Add top-p to run_eval.py (#16844)

docs/basic_usage/deepseek_v32.md

@@ -190,15 +190,25 @@ Output throughput: 4418.617 token/s
 
 Accuracy benchmark on long context can be tested on GPQA-diamond dataset with long output tokens and thinking enabled:
 ```bash
-python3 -m sglang.test.run_eval --port 30000 --eval-name gpqa --num-examples 198 --max-tokens 120000 --repeat 8 --thinking-mode deepseek-v3
+python3 -m sglang.test.run_eval --port 30000 --eval-name gpqa --num-examples 198 --max-tokens 128000 --repeat 8 --thinking-mode deepseek-v3
 ```
 
-The mean accuracy over 8 runs shows 0.797, which matches the number 79.9 in official tech report.
+The mean accuracy over 8 runs shows 0.797, which matches the number 0.799 in official tech report.
 ```bash
 Repeat: 8, mean: 0.797
 Scores: ['0.808', '0.798', '0.808', '0.798', '0.783', '0.788', '0.803', '0.793']
 ```
 
+For Deepseek V3.2, Deepseek recommends setting the sampling parameters to temperature = 1.0, top_p = 0.95:
+
+```bash
+python3 -m sglang.test.run_eval --port 30000 --eval-name gpqa --num-examples 198 --max-tokens 128000 --repeat 8 --top-p 0.95 --temperature 1.0 --thinking-mode deepseek-v3
+
+Repeat: 8, mean: 0.840
+Scores: ['0.848', '0.808', '0.848', '0.838', '0.879', '0.813', '0.838', '0.848']
+```
+which matches the official score, 0.824, as reported in the [Deepseek-V3.2 technical report](https://huggingface.co/deepseek-ai/DeepSeek-V3.2/blob/main/assets/paper.pdf).
+
 ### Accuracy Test with `aime 2025`
 
 Prepare the environment by installing NeMo-Skills in the docker or your own virtual environment:

python/sglang/test/run_eval.py

@@ -36,6 +36,7 @@ def run_eval_once(args, base_url: str, eval_obj: Eval) -> dict:
     sampler = ChatCompletionSampler(
         model=args.model,
         max_tokens=getattr(args, "max_tokens", 2048),
+        top_p=getattr(args, "top_p", 1.0),
         base_url=base_url,
         temperature=getattr(args, "temperature", 0.0),
         reasoning_effort=getattr(args, "reasoning_effort", None),
@@ -233,6 +234,7 @@ if __name__ == "__main__":
     parser.add_argument("--num-threads", type=int, default=512)
     parser.add_argument("--max-tokens", type=int, default=2048)
     parser.add_argument("--temperature", type=float, default=0.0)
+    parser.add_argument("--top-p", type=float, default=1.0)
     parser.add_argument("--reasoning-effort", type=str)
     parser.add_argument(
         "--thinking-mode",

python/sglang/test/simple_eval_common.py

@@ -91,6 +91,7 @@ class ChatCompletionSampler(SamplerBase):
         model: Optional[str] = None,
         system_message: Optional[str] = None,
         temperature: float = 0.0,
+        top_p: float = 1.0,
         reasoning_effort: Optional[str] = None,
         max_tokens: int = 2048,
         extra_body: Optional[Dict[str, Any]] = None,
@@ -103,6 +104,7 @@ class ChatCompletionSampler(SamplerBase):
         self.model = model
         self.system_message = system_message
         self.temperature = temperature
+        self.top_p = top_p
         self.max_tokens = max_tokens
         self.reasoning_effort = reasoning_effort
         self.extra_body = extra_body
@@ -144,6 +146,7 @@ class ChatCompletionSampler(SamplerBase):
                     model=self.model,
                     messages=message_list,
                     temperature=self.temperature,
+                    top_p=self.top_p,
                     max_tokens=self.max_tokens,
                     reasoning_effort=self.reasoning_effort,
                     extra_body=self.extra_body,