fix ci ut

Author: liuruyan

fastdeploy/entrypoints/llm.py

@@ -450,11 +450,7 @@ def _build_prompt_logprobs(
                                    tensors.
         """
 
-        token_ids, logprobs, ranks = (
-            prompt_logprobs_tensors.logprob_token_ids,
-            prompt_logprobs_tensors.logprobs,
-            prompt_logprobs_tensors.selected_token_ranks,
-        )
+        token_ids, logprobs, ranks = prompt_logprobs_tensors[:3]
 
         # Detokenize non-incrementally.
         # Output is flat: [num_tok, num_lps] -> [num_tok * num_lps]

fastdeploy/entrypoints/openai/serving_chat.py

@@ -968,11 +968,7 @@ def _build_prompt_logprobs(
                                    tensors.
         """
 
-        token_ids, logprobs, ranks = (
-            prompt_logprobs_tensors.logprob_token_ids,
-            prompt_logprobs_tensors.logprobs,
-            prompt_logprobs_tensors.selected_token_ranks,
-        )
+        token_ids, logprobs, ranks = prompt_logprobs_tensors[:3]
 
         # Normalize to plain Python lists (support both Tensor and list inputs)
         if hasattr(token_ids, "tolist"):

fastdeploy/entrypoints/openai/serving_completion.py

@@ -900,11 +900,7 @@ def _build_prompt_logprobs(
                                    tensors.
         """
 
-        token_ids, logprobs, ranks = (
-            prompt_logprobs_tensors.logprob_token_ids,
-            prompt_logprobs_tensors.logprobs,
-            prompt_logprobs_tensors.selected_token_ranks,
-        )
+        token_ids, logprobs, ranks = prompt_logprobs_tensors[:3]
 
         # Normalize to plain Python lists (support both Tensor and list inputs)
         if hasattr(token_ids, "tolist"):

tests/ce/server/test_logprobs.py

@@ -3,6 +3,17 @@
 from core import TEMPLATE, URL, build_request_payload, send_request
 
 
+def _strip_logits_stats(obj):
+    """Recursively remove 'logits_stats' keys from logprobs response."""
+    if isinstance(obj, dict):
+        obj.pop("logits_stats", None)
+        for v in obj.values():
+            _strip_logits_stats(v)
+    elif isinstance(obj, list):
+        for item in obj:
+            _strip_logits_stats(item)
+
+
 def test_unstream_with_logprobs():
     """
     测试非流式响应开启 logprobs 后，返回的 token 概率信息是否正确。
@@ -21,6 +32,7 @@ def test_unstream_with_logprobs():
     response = send_request(URL, payload)
     print(json.dumps(response.json(), indent=2, ensure_ascii=False))
     resp_json = response.json()
+    _strip_logits_stats(resp_json)
 
     # 校验返回内容与概率信息
     assert resp_json["choices"][0]["message"]["content"] == "牛顿的"
@@ -99,6 +111,7 @@ def test_stream_with_logprobs():
             print(json.dumps(result_chunk, indent=2, ensure_ascii=False))
             break
 
+    _strip_logits_stats(result_chunk)
     # 校验概率字段
     assert result_chunk["choices"][0]["delta"]["content"] == "牛顿"
     assert result_chunk["choices"][0]["logprobs"]["content"][0]["token"] == "牛顿"
@@ -184,6 +197,7 @@ def test_stream_with_temp_scaled_logprobs():
             print(json.dumps(result_chunk, indent=2, ensure_ascii=False))
             break
 
+    _strip_logits_stats(result_chunk)
     # 校验概率字段
     assert result_chunk["choices"][0]["delta"]["content"] == "牛顿"
     assert result_chunk["choices"][0]["logprobs"]["content"][0]["token"] == "牛顿"
@@ -229,6 +243,7 @@ def test_stream_with_top_p_normalized_logprobs():
             print(json.dumps(result_chunk, indent=2, ensure_ascii=False))
             break
 
+    _strip_logits_stats(result_chunk)
     # 校验概率字段
     assert result_chunk["choices"][0]["delta"]["content"] == "牛顿"
     assert result_chunk["choices"][0]["logprobs"]["content"][0]["token"] == "牛顿"

tests/e2e/4cards_cases/test_ernie_21b_tp1_dp4.py

@@ -23,6 +23,18 @@
 import pytest
 import requests
 
+
+def _strip_logits_stats(obj):
+    """Recursively remove 'logits_stats' keys from logprobs response."""
+    if isinstance(obj, dict):
+        obj.pop("logits_stats", None)
+        for v in obj.values():
+            _strip_logits_stats(v)
+    elif isinstance(obj, list):
+        for item in obj:
+            _strip_logits_stats(item)
+
+
 tests_dir = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", ".."))
 sys.path.insert(0, tests_dir)
 
@@ -606,6 +618,7 @@ def test_non_stream_with_logprobs(api_url):
     resp_json = send_request(url=api_url, payload=payload).json()
 
     logprobs = resp_json["choices"][0]["logprobs"]
+    _strip_logits_stats(logprobs)
 
     base_path = os.getenv("MODEL_PATH")
     if base_path:

tests/e2e/4cards_cases/test_ernie_21b_tp1_dp4_mtp.py

@@ -23,6 +23,18 @@
 import pytest
 import requests
 
+
+def _strip_logits_stats(obj):
+    """Recursively remove 'logits_stats' keys from logprobs response."""
+    if isinstance(obj, dict):
+        obj.pop("logits_stats", None)
+        for v in obj.values():
+            _strip_logits_stats(v)
+    elif isinstance(obj, list):
+        for item in obj:
+            _strip_logits_stats(item)
+
+
 tests_dir = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", ".."))
 sys.path.insert(0, tests_dir)
 
@@ -512,6 +524,7 @@ def test_non_stream_with_logprobs(api_url):
     resp_json = send_request(url=api_url, payload=payload).json()
 
     logprobs = resp_json["choices"][0]["logprobs"]
+    _strip_logits_stats(logprobs)
 
     base_path = os.getenv("MODEL_PATH")