chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Re-org eagle unit tests (#14909)

Browse Source
This commit is contained in:
Liangsheng Yin
2025-12-12 12:25:39 +09:00
committed by GitHub
parent 10146af099
commit c660d8dfd0
5 changed files with 155 additions and 229 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
python/sglang/srt/model_executor/model_runner.py
View File

@@ -237,9 +237,6 @@ def add_chunked_prefix_cache_attention_backend(backend_name):
)
# Use a small KV cache pool size for tests in CI
SGLANG_CI_SMALL_KV_SIZE = os.getenv("SGLANG_CI_SMALL_KV_SIZE", None)
# Detect stragger ranks in model loading
UNBALANCED_MODEL_LOADING_TIMEOUT_S = 480 # leave more time for post data processing
@@ -1716,8 +1713,10 @@ class ModelRunner:
log_info_on_rank0(logger, f"Using KV cache dtype: {self.kv_cache_dtype}")
self.max_total_num_tokens = self.profile_max_num_token(total_gpu_memory)
if SGLANG_CI_SMALL_KV_SIZE:
self.max_total_num_tokens = int(SGLANG_CI_SMALL_KV_SIZE)
if (small_kv_size := envs.SGLANG_CI_SMALL_KV_SIZE.get()) > 0:
# Use a small KV cache pool size for local tests
self.max_total_num_tokens = small_kv_size
if max_num_reqs is None:
max_num_reqs = min(