AGENT SKILL

sdk-client

Use LeptonAI's Python SDK Client and v2 API resource/model objects safely from scripts and notebooks.

License 许可
Apache-2.0
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
sdk-client
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Use LeptonAI's Python SDK Client and v2 API resource/model objects safely from scripts and notebooks.
disable-model-invocation
true

02 资源文件 6

03 SKILL.md 正文

sdk-client

Use this sub-skill when the task is to call a deployed Lepton endpoint from Python, inspect endpoint OpenAPI-derived methods, or build SDK scripts/notebooks with leptonai.api.v2 resource objects and Pydantic models.

Route here for

  • Client(workspace_or_url, deployment=None, token=None, stream=None, chunk_size=None, timeout=None, no_check=False, http2=True) endpoint-call code.
  • local(port), Client.local(port), current(), and Client.current() usage.
  • Dynamic PathTree behavior for OpenAPI paths exposed as Python attributes.
  • APIClient(...) workspace API resource objects and their method surfaces.
  • APIResourse.safe_json(...) model serialization and v2 Pydantic spec construction.
  • Pure parsing helpers make_mounts_from_strings(...) and make_env_vars_from_strings(...).

Route elsewhere

  • Workspace login, workspace record maintenance, token refresh, or auth setup: use the workspace-and-auth route.
  • Shell-first workload recipes such as endpoint/job/pod/Ray/finetune create/list/update/delete: use the workload-management route.
  • Storage, secrets, ingress, or canary routing plans: use the storage/ingress route even when SDK model helpers appear in the plan.

Safety rule

Creating a Client(...) for a real URL or workspace endpoint performs network checks and tries to fetch openapi.json. APIClient resource methods perform authenticated workspace API requests. Do not run live or credentialed cloud calls unless the user has explicitly authorized the operation and supplied an appropriate workspace context. The bundled scripts in this sub-skill are designed to avoid network calls.

Quick starts

Inspect the installed SDK without network calls

python scripts/sdk_introspection.py
python scripts/client_path_tree_demo.py

The first script prints the installed package version, important signatures, and APIClient resource method surfaces without constructing a live workspace client. The second script demonstrates PathTree and local(...) behavior with fake callables only.

Call a deployed endpoint from Python

import os
from leptonai.client import Client, local

# Local endpoint URL plan; constructing the client contacts the local service.
c = Client(local(port=8080), timeout=60, no_check=True)
print(list(c.paths()))

# Workspace endpoint URL plan; pass an auth token explicitly if the endpoint is private.
remote = Client(
    "workspace-id",
    "endpoint-name",
    token=os.environ.get("LEPTON_WORKSPACE_TOKEN"),
    timeout=None,
)
# OpenAPI path /run becomes remote.run(...)
result = remote.run(inputs="hello")

For dynamic path rules, streaming return behavior, and failure modes, read client-api.md.

Build v2 specs without cloud calls

from leptonai.api.v2.api_resource import APIResourse
from leptonai.api.v2.spec_utils import make_env_vars_from_strings, make_mounts_from_strings
from leptonai.api.v2.types.common import Metadata
from leptonai.api.v2.types.deployment import LeptonDeployment, LeptonDeploymentUserSpec

spec = LeptonDeployment(
    metadata=Metadata(name="example"),
    spec=LeptonDeploymentUserSpec(
        envs=make_env_vars_from_strings(["MODE=test"], ["API_SECRET"]),
        mounts=make_mounts_from_strings(["/data:/mnt/data:node-local"]),
    ),
)
payload = APIResourse.safe_json(APIResourse.__new__(APIResourse), spec)

For resource surfaces, serialization caveats, and spec helpers, read api-resources.md and specs-and-models.md.

Troubleshooting index

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351