fix : FSDP 优化器顺序、温度除零、调度器静默死亡、ref模型设备

- executor: use_orig_params 硬编码 True，FSDP 不替换 Parameter 对象
- strategy: DPO/GRPO ref 模型创建后移到 device
- sample: TemperatureStrategy clamp 1e-8，engine 验证改为 >0
- scheduler: 异常不 re-raise 避免 daemon 静默死亡，stop() 发回调给 waiting 任务

astrai/inference/core/scheduler.py

@@ -71,6 +71,7 @@ class InferenceScheduler:
         )
 
         self._running = False
+        self._fatal_error: Optional[Exception] = None
 
     def add_task(self, prompt: str, **kwargs) -> str:
         return self._task_mgr.add_task(prompt, **kwargs)
@@ -175,6 +176,8 @@ class InferenceScheduler:
                                     t.stream_callback(STOP)
 
         except Exception as e:
+            self._fatal_error = e
+            self._running = False
             logger.error(f"Scheduler loop crashed: {e}", exc_info=True)
             for task in self._task_mgr.get_active_tasks():
                 if task.stream_callback:
@@ -184,7 +187,6 @@ class InferenceScheduler:
                 if task.stream_callback:
                     task.stream_callback(STOP)
             self._task_mgr.clear_queues()
-            raise
 
     def start(self):
         if not self._running:
@@ -199,7 +201,12 @@ class InferenceScheduler:
         if hasattr(self, "_loop_thread"):
             self._loop_thread.join(timeout=2.0)
         for task in self._task_mgr.get_active_tasks():
+            if task.stream_callback:
+                task.stream_callback(STOP)
             self._page_cache.task_free(task.task_id)
+        for task in self._task_mgr.get_waiting_tasks():
+            if task.stream_callback:
+                task.stream_callback(STOP)
         self._task_mgr.clear_queues()
         if torch.cuda.is_available():
             torch.cuda.empty_cache()

astrai/inference/engine.py

@@ -79,8 +79,8 @@ class GenerationRequest:
             raise ValueError("top_k must be a non-negative integer")
         if not (0.0 <= top_p <= 1.0):
             raise ValueError("top_p must be a float between 0.0 and 1.0")
-        if not (isinstance(temperature, (int, float)) and temperature >= 0):
+        if not (isinstance(temperature, (int, float)) and temperature > 0):
-            raise ValueError("temperature must be a non-negative number")
+            raise ValueError("temperature must be a positive number")
 
         self.messages = messages
         self.top_k = top_k

astrai/inference/sample.py

@@ -44,10 +44,12 @@ class TemperatureStrategy(BaseSamplingStrategy):
     def apply(self, logits, filter_value=-float("inf")):
         t = self.temperature
         if isinstance(t, Tensor):
+            t = t.to(logits.device, non_blocking=True).view(-1, 1)
+            t = torch.clamp(t, min=1e-8)
             if (t != 1.0).any():
-                logits = logits / t.to(logits.device, non_blocking=True).view(-1, 1)
-        elif t != 1.0:
                 logits = logits / t
+        elif t != 1.0:
+            logits = logits / max(t, 1e-8)
         return logits
 
 

astrai/parallel/executor.py

@@ -218,7 +218,6 @@ class FSDPExecutor(BaseExecutor):
         sync_module_states: bool = False,
         forward_prefetch: bool = False,
         limit_all_gathers: bool = True,
-        use_orig_params: bool = False,
         ignored_states=None,
         device_mesh=None,
     ):
@@ -237,7 +236,7 @@ class FSDPExecutor(BaseExecutor):
                 sync_module_states=sync_module_states,
                 forward_prefetch=forward_prefetch,
                 limit_all_gathers=limit_all_gathers,
-                use_orig_params=use_orig_params,
+                use_orig_params=True,
                 ignored_states=ignored_states,
                 device_mesh=device_mesh,
             ).items()

astrai/trainer/strategy.py

@@ -219,7 +219,7 @@ class DPOStrategy(BaseStrategy):
         super().__init__(model, device, **kwargs)
         self.ref_model = create_ref_model(
             self.model_fn, self.executor.unwrap_model(model)
-        )
+        ).to(device=self.device)
         self.beta = beta
         self.reduction = reduction
 
@@ -275,7 +275,7 @@ class GRPOStrategy(BaseStrategy):
         super().__init__(model, device, **kwargs)
         self.ref_model = create_ref_model(
             self.model_fn, self.executor.unwrap_model(model)
-        )
+        ).to(device=self.device)
         self.clip_eps = clip_eps
         self.kl_coef = kl_coef
         self.group_size = group_size