Exécutez n'importe quel Skill dans Manus
en un clic

Exécutez n'importe quel Skill dans Manus en un clic

helios-native-mlx-ab-profiling

Étoiles2

Forks0

Mis à jour30 juin 2026 à 22:59

Profile and optimize Helios native MLX/C++ execution paths with A/B evidence, memory gates, and feature-flagged experimental kernels.

Installation

Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.

Exécuter dans Manus

Source

jscott3201

jscott3201/Helios

Ouvrir le dépôt GitHub Voir les dépôts du créateur

Téléchargement

Exécuter dans Manus

Métiers associésSOC

Basé sur la classification professionnelle SOC

Développeurs de logicielsProfessions informatiques et mathématiques·SOC 15-1252

SKILL.md

readonly

name	helios-native-mlx-ab-profiling
description	Profile and optimize Helios native MLX/C++ execution paths with A/B evidence, memory gates, and feature-flagged experimental kernels.

Helios native MLX A/B profiling skill

Use this skill for native graph, MLX, Metal, KV cache, compressed attention, LoRA hot path, and prefill/decode optimization.

Rules

Establish baseline records before changing hot-path code.
Separate model load, prefill, decode, eval/synchronization, cache import/export, and server overhead.
Keep unsafe/FFI changes narrow and tested.
Avoid default-on experimental kernels.
Record p50, p95, p99, peak MLX memory, active KV bytes, and RSS.
Treat no-go reports as valid outcomes for high-risk ideas.

Red flags

Speedup measured only on repeated-token prompts.
Active memory reduction claimed when compressed payload is decompressed into BF16 active state.
MTP speedup claimed without acceptance and exactness evidence.
Server speedup measured through stub metrics.

Plus depuis ce dépôt

même dépôt

helios-mtp-acceptance-analysis

jscott3201/Helios

Diagnose Gemma 4 MTP acceptance and speculative decoding behavior in Helios using real-context prompts, trace records, and exactness gates.

2026-06-302

helios-real-workload-ab

jscott3201/Helios

Build and run Helios real-context A/B benchmarks with deterministic workload manifests, correctness gates, and decision artifacts.

2026-06-302

gemma4d-16gb-mac-profiling

jscott3201/Helios

Use for tiny16 MacBook profiling: memory pressure, context-length limits, TTFT/decode metrics, cache effectiveness, MTP acceptance, and release-gate benchmark reports.

2026-06-302

gemma4d-dynamic-adapters

jscott3201/Helios

Use for dynamic LoRA/QLoRA adapter import, validation, loading, routing, adapter-aware KV cache keys, aLoRA planning, and adapter memory residency.

2026-06-302

gemma4d-kv-cache-offload

jscott3201/Helios

Use for Gemma4D KV cache, prefix cache, SSD cold tier, cache manifests, cache-key correctness, quantized/compressed KV experiments, and restore parity.

2026-06-302

gemma4d-milestone-execution

jscott3201/Helios

Use to execute one Gemma4D milestone end-to-end with scoped tasks, verification commands, evidence capture, and blocker reporting. Trigger on milestone implementation, Codex Goals, task planning, or release gates.

2026-06-302

name	helios-native-mlx-ab-profiling
description	Profile and optimize Helios native MLX/C++ execution paths with A/B evidence, memory gates, and feature-flagged experimental kernels.

Helios native MLX A/B profiling skill

Use this skill for native graph, MLX, Metal, KV cache, compressed attention, LoRA hot path, and prefill/decode optimization.

Rules

Establish baseline records before changing hot-path code.
Separate model load, prefill, decode, eval/synchronization, cache import/export, and server overhead.
Keep unsafe/FFI changes narrow and tested.
Avoid default-on experimental kernels.
Record p50, p95, p99, peak MLX memory, active KV bytes, and RSS.
Treat no-go reports as valid outcomes for high-risk ideas.

Red flags

Speedup measured only on repeated-token prompts.
Active memory reduction claimed when compressed payload is decompressed into BF16 active state.
MTP speedup claimed without acceptance and exactness evidence.
Server speedup measured through stub metrics.