[Doc] Explicitly state that PP isn't compatible with speculative decoding yet (#10975)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

vllm/model_executor/models/granite.py

@@ -400,16 +400,17 @@ class GraniteForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
                 self.lm_head.weight = self.model.embed_tokens.weight
 
             logit_scale = getattr(config, "logit_scale", 1.0)
-
             if hasattr(config, "logits_scaling"):
                 logit_scale /= config.logits_scaling
+
             self.logits_processor = LogitsProcessor(self.unpadded_vocab_size,
                                                     config.vocab_size,
                                                     scale=logit_scale)
-            self.sampler = get_sampler()
         else:
             self.lm_head = PPMissingLayer()
 
+        self.sampler = get_sampler()
+
     def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.model.get_input_embeddings(input_ids)