Minor fix in compiler & format (#545)

python/sglang/srt/managers/controller/infer_batch.py

@@ -498,9 +498,10 @@ class Batch:
                         req.output_ids = cur_output_ids
                         continue
 
-                    jump_forward_str, next_state = (
-                        req.jump_forward_map.jump_forward_symbol(cur_state)
-                    )
+                    (
+                        jump_forward_str,
+                        next_state,
+                    ) = req.jump_forward_map.jump_forward_symbol(cur_state)
 
                     # Make the incrementally decoded text part of jump_forward_str
                     # so that the UTF-8 will not corrupt

python/sglang/srt/managers/controller/tp_worker.py

@@ -283,13 +283,14 @@ class ModelTpServer:
                 (recv_req.image_hash >> 64) % self.model_config.vocab_size,
             ]
             req.image_size = recv_req.image_size
-            req.origin_input_ids, req.image_offset = (
-                self.model_runner.model.pad_input_ids(
-                    req.origin_input_ids_unpadded,
-                    req.pad_value,
-                    req.pixel_values.shape,
-                    req.image_size,
-                )
+            (
+                req.origin_input_ids,
+                req.image_offset,
+            ) = self.model_runner.model.pad_input_ids(
+                req.origin_input_ids_unpadded,
+                req.pad_value,
+                req.pixel_values.shape,
+                req.image_size,
             )
         req.sampling_params = recv_req.sampling_params
         req.return_logprob = recv_req.return_logprob

python/sglang/srt/managers/io_struct.py

@@ -35,7 +35,6 @@ class GenerateReqInput:
     stream: bool = False
 
     def post_init(self):
-
         if (self.text is None and self.input_ids is None) or (
             self.text is not None and self.input_ids is not None
         ):

python/sglang/srt/managers/tokenizer_manager.py

@@ -334,15 +334,15 @@ class TokenizerManager:
                 ret["meta_info"]["decode_token_logprobs"], return_text_in_logprobs
             )
         if top_logprobs_num > 0:
-            ret["meta_info"]["prefill_top_logprobs"] = (
-                self.detokenize_top_logprobs_tokens(
-                    ret["meta_info"]["prefill_top_logprobs"], return_text_in_logprobs
-                )
+            ret["meta_info"][
+                "prefill_top_logprobs"
+            ] = self.detokenize_top_logprobs_tokens(
+                ret["meta_info"]["prefill_top_logprobs"], return_text_in_logprobs
             )
-            ret["meta_info"]["decode_top_logprobs"] = (
-                self.detokenize_top_logprobs_tokens(
-                    ret["meta_info"]["decode_top_logprobs"], return_text_in_logprobs
-                )
+            ret["meta_info"][
+                "decode_top_logprobs"
+            ] = self.detokenize_top_logprobs_tokens(
+                ret["meta_info"]["decode_top_logprobs"], return_text_in_logprobs
             )
         return ret
 

python/sglang/srt/models/chatglm.py

@@ -36,7 +36,6 @@ LoraConfig = None
 
 
 class GLMAttention(nn.Module):
-
     def __init__(
         self,
         config,
@@ -294,7 +293,6 @@ class GLMTransformer(nn.Module):
 
 
 class ChatGLMModel(nn.Module):
-
     def __init__(
         self,
         config,

python/sglang/srt/models/grok.py

@@ -521,7 +521,6 @@ class Grok1DecoderLayer(nn.Module):
         hidden_states: torch.Tensor,
         input_metadata: InputMetadata,
     ) -> torch.Tensor:
-
         hidden_states = (
             self.post_attn_norm(
                 self.self_attn(

python/sglang/srt/models/llama2.py

@@ -160,9 +160,9 @@ class LlamaDecoderLayer(nn.Module):
         if rope_scaling is not None and getattr(
             config, "original_max_position_embeddings", None
         ):
-            rope_scaling["original_max_position_embeddings"] = (
-                config.original_max_position_embeddings
-            )
+            rope_scaling[
+                "original_max_position_embeddings"
+            ] = config.original_max_position_embeddings
         max_position_embeddings = getattr(config, "max_position_embeddings", 8192)
         self.self_attn = LlamaAttention(
             hidden_size=self.hidden_size,