We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
1 parent ff1dfea commit 8c8a299Copy full SHA for 8c8a299
vllm/distributed/kv_transfer/kv_connector/simple_connector.py
@@ -215,9 +215,8 @@ def send_kv_caches_and_hidden_states(
215
values = torch.empty_like(keys)
216
217
# Copy the relevant parts from kvcache
218
- for layer_idx, layer_id in enumerate(range(start_layer,
219
- end_layer)):
220
- kv_cache = kv_caches[layer_id - start_layer]
+ for layer_idx in range(end_layer - start_layer):
+ kv_cache = kv_caches[layer_idx]
221
222
if self.is_deepseek_mla and self.use_mla_opt:
223
key_cache = kv_cache.reshape(-1, num_heads, head_size)
0 commit comments