py b/qwen3_6_scripts/protocol.py
index a51a1c5..3a041b4 100644
--- a/qwen3_6_scripts/protocol.py
+++ b/qwen3_6_scripts/protocol.py
@@ -180,6 +180,8 @@ class ChatCompletionRequest(OpenAIBaseModel):
     logprobs: Optional[bool] = False
     top_logprobs: Optional[int] = 0
     max_tokens: Optional[int] = None
+    # OpenAI newer API field — treat as alias for max_tokens
+    max_completion_tokens: Optional[int] = None
     n: Optional[int] = 1
     presence_penalty: Optional[float] = 0.0
     response_format: Optional[ResponseFormat] = None
@@ -193,6 +195,7 @@ class ChatCompletionRequest(OpenAIBaseModel):
     tool_choice: Optional[Union[Literal["none"], Literal["auto"],
                                 ChatCompletionNamedToolChoiceParam]] = "none"
     thinking: Optional[Union[bool, str, Dict[str, Any]]] = None
+    reasoning_effort: Optional[str] = None
 
     # NOTE this will be ignored by VLLM -- the model determines the behavior
     parallel_tool_calls: Optional[bool] = False
diff --git a/qwen3_6_scripts/serving_chat.py b/qwen3_6_scripts/serving_chat.py
index 4ed841b..366253b 100644
--- a/qwen3_6_scripts/serving_chat.py
+++ b/qwen3_6_scripts/serving_chat.py
@@ -394,6 +394,9 @@ class OpenAIServingChat(OpenAIServing):
             assert prompt_inputs is not None
 
             sampling_params: Union[SamplingParams, BeamSearchParams]
+            # OpenAI API: max_completion_tokens takes precedence over max_tokens
+            if request.max_completion_tokens is not None and request.max_tokens is None:
+                request.max_tokens = request.max_completion_tokens
             default_max_tokens = self.max_model_len - len(
                 prompt_inputs["prompt_token_ids"])
             if request.use_beam_search: