misc: add pre-commit config (#637)

python/sglang/srt/models/llama2.py

@@ -163,9 +163,9 @@ 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
+            )
         rope_is_neox_style = getattr(config, "rope_is_neox_style", True)
         max_position_embeddings = getattr(config, "max_position_embeddings", 8192)
         self.self_attn = LlamaAttention(

python/sglang/srt/models/qwen2.py

@@ -313,7 +313,10 @@ class Qwen2ForCausalLM(nn.Module):
                 param = params_dict[name]
                 weight_loader = getattr(param, "weight_loader", default_weight_loader)
                 weight_loader(param, loaded_weight)
-                if self.config.tie_word_embeddings and name=="model.embed_tokens.weight":
+                if (
+                    self.config.tie_word_embeddings
+                    and name == "model.embed_tokens.weight"
+                ):
                     weight_loader(params_dict["lm_head.weight"], loaded_weight)
 
 

python/sglang/srt/models/qwen2_moe.py

@@ -401,9 +401,11 @@ class Qwen2MoeForCausalLM(nn.Module):
             # These are the weights for the experts
             # (param_name, weight_name, expert_id, shard_id)
             (
-                "experts.w13_weight"
-                if weight_name in ["gate_proj", "up_proj"]
-                else "experts.w2_weight",
+                (
+                    "experts.w13_weight"
+                    if weight_name in ["gate_proj", "up_proj"]
+                    else "experts.w2_weight"
+                ),
                 f"experts.{expert_id}.{weight_name}.weight",
                 expert_id,
                 shard_id,
@@ -418,7 +420,7 @@ class Qwen2MoeForCausalLM(nn.Module):
         for name, loaded_weight in weights:
             if "rotary_emb.inv_freq" in name:
                 continue
-            for (param_name, weight_name, shard_id) in stacked_params_mapping:
+            for param_name, weight_name, shard_id in stacked_params_mapping:
                 # Skip non-stacked layers and experts (experts handled below).
                 if weight_name not in name:
                     continue