remove unused kernel in pytorch engine (#1237)

lmdeploy/pytorch/engine/engine.py

@@ -834,6 +834,20 @@ async def _add_messages(session_ids, token_ids):
         output_token_len = [len(token_ids) for token_ids in output_token_ids]
         return (status, output_token_ids, output_token_len)
 
+    def batched_infer(self,
+                      session_ids: List[int],
+                      token_ids: List[List[int]] = None,
+                      gen_config: EngineGenerationConfig = None,
+                      adapter_names: List[str] = None,
+                      keep_cache: bool = False):
+        """batched infer."""
+        coro = self.async_batched_infer(session_ids,
+                                        token_ids,
+                                        gen_config=gen_config,
+                                        adapter_names=adapter_names,
+                                        keep_cache=keep_cache)
+        return self.req_sender.run_until_complete(coro)
+
     def decode(self,
                input_ids,
                steps: List[int] = None,

lmdeploy/pytorch/kernels/__init__.py

@@ -1,18 +1,15 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 from .alibi_pagedattention import alibi_paged_attention_fwd
 from .apply_rotary_pos_emb import apply_rotary_pos_emb
-from .biased_pagedattention import biased_paged_attention_fwd
 from .fill_kv_cache import fill_kv_cache
-from .flashattention_nopad import context_attention_fwd
 from .fused_rotary_emb import fused_rotary_emb
 from .multinomial_sampling import multinomial_sampling
 from .pagedattention import paged_attention_fwd
 from .rerope_attention import rerope_attention_fwd
 from .rms_norm import rms_norm
 
 __all__ = [
-    'apply_rotary_pos_emb', 'context_attention_fwd', 'fused_rotary_emb',
-    'paged_attention_fwd', 'biased_paged_attention_fwd',
+    'apply_rotary_pos_emb', 'fused_rotary_emb', 'paged_attention_fwd',
     'alibi_paged_attention_fwd', 'fill_kv_cache', 'multinomial_sampling',
     'rms_norm', 'rerope_attention_fwd'
 ]