fix(api): validate and normalize request seeds

lightllm/server/api_models.py

@@ -6,6 +6,8 @@
 from typing import Any, Dict, List, Optional, Union, Literal, ClassVar
 from transformers import GenerationConfig
 
+MAX_SEED = (1 << 63) - 1
+
 
 class ImageURL(BaseModel):
     url: str
@@ -151,7 +153,7 @@ class CompletionRequest(BaseModel):
     top_k: Optional[int] = -1
     repetition_penalty: Optional[float] = 1.0
     ignore_eos: Optional[bool] = False
-    seed: Optional[int] = -1
+    seed: Optional[int] = Field(default=None, ge=-1, le=MAX_SEED)
 
     # Class variables to store loaded default values
     _loaded_defaults: ClassVar[Dict[str, Any]] = {}
@@ -231,7 +233,7 @@ class ChatCompletionRequest(BaseModel):
     top_k: Optional[int] = -1
     repetition_penalty: Optional[float] = 1.0
     ignore_eos: Optional[bool] = False
-    seed: Optional[int] = -1
+    seed: Optional[int] = Field(default=None, ge=-1, le=MAX_SEED)
     role_settings: Optional[Dict[str, str]] = None
     character_settings: Optional[List[Dict[str, str]]] = None
 

lightllm/server/core/objs/py_sampling_params.py

@@ -6,6 +6,7 @@
 from typing import List, Optional, Union, Tuple
 from transformers import GenerationConfig
 from lightllm.server.req_id_generator import MAX_BEST_OF
+from .sampling_params import normalize_seed, validate_seed
 
 
 _SAMPLING_EPS = 1e-5
@@ -92,7 +93,7 @@ def __init__(
         self.invalid_token_ids = invalid_token_ids
         self.group_request_id = group_request_id
         self.suggested_dp_index = suggested_dp_index
-        self.seed = seed
+        self.seed = normalize_seed(seed)
         if self.do_sample is False:
             self.temperature = 1.0
             self.top_p = 1.0
@@ -154,6 +155,7 @@ def verify(self):
             raise ValueError(
                 f"min_new_tokens must <= max_new_tokens, but got min {self.min_new_tokens}, max {self.max_new_tokens}."
             )
+        validate_seed(self.seed)
 
         if len(self.exponential_decay_length_penalty) != 2:
             raise ValueError(

lightllm/server/core/objs/sampling_params.py

@@ -20,6 +20,16 @@
 JSON_SCHEMA_MAX_LENGTH = int(os.getenv("LIGHTLLM_JSON_SCHEMA_MAX_LENGTH", 2048))
 INVALID_TOKEN_IDS_MAX_LENGTH = int(os.getenv("LIGHTLLM_INVALID_TOKEN_IDS_MAX_LENGTH", 10))
 MAX_PROMPT_LOGPROBS = int(os.getenv("LIGHTLLM_MAX_PROMPT_LOGPROBS", 1024))
+MAX_SEED = (1 << 63) - 1
+
+
+def normalize_seed(seed: Optional[int]) -> int:
+    return -1 if seed is None else seed
+
+
+def validate_seed(seed: int) -> None:
+    if seed < -1 or seed > MAX_SEED:
+        raise ValueError(f"seed must be -1 (random), or an integer in [0, {MAX_SEED}], got {seed}")
 
 
 class StopSequence(ctypes.Structure):
@@ -344,7 +354,11 @@ def init(self, tokenizer, **kwargs):
         self.add_special_tokens = kwargs.get("add_special_tokens", True)
         self.add_spaces_between_special_tokens = kwargs.get("add_spaces_between_special_tokens", True)
         self.print_eos_token = kwargs.get("print_eos_token", False)
-        self.seed = kwargs.get("seed", -1)
+        seed = normalize_seed(kwargs.get("seed"))
+        # ctypes silently wraps out-of-range integers, so validate the Python
+        # value before assigning it to the c_int64 field.
+        validate_seed(seed)
+        self.seed = seed
         prompt_logprobs = kwargs.get("prompt_logprobs", None)
         self.prompt_logprobs = -1 if prompt_logprobs is None else int(prompt_logprobs)
 
@@ -448,6 +462,7 @@ def verify(self):
             raise ValueError(f"prompt_logprobs must be in [-1, {MAX_PROMPT_LOGPROBS}], got {self.prompt_logprobs}")
         if self.prompt_logprobs >= 0 and not get_env_start_args().enable_prompt_logprobs:
             raise ValueError("prompt_logprobs requires --enable_prompt_logprobs")
+        validate_seed(self.seed)
         self._verify_allowed_token_ids()
         self._verify_grammar_constraint()