Fix decode OOM caused by retraction (#14939)

This commit is contained in:
Liangsheng Yin
2025-12-13 13:59:17 +09:00
committed by GitHub
parent 8698867479
commit 01e3b3f3a3
4 changed files with 10 additions and 4 deletions

View File

@@ -1666,7 +1666,9 @@ class ModelRunner:
self.max_total_num_tokens = self.profile_max_num_token(total_gpu_memory)
if (small_kv_size := envs.SGLANG_CI_SMALL_KV_SIZE.get()) > 0:
# Use a small KV cache pool size for local tests
logger.info(
f"Use a small KV cache pool size ({small_kv_size}) for local tests"
)
self.max_total_num_tokens = small_kv_size
if max_num_reqs is None: