fix param check

Author: qwes5s5

fastdeploy/engine/sampling_params.py

@@ -209,17 +209,15 @@ def _verify_args(self) -> None:
             )
 
         if os.getenv("FD_USE_GET_SAVE_OUTPUT_V1", "0") == "0":
-            if self.logprobs is not None and self.logprobs < 0:
-                raise ValueError(f"logprobs must be greater than 0, got {self.logprobs}.")
-            if self.logprobs is not None and self.logprobs > 20:
-                raise ValueError("Invalid value for 'top_logprobs': must be less than or equal to 20.")
+            if self.logprobs is not None and (self.logprobs < 0 or self.logprobs > 20):
+                raise ValueError("Invalid value for 'top_logprobs': must be between 0 and 20.")
             if self.prompt_logprobs is not None:
                 raise ValueError("prompt_logprobs is not support when FD_USE_GET_SAVE_OUTPUT_V1 is disabled.")
         elif os.getenv("FD_USE_GET_SAVE_OUTPUT_V1", "0") == "1":
             if self.logprobs is not None and self.logprobs < -1:
-                raise ValueError(f"logprobs must be greater than -1, got {self.logprobs}.")
+                raise ValueError(f"logprobs must be a non-negative value or -1, got {self.logprobs}.")
             if self.prompt_logprobs is not None and self.prompt_logprobs < -1:
-                raise ValueError(f"prompt_logprobs must be greater than or equal to -1, got {self.prompt_logprobs}.")
+                raise ValueError(f"prompt_logprobs a must be non-negative value or -1, got {self.prompt_logprobs}.")
 
         if not 0 <= self.seed <= 922337203685477580:
             raise ValueError("seed must be in [0, 922337203685477580], got " f"{self.seed}.")

fastdeploy/entrypoints/engine_client.py

@@ -357,11 +357,15 @@ def valid_parameters(self, data):
                 api_server_logger.error(err_msg)
                 raise ParameterError("prompt_logprobs", err_msg)
 
-            if prompt_logprobs == -1:
-                prompt_logprobs = self.ori_vocab_size
+            if prompt_logprobs == -1 and self.ori_vocab_size > max_logprobs:
+                err_msg = f"The requested value of ({self.ori_vocab_size}) for prompt_logprobs (-1) exceeds the maximum allowed value of ({max_logprobs})"
+                api_server_logger.error(err_msg)
+                raise ValueError("prompt_logprobs", err_msg)
 
             if prompt_logprobs < -1:
-                err_msg = f"Invalid 'prompt_logprobs': must be >= -1, got {prompt_logprobs}."
+                err_msg = (
+                    f"prompt_logprobs must be a non-negative value or -1; the current value is {prompt_logprobs}."
+                )
                 api_server_logger.error(err_msg)
                 raise ValueError("prompt_logprobs", err_msg)
 
@@ -384,19 +388,18 @@ def valid_parameters(self, data):
                 raise ParameterError("top_logprobs", err_msg)
 
             if os.getenv("FD_USE_GET_SAVE_OUTPUT_V1", "0") == "0":
-                if top_logprobs < 0:
-                    err_msg = "Invalid value for 'top_logprobs': must be >= 0."
-                    raise ValueError("top_logprobs", err_msg)
-
-                if top_logprobs > 20:
-                    err_msg = "Invalid value for 'top_logprobs': must be <= 20."
+                if top_logprobs < 0 or top_logprobs > 20:
+                    err_msg = f"top_logprobs must be between 0 and 20; the current value is {top_logprobs}."
+                    api_server_logger.error(err_msg)
                     raise ValueError("top_logprobs", err_msg)
             else:
-                if top_logprobs == -1:
-                    top_logprobs = self.ori_vocab_size
+                if top_logprobs == -1 and self.ori_vocab_size > max_logprobs:
+                    err_msg = f"The requested value of ({self.ori_vocab_size}) for top_logprobs (-1) exceeds the maximum allowed value of ({max_logprobs})"
+                    api_server_logger.error(err_msg)
+                    raise ValueError("top_logprobs", err_msg)
 
                 if top_logprobs < -1:
-                    err_msg = f"Invalid 'top_logprobs': must be >= -1, got {top_logprobs}."
+                    err_msg = f"top_logprobs must be a non-negative value or -1; the current value is {top_logprobs}."
                     api_server_logger.error(err_msg)
                     raise ValueError("top_logprobs", err_msg)
 

fastdeploy/entrypoints/llm.py

@@ -351,8 +351,10 @@ def _add_request(
 
             if current_sampling_params.logprobs is not None:
                 num_logprobs = current_sampling_params.logprobs
-                if num_logprobs == -1:
-                    num_logprobs = ori_vocab_size
+                if num_logprobs == -1 and ori_vocab_size > max_logprobs:
+                    raise ValueError(
+                        f"Number of logprobs(-1) requested ({ori_vocab_size}) exceeds maximum allowed value ({max_logprobs})."
+                    )
                 if num_logprobs > max_logprobs:
                     raise ValueError(
                         f"Number of logprobs requested ({num_logprobs}) exceeds maximum allowed value ({max_logprobs})."
@@ -363,8 +365,10 @@ def _add_request(
                 if kwargs.get("stream"):
                     raise ValueError("prompt_logprobs is not supported with streaming.")
                 num_prompt_logprobs = current_sampling_params.prompt_logprobs
-                if num_prompt_logprobs == -1:
-                    num_prompt_logprobs = ori_vocab_size
+                if num_prompt_logprobs == -1 and ori_vocab_size > max_logprobs:
+                    raise ValueError(
+                        f"Number of prompt_logprobs(-1) requested ({ori_vocab_size}) exceeds maximum allowed value ({max_logprobs})."
+                    )
                 if num_prompt_logprobs > max_logprobs:
                     raise ValueError(
                         f"Number of logprobs requested ({num_prompt_logprobs}) exceeds maximum allowed value ({max_logprobs})."
@@ -561,7 +565,7 @@ def _run_engine(
                         result.outputs.logprobs = self._build_sample_logprobs(
                             result.outputs.top_logprobs, topk_logprobs
                         )
-                    if result.prompt_logprobs and num_prompt_logprobs:
+                    if result.prompt_logprobs is not None and num_prompt_logprobs is not None:
                         if num_prompt_logprobs == -1:
                             num_prompt_logprobs = self.llm_engine.cfg.model_config.ori_vocab_size
                         result.prompt_logprobs = self._build_prompt_logprobs(

fastdeploy/entrypoints/openai/protocol.py

@@ -788,7 +788,7 @@ def check_logprobs(cls, data):
             if top_logprobs < -1:
                 raise ValueError("`top_logprobs` must be a greater than -1.")
 
-            if top_logprobs > 0 and not data.get("logprobs"):
+            if not data.get("logprobs"):
                 raise ValueError("when using `top_logprobs`, `logprobs` must be set to true.")
 
         if (prompt_logprobs := data.get("prompt_logprobs")) is not None:

fastdeploy/entrypoints/openai/serving_chat.py

@@ -299,7 +299,7 @@ async def chat_completion_stream_generator(
                         for i in range(num_choices):
                             prompt_logprobs_res: Optional[PromptLogprobs] = None
                             prompt_logprobs_tensors = res.get("prompt_logprobs", None)
-                            if request.prompt_logprobs and prompt_logprobs_tensors is not None:
+                            if request.prompt_logprobs is not None and prompt_logprobs_tensors is not None:
                                 num_prompt_logprobs = (
                                     request.prompt_logprobs
                                     if request.prompt_logprobs != -1
@@ -583,7 +583,7 @@ async def chat_completion_full_generator(
                             if draft_logprobs_res and draft_logprobs_res.content is not None:
                                 draft_logprob_contents[idx].extend(draft_logprobs_res.content)
                     prompt_logprobs_tensors = data.get("prompt_logprobs", None)
-                    if request.prompt_logprobs and prompt_logprobs_tensors is not None:
+                    if request.prompt_logprobs is not None and prompt_logprobs_tensors is not None:
                         num_prompt_logprobs = (
                             request.prompt_logprobs
                             if request.prompt_logprobs != -1

fastdeploy/entrypoints/openai/serving_completion.py

@@ -445,7 +445,7 @@ async def completion_stream_generator(
                     prompt_logprobs_res: Optional[PromptLogprobs] = None
                     if first_iteration[idx]:
                         prompt_logprobs_tensors = res.get("prompt_logprobs", None)
-                        if request.prompt_logprobs and prompt_logprobs_tensors is not None:
+                        if request.prompt_logprobs is not None and prompt_logprobs_tensors is not None:
                             num_prompt_logprobs = (
                                 request.prompt_logprobs
                                 if request.prompt_logprobs != -1
@@ -495,7 +495,7 @@ async def completion_stream_generator(
                     output_draft_top_logprobs = output["draft_top_logprobs"]
                     logprobs_res: Optional[CompletionLogprobs] = None
                     draft_logprobs_res: Optional[CompletionLogprobs] = None
-                    if request.logprobs and output_top_logprobs is not None:
+                    if request.logprobs is not None and output_top_logprobs is not None:
                         num_logprobs = (
                             request.logprobs if request.logprobs != -1 else self.engine_client.ori_vocab_size
                         )
@@ -644,7 +644,7 @@ def request_output_to_completion_response(
                 )
             prompt_logprobs_res: Optional[PromptLogprobs] = None
             prompt_logprobs_tensors = final_res.get("prompt_logprobs_tensors", None)
-            if request.prompt_logprobs and prompt_logprobs_tensors is not None:
+            if request.prompt_logprobs is not None and prompt_logprobs_tensors is not None:
                 num_prompt_logprobs = (
                     request.prompt_logprobs if request.prompt_logprobs != -1 else self.engine_client.ori_vocab_size
                 )

tests/engine/test_sampling_params.py

@@ -61,7 +61,7 @@ def test_logprobs_invalid_less_than_minus_one(self):
                 params = SamplingParams(logprobs=-2)
                 params._verify_args()
 
-            self.assertIn("logprobs must be greater than -1", str(cm.exception))
+            self.assertIn("logprobs must be a non-negative value or -1", str(cm.exception))
             self.assertIn("got -2", str(cm.exception))
 
     def test_logprobs_invalid_less_than_zero(self):
@@ -71,8 +71,7 @@ def test_logprobs_invalid_less_than_zero(self):
                 params = SamplingParams(logprobs=-1)
                 params._verify_args()
 
-            self.assertIn("logprobs must be greater than 0", str(cm.exception))
-            self.assertIn("got -1", str(cm.exception))
+            self.assertIn("Invalid value for 'top_logprobs': must be between 0 and 20", str(cm.exception))
 
     def test_logprobs_greater_than_20_with_v1_disabled(self):
         """Test logprobs greater than 20 when FD_USE_GET_SAVE_OUTPUT_V1 is disabled"""
@@ -81,7 +80,7 @@ def test_logprobs_greater_than_20_with_v1_disabled(self):
                 params = SamplingParams(logprobs=21)
                 params._verify_args()
 
-            self.assertEqual("Invalid value for 'top_logprobs': must be less than or equal to 20.", str(cm.exception))
+            self.assertEqual("Invalid value for 'top_logprobs': must be between 0 and 20.", str(cm.exception))
 
     def test_logprobs_greater_than_20_with_v1_enabled(self):
         """Test logprobs greater than 20 when FD_USE_GET_SAVE_OUTPUT_V1 is enabled"""
@@ -127,7 +126,7 @@ def test_prompt_logprobs_invalid_less_than_minus_one(self):
                 params = SamplingParams(prompt_logprobs=-2)
                 params._verify_args()
 
-            self.assertIn("prompt_logprobs must be greater than or equal to -1", str(cm.exception))
+            self.assertIn("prompt_logprobs a must be non-negative value or -1", str(cm.exception))
             self.assertIn("got -2", str(cm.exception))
 
     def test_combined_logprobs_and_prompt_logprobs(self):
@@ -234,7 +233,7 @@ def test_error_message_formatting(self):
                 params._verify_args()
 
             error_msg = str(cm.exception)
-            self.assertIn("logprobs must be greater than -1", error_msg)
+            self.assertIn("logprobs must be a non-negative value or -1", error_msg)
             self.assertIn("got -5", error_msg)
 
         # Test logprobs error message when FD_USE_GET_SAVE_OUTPUT_V1 is "0"
@@ -244,8 +243,7 @@ def test_error_message_formatting(self):
                 params._verify_args()
 
             error_msg = str(cm.exception)
-            self.assertIn("logprobs must be greater than 0", error_msg)
-            self.assertIn("got -1", error_msg)
+            self.assertIn("Invalid value for 'top_logprobs': must be between 0 and 20", error_msg)
 
         # Test prompt_logprobs error message when FD_USE_GET_SAVE_OUTPUT_V1 is "1"
         with patch.dict(os.environ, {"FD_USE_GET_SAVE_OUTPUT_V1": "1"}):
@@ -254,7 +252,7 @@ def test_error_message_formatting(self):
                 params._verify_args()
 
             error_msg = str(cm.exception)
-            self.assertIn("prompt_logprobs must be greater than or equal to -1", error_msg)
+            self.assertIn("prompt_logprobs a must be non-negative value or -1", error_msg)
             self.assertIn("got -10", error_msg)
 
         # Test prompt_logprobs not supported error message when FD_USE_GET_SAVE_OUTPUT_V1 is "0"

tests/entrypoints/test_engine_client.py

@@ -133,6 +133,7 @@ def test_max_logprobs_invalid_values(self):
 
         self.assertIn("max_logprobs", str(context.exception))
         self.assertIn("must be >= -1", str(context.exception))
+        self.assertIn("got -2", str(context.exception))
 
     def test_max_logprobs_exceeds_vocab_size(self):
         """Test max_logprobs exceeding vocab_size"""
@@ -146,7 +147,7 @@ def test_max_logprobs_exceeds_vocab_size(self):
         self.assertIn("max_logprobs", str(context.exception))
         self.assertIn("must be <= vocab_size", str(context.exception))
         self.assertIn("1000", str(context.exception))
-        self.assertIn("1500", str(context.exception))
+        self.assertIn("got 1500", str(context.exception))
 
     def test_max_logprobs_unlimited(self):
         """Test max_logprobs = -1 (unlimited) sets to ori_vocab_size"""
@@ -237,7 +238,8 @@ def test_prompt_logprobs_invalid_values(self):
                 self.engine_client.valid_parameters(data)
 
             self.assertIn("prompt_logprobs", str(context.exception))
-            self.assertIn("must be >= -1", str(context.exception))
+            self.assertIn("must be a non-negative value or -1", str(context.exception))
+            self.assertIn("current value is -2", str(context.exception))
 
     def test_prompt_logprobs_exceeds_max_logprobs(self):
         """Test prompt_logprobs exceeding max_logprobs"""
@@ -252,6 +254,8 @@ def test_prompt_logprobs_exceeds_max_logprobs(self):
 
             self.assertIn("prompt_logprobs", str(context.exception))
             self.assertIn("exceeds maximum allowed value", str(context.exception))
+            self.assertIn("15", str(context.exception))
+            self.assertIn("10", str(context.exception))
 
     def test_top_logprobs_validation_with_fd_use_get_save_output_v1_enabled(self):
         """Test top_logprobs validation when FD_USE_GET_SAVE_OUTPUT_V1 is enabled"""
@@ -275,7 +279,8 @@ def test_top_logprobs_validation_with_fd_use_get_save_output_v1_enabled(self):
             data = {"logprobs": True, "top_logprobs": -2, "request_id": "test"}
             with self.assertRaises(ValueError) as context:
                 self.engine_client.valid_parameters(data)
-            self.assertIn("must be >= -1", str(context.exception))
+            self.assertIn("must be a non-negative value or -1", str(context.exception))
+            self.assertIn("current value is -2", str(context.exception))
 
             # Test value exceeding max_logprobs - should raise ValueError
             data = {"logprobs": True, "top_logprobs": 25, "request_id": "test"}
@@ -293,13 +298,15 @@ def test_top_logprobs_validation_with_fd_use_get_save_output_v1_disabled(self):
             data = {"logprobs": True, "top_logprobs": -1, "request_id": "test"}
             with self.assertRaises(ValueError) as context:
                 self.engine_client.valid_parameters(data)
-            self.assertIn("must be >= 0", str(context.exception))
+            self.assertIn("top_logprobs must be between 0 and 20", str(context.exception))
+            self.assertIn("current value is -1", str(context.exception))
 
             # Test value > 20 - should raise ValueError
             data = {"logprobs": True, "top_logprobs": 25, "request_id": "test"}
             with self.assertRaises(ValueError) as context:
                 self.engine_client.valid_parameters(data)
-            self.assertIn("must be <= 20", str(context.exception))
+            self.assertIn("top_logprobs must be between 0 and 20", str(context.exception))
+            self.assertIn("current value is 25", str(context.exception))
 
             # Test valid value
             data = {"logprobs": True, "top_logprobs": 10, "request_id": "test"}