Rename InputMetadata -> ForwardBatch (#1543)

python/sglang/srt/models/minicpm.py

@@ -39,7 +39,7 @@ from sglang.srt.layers.linear import (
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.model_executor.forward_batch_info import InputMetadata
+from sglang.srt.model_executor.forward_batch_info import ForwardBatch
 
 
 class MiniCPMMLP(nn.Module):
@@ -148,7 +148,7 @@ class MiniCPMAttention(nn.Module):
         self,
         positions: torch.Tensor,
         hidden_states: torch.Tensor,
-        input_metadata: InputMetadata,
+        forward_batch: ForwardBatch,
     ) -> torch.Tensor:
         qkv, _ = self.qkv_proj(hidden_states)
         q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
@@ -156,7 +156,7 @@ class MiniCPMAttention(nn.Module):
         q, k = q.float(), k.float()
         q, k = self.rotary_emb(positions, q, k)
         q, k = q.to(orig_dtype), k.to(orig_dtype)
-        attn_output = self.attn(q, k, v, input_metadata)
+        attn_output = self.attn(q, k, v, forward_batch)
         output, _ = self.o_proj(attn_output)
         return output
 
@@ -199,7 +199,7 @@ class MiniCPMDecoderLayer(nn.Module):
         self,
         positions: torch.Tensor,
         hidden_states: torch.Tensor,
-        input_metadata: InputMetadata,
+        forward_batch: ForwardBatch,
         residual: Optional[torch.Tensor],
     ) -> Tuple[torch.Tensor, torch.Tensor]:
         # Self Attention
@@ -208,7 +208,7 @@ class MiniCPMDecoderLayer(nn.Module):
         hidden_states = self.self_attn(
             positions=positions,
             hidden_states=hidden_states,
-            input_metadata=input_metadata,
+            forward_batch=forward_batch,
         )
         hidden_states = residual + hidden_states * (
             self.config.scale_depth / math.sqrt(self.config.num_hidden_layers)
@@ -252,7 +252,7 @@ class MiniCPMModel(nn.Module):
         self,
         input_ids: torch.Tensor,
         positions: torch.Tensor,
-        input_metadata: InputMetadata,
+        forward_batch: ForwardBatch,
         input_embeds: torch.Tensor = None,
     ) -> torch.Tensor:
         if input_embeds is None:
@@ -266,7 +266,7 @@ class MiniCPMModel(nn.Module):
             hidden_states, residual = layer(
                 positions,
                 hidden_states,
-                input_metadata,
+                forward_batch,
                 residual,
             )
         hidden_states = self.norm(hidden_states)
@@ -303,19 +303,19 @@ class MiniCPMForCausalLM(nn.Module):
         self,
         input_ids: torch.Tensor,
         positions: torch.Tensor,
-        input_metadata: InputMetadata,
+        forward_batch: ForwardBatch,
         input_embeds: torch.Tensor = None,
     ) -> torch.Tensor:
         if input_embeds is not None:
             input_embeds = input_embeds * self.config.scale_emb
-        hidden_states = self.model(input_ids, positions, input_metadata, input_embeds)
+        hidden_states = self.model(input_ids, positions, forward_batch, input_embeds)
         hidden_states = hidden_states / self.scale_width
         if self.config.tie_word_embeddings:
             lm_head_weight = self.model.embed_tokens.weight
         else:
             lm_head_weight = self.lm_head.weight
         return self.logits_processor(
-            input_ids, hidden_states, lm_head_weight, input_metadata
+            input_ids, hidden_states, lm_head_weight, forward_batch
         )
 
     def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):