Aevonix/FlashAlign

选用指南

AI 根据 README 整理 · 2026/11/1 04:57:12

使用方式

通过命令行安装和配置，支持多种模式（如 Verified contacts 和 Always On），并可通过 Python 客户端或自定义 harness 集成。

查看 README 依据

The wizard detects local models and harnesses, selects a matching profile, prepares the engine and creates a deployment signing key. ```sh git clone https://github.com/Aevonix/FlashAlign.git cd FlashAlign bash ./install.sh --auto ``` Choose **Verified contacts** or **Always On**. In Verified contacts mode, choose whether tools inherit the profile. Setup checks activation and backs up changed configuration. Chat integrations restart their selected gateways; local terminal setup prints its launch command. All five starter profiles are included. No profile download or private

Use `--model-path /path/to/model` to select a model. Run `bash ./install.sh` to connect to an engine that is already running. See the [engine selection guide](docs/engine-setup.md#engine-selection) for supported model and engine combinations. For standalone setup with a local Qwen checkpoint: ```sh bash ./install.sh --engine-only --backend vllm --model-path /models/qwen38-27b --yes ``` See [direct access](docs/direct-engine.md) for the Python client and [custom harnesses](docs/custom-harness.md) for integration hooks. To use profiles from another folder: ```sh

部署要求

需要 Linux 和 NVIDIA GPU，Python 3.10 或更高版本；支持特定引擎（如 TensorFold、vLLM）和模型。

查看 README 依据

## Compatibility | Component | Requirement | | --- | --- | | Inference host | Linux and NVIDIA GPU; matched CPU and GPU architecture for TensorFold bundles | | Python | 3.10 or later for setup and adapters | | Hermes | Native ingress, request and tool-execution hooks | | OpenClaw | Compatible native plugin API; Node 24.22.1+ on 24.x or 26.12.1+ on 26.x | | Model API | OpenAI-compatible Chat Completions and the FlashAlign engine contract | | Model profile | A matching profile loaded by the engine | See [compatibility](docs/compatibility.md) for channel, engine and delegation requirements.

已知限制

当前主要专注减少拒绝，未来可能增加增加拒绝的配置文件；实验性测试配置文件被标记为实验性，且加载新模型或配置文件可能需要重启引擎。

查看 README 依据

The V1 starter profiles focus on refusal reduction. Future profiles could also increase refusals for public or higher-risk tasks. Restricting abliteration to verified users and channels reduces the risk of an abliterated model being abused through malicious prompt injection. Tool outputs within an authorized request still rely on the harness's protections. ## Get started Download a supported model first. The [model download guide](docs/model-downloads.md) has pinned commands for all five starter profiles and their supported formats.

| Gemma 4 26B A4B | TensorFold, vLLM, SGLang, llama.cpp | See [compatibility](docs/compatibility.md#engines-and-models) for hardware requirements, checkpoint formats and feature limits. The [validation results](docs/validation.md) record the configurations tested. The included [starter profiles](profiles/README.md) remain experimental; setup checks that the selected profile matches the model and engine. Switching happens per request among profiles already loaded by the engine. Loading a new model or profile happens beforehand and can require an engine restart.