[feat] add native kvcache offload

01267596 · 01267596 · commit 04d79299e369 · 2025-11-05T08:37:01.000Z
Signed-off-by: 01267596 &lt;xiongkai123@cmbchina.com&gt;
diff --git a/vllm_ascend/spec_decode/draft_proposer.py b/vllm_ascend/spec_decode/draft_proposer.py
@@ -254,7 +254,6 @@ def merge_next_token_ids_into_token_ids(
     new_token_ids = extend_flat_seqs(
         seqs=input_token_ids, end_locs=query_end_locs, new_vals=next_token_ids
     )
-    logger.warning("new_token_ids: {}".format(new_token_ids))
 
     # append new positions
     positions_to_append = input_positions[query_end_locs] + 1

Original file line number	Diff line number	Diff line change
`@@ -254,7 +254,6 @@ def merge_next_token_ids_into_token_ids(`
`254`	`254`	`new_token_ids = extend_flat_seqs(`
`255`	`255`	`seqs=input_token_ids, end_locs=query_end_locs, new_vals=next_token_ids`
`256`	`256`	`)`
`257`		`- logger.warning("new_token_ids: {}".format(new_token_ids))`
`258`	`257`
`259`	`258`	`# append new positions`
`260`	`259`	`positions_to_append = input_positions[query_end_locs] + 1`