Apply suggestions from code review

Co-authored-by: OlivierDehaene <[email protected]>
huggingface · Dec 11, 2023 · 0006fab · 0006fab
1 parent b6519b5
commit 0006fab
Show file tree

Hide file tree

Showing 2 changed files with 0 additions and 5 deletions.
diff --git a/server/text_generation_server/models/flash_causal_lm.py b/server/text_generation_server/models/flash_causal_lm.py
@@ -422,8 +422,6 @@ def filter(self, request_ids: List[int]) -> "FlashCausalLMBatch":
         block_tables_tensor = self.block_tables_tensor[indices]
         input_lengths_tensor = self.input_lengths_tensor[indices]
         slots = self.slots[slot_filtering_indices]
-        if slot_indices.max().item() > slots.shape[0]:
-            import ipdb;ipdb.set_trace()
         next_token_chooser = self.next_token_chooser.filter(indices)
         top_n_tokens_tensor = self.top_n_tokens_tensor[indices]
         speculative_ids = self.speculative_ids[indices] if self.speculative_ids is not None else None

diff --git a/server/text_generation_server/utils/layers.py b/server/text_generation_server/utils/layers.py
@@ -742,9 +742,6 @@ def get_cos_sin(
 
             self._update_cos_sin_cache(dtype, position_ids.device, max_s)
 
-            if position_ids.max().item() >= max_s:
-                import ipdb;ipdb.set_trace()
-
             cos = torch.index_select(self._cos_cached, 0, position_ids)
             sin = torch.index_select(self._sin_cached, 0, position_ids)
             # Note: this unsqueeze is not necessary on RoCm + VLLM ROPE implementation, but we leave it as is to avoid yet an other controlflow.