[Feature] Speculative decoding support lookahead (#9873)

Co-authored-by: a4zhangfei <a4zhangfei@qq.com>
Co-authored-by: Qiaolin-Yu <liin1211@outlook.com>
This commit is contained in:
Zhihao Zhang
2025-09-18 16:42:41 -07:00
committed by GitHub
co-authored by a4zhangfei Qiaolin-Yu
parent 2a2ff9a840
commit e7bc600304
30 changed files with 2058 additions and 32 deletions
+1
View File
@@ -80,6 +80,7 @@ DEFAULT_STANDALONE_SPECULATIVE_TARGET_MODEL_FOR_TEST = (
"meta-llama/Llama-3.1-8B-Instruct"
)
DEFAULT_STANDALONE_SPECULATIVE_DRAFT_MODEL_FOR_TEST = "meta-llama/Llama-3.2-1B-Instruct"
DEFAULT_LOOKAHEAD_SPECULATIVE_TARGET_MODEL_FOR_TEST = "Qwen/Qwen2.5-Coder-7B-Instruct"
# Other use cases
DEFAULT_MODEL_NAME_FOR_TEST_LOCAL_ATTENTION = (