[NewFeture]add ep rollout model init and update/clear ep buffer (#4039)

* fix gid

* merge

* fix test

* fix bug

* fix

* fix ci
This commit is contained in:
gaoziyuan
2025-09-17 20:24:53 +08:00
committed by GitHub
parent 0d3a57a2c6
commit 896e3bb606
12 changed files with 348 additions and 293 deletions

View File

@@ -5,6 +5,7 @@ from fastdeploy.config import (
CacheConfig,
FDConfig,
GraphOptimizationConfig,
LoadConfig,
ParallelConfig,
SchedulerConfig,
)
@@ -15,10 +16,12 @@ class TestConfig(unittest.TestCase):
parallel_config = ParallelConfig({"tensor_parallel_size": 16, "expert_parallel_size": 1})
graph_opt_config = GraphOptimizationConfig({})
cache_config = CacheConfig({})
load_config = LoadConfig({})
scheduler_config = SchedulerConfig({})
fd_config = FDConfig(
parallel_config=parallel_config,
graph_opt_config=graph_opt_config,
load_config=load_config,
cache_config=cache_config,
scheduler_config=scheduler_config,
ips=["1.1.1.1", "0.0.0.0"],
@@ -31,10 +34,12 @@ class TestConfig(unittest.TestCase):
parallel_config = ParallelConfig({})
graph_opt_config = GraphOptimizationConfig({})
cache_config = CacheConfig({})
load_config = LoadConfig({})
scheduler_config = SchedulerConfig({})
fd_config = FDConfig(
parallel_config=parallel_config,
graph_opt_config=graph_opt_config,
load_config=load_config,
cache_config=cache_config,
scheduler_config=scheduler_config,
ips="0.0.0.0",
@@ -46,12 +51,14 @@ class TestConfig(unittest.TestCase):
parallel_config = ParallelConfig({})
graph_opt_config = GraphOptimizationConfig({})
cache_config = CacheConfig({})
load_config = LoadConfig({})
cache_config.enable_chunked_prefill = True
scheduler_config = SchedulerConfig({})
fd_config = FDConfig(
parallel_config=parallel_config,
graph_opt_config=graph_opt_config,
cache_config=cache_config,
load_config=load_config,
scheduler_config=scheduler_config,
ips="0.0.0.0",
test_mode=True,
@@ -64,6 +71,7 @@ class TestConfig(unittest.TestCase):
parallel_config=parallel_config,
graph_opt_config=graph_opt_config,
cache_config=cache_config,
load_config=load_config,
scheduler_config=scheduler_config,
ips="0.0.0.0",
test_mode=True,
@@ -77,11 +85,13 @@ class TestConfig(unittest.TestCase):
cache_config = CacheConfig({})
cache_config.cache_transfer_protocol = "rdma,ipc"
cache_config.pd_comm_port = "2334"
load_config = LoadConfig({})
scheduler_config = SchedulerConfig({})
fd_config = FDConfig(
parallel_config=parallel_config,
graph_opt_config=graph_opt_config,
cache_config=cache_config,
load_config=load_config,
scheduler_config=scheduler_config,
splitwise_role="prefill",
test_mode=True,