القائمة
Use when modifying python/src/lib.rs (PyO3 bindings), python/pegaflow/connector/ (vLLM KV connector), python/pegaflow/sglang/ (SGLang radix cache), .pyi type stubs, or building with maturin.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Use when configuring, deploying, or troubleshooting pegaflow-server/pegaflow-metaserver. Covers CLI flags, pool sizing, SSD cache, NUMA, metrics, and multi-node setup.
Bump pegaflow-llm package version. Use when the user asks to bump version, release a new version, increment version, or update package version.
Debug vLLM async KV loading and PegaFlow connector behavior. Use when investigating async KV loads, WAITING_FOR_REMOTE_KVS states, load/save intents, connector metadata flow, prefetch behavior, or preemption interactions in vLLM scheduler/worker code.
| name | python-binding |
| description | Use when modifying python/src/lib.rs (PyO3 bindings), python/pegaflow/connector/ (vLLM KV connector), python/pegaflow/sglang/ (SGLang radix cache), .pyi type stubs, or building with maturin. |
cd python
maturin develop # Dev build
maturin develop --release # Release build
Important: When modifying python/src/lib.rs (PyO3 bindings), update the type stub file python/pegaflow/pegaflow.pyi to keep type hints in sync.
python/src/lib.rs: PyO3 bindings exposing PegaEngine and gRPC clientpython/pegaflow/pegaflow.pyi: Type stubs — must stay in sync with lib.rspython/pegaflow/connector/scheduler.py: vLLM scheduler-side connectorpython/pegaflow/connector/worker.py: vLLM worker-side connectorpython/pegaflow/sglang/pegaflow_radix_cache.py: SGLang radix cachepython/pegaflow/ipc_wrapper.py: CUDA IPC handle wrapperConfigure vLLM to use PegaFlow:
from vllm.distributed.kv_transfer.kv_transfer_agent import KVTransferConfig
kv_transfer_config = KVTransferConfig(
kv_connector="PegaKVConnector",
kv_role="kv_both",
kv_connector_module_path="pegaflow.connector",
)
Connector is split into scheduler-side (scheduler.py) and worker-side (worker.py).
Drop-in replacement for SGLang's RadixCache using PegaEngine for distributed KV cache.
from pegaflow.sglang.peagflow_radix_cache import PeagflowRadixCache
kv_cache = PeagflowRadixCache(
params=cache_params, # CacheInitParams
model_config=model_config, # ModelConfig
tp_size=tp_size,
rank=tp_rank,
)
Behavioral differences vs default RadixCache: