EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[CI] fix port conflicts (#5789)

Browse Source
This commit is contained in:
Lianmin Zheng
2025-04-27 05:17:44 -07:00
committed by GitHub
parent 3c4e0ee64d
commit 35ca04d2fa
7 changed files with 55 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
test/srt/test_update_weights_from_distributed.py
View File

@@ -129,7 +129,7 @@ def init_process_hf(
hf_instruct_params = []
hf_base_params = []
print("get parameter in hf instruct model and base model")
print("[hf] get parameter in hf instruct model and base model")
for parameter_name in checking_parameters:
hf_instruct_params.append(
hf_instruct_model.get_parameter(parameter_name)[:truncate_size]
@@ -152,10 +152,12 @@ def init_process_hf(
param_queue.put(("hf_base_params", hf_base_params))
# Init weight update group for rank 0 (the training engine in RLHF).
print(f"rank {rank} world_size: {world_size} init custom process group")
port = 60000 + int(os.environ.get("CUDA_VISIBLE_DEVICES", "0")[0]) * 100
init_method = f"tcp://localhost:{port}"
print(f"[hf] {rank=} {world_size=} init custom process group. {init_method=}")
group = init_custom_process_group(
backend="nccl",
init_method="tcp://localhost:65500",
init_method=init_method,
world_size=world_size,
rank=rank,
group_name="test_parameter_update_group",
@@ -184,7 +186,7 @@ def init_process_hf(
# Measure the latency of broadcasting/weights update.
broadcast_time = time_end_broadcast - time_begin_broadcast
print(f"rank {rank} broadcast parameter time: {broadcast_time:.3f}s")
print(f"[hf] {rank=} {broadcast_time=:.3f}s")
param_queue.put(("broadcast_time", broadcast_time))
# Delete the huggingface models to free up memory.
@@ -210,17 +212,21 @@ def init_process_sgl(
torch.cuda.synchronize()
base_gpu_id = 1 if rank == 1 else 1 + tp_size
if backend == "Engine":
print(f"[sgl] rank {rank} init engine")
engine = sgl.Engine(
model_path=model_name,
random_seed=42,
base_gpu_id=base_gpu_id,
tp_size=tp_size,
cuda_graph_max_bs=2,
)
else:
if rank == 1:
url = DEFAULT_URL_FOR_TEST
else:
url = DEFAULT_URL_FOR_TEST.replace("2157", "2159")
host, port = DEFAULT_URL_FOR_TEST.split(":")
url = ":".join(host, str(int(port) + 10000))
print(f"[sgl] rank {rank} init server on url: {url}")
process = popen_launch_server(
model_name,
url,
@@ -230,13 +236,11 @@ def init_process_sgl(
str(base_gpu_id),
"--tp-size",
str(tp_size),
"--cuda-graph-max-bs",
2,
),
)
torch.cuda.synchronize()
if backend == "Engine":
print(f"rank {rank} init engine")
else:
print(f"rank {rank} init server on url: {url}")
# Get weights of instruct model, i.e. pre-training weights.
instruct_params = []
@@ -252,11 +256,13 @@ def init_process_sgl(
param_queue.put((f"sgl_dp_{rank}_instruct_params", instruct_params))
port = 60000 + int(os.environ.get("CUDA_VISIBLE_DEVICES", "0")[0]) * 100
# Init weight update group with the training engine.
if backend == "Engine":
engine.init_weights_update_group(
master_address="localhost",
master_port="65500",
master_port=str(port),
rank_offset=base_gpu_id,
world_size=world_size,
group_name="test_parameter_update_group",
@@ -267,7 +273,7 @@ def init_process_sgl(
f"{url}/init_weights_update_group",
json={
"master_address": "localhost",
"master_port": "65500",
"master_port": str(port),
"rank_offset": base_gpu_id,
"world_size": world_size,
"group_name": "test_parameter_update_group",
@@ -311,7 +317,7 @@ def init_process_sgl(
# Measure the latency of broadcast/weights update.
update_time = time_end_update - time_begin_update
print(
f"fully update model_name {model_name} rank {rank} parameter from distributed time: {update_time:.3f}s"
f"[sgl] fully update model_name {model_name} rank {rank} parameter from distributed time: {update_time:.3f}s"
)
param_queue.put((f"update_sgl_dp_{rank}_time", update_time))