メニュー
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.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
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: