← 返回 Skills 市场
LoRA Toolkit
作者
loutai0307-prog
· GitHub ↗
· v1.0.0
· MIT-0
82
总下载
0
收藏
1
当前安装
1
版本数
在 OpenClaw 中安装
/install bytesagain-lora-toolkit
功能描述
Configure, estimate, and generate LoRA fine-tuning scripts for LLMs. Input: base model name, dataset size, GPU spec. Output: training config, PEFT script, co...
使用说明 (SKILL.md)
LoRA Toolkit
Configure and generate LoRA fine-tuning scripts for large language models. Supports Llama, Mistral, Qwen, Phi and other HuggingFace-compatible models.
Commands
config
Generate a LoRA training configuration for your model and hardware.
bash scripts/script.sh config --model llama3-8b --gpu 24gb --dataset 10000
Parameters:
--model— base model (llama3-8b, mistral-7b, qwen2-7b, phi3-mini, llama3-70b)--gpu— VRAM size (8gb, 16gb, 24gb, 40gb, 80gb)--dataset— number of training samples
estimate
Estimate VRAM usage, training time, and cost before starting.
bash scripts/script.sh estimate --model mistral-7b --gpu 16gb --dataset 5000 --epochs 3
generate
Generate a ready-to-run Python training script using HuggingFace PEFT + TRL.
bash scripts/script.sh generate --model llama3-8b --output train.py
validate
Check dataset format compatibility (Alpaca / ShareGPT / OpenAI Chat format).
bash scripts/script.sh validate --file dataset.json --format alpaca
recommend
Recommend the best base model for your use case and hardware.
bash scripts/script.sh recommend --task chat --gpu 16gb --language en
help
Show all commands.
bash scripts/script.sh help
LoRA Parameters Reference
| Model Size | Recommended Rank | Alpha | VRAM (4-bit) |
|---|---|---|---|
| 7B | 16–32 | 32–64 | 8–12 GB |
| 13B | 16 | 32 | 14–18 GB |
| 70B | 8–16 | 16–32 | 40–48 GB |
Supported Dataset Formats
- Alpaca:
{"instruction": "...", "input": "...", "output": "..."} - ShareGPT:
{"conversations": [{"from": "human", "value": "..."}, ...]} - OpenAI Chat:
{"messages": [{"role": "user", "content": "..."}, ...]}
Requirements
- Python 3.8+
- Optional:
pip install transformers peft trl datasetsfor script execution
Feedback
https://bytesagain.com/feedback/ Powered by BytesAgain | bytesagain.com
安全使用建议
This skill appears coherent for generating LoRA configs and training scripts, but take the following precautions before running anything it produces: 1) Review the generated train.py and any other files the script writes. The training script sets trust_remote_code=True when loading pretrained models — that means code fetched from a model repository can run on your machine. Only load models from sources you trust. 2) Verify dataset contents (no secrets) before using them in training. 3) Be aware of resource and cost implications if you run the training (GPU hours, cloud billing). 4) If you want stricter safety, remove or change trust_remote_code in the generated script and pin models to trusted repos/commits. If you need more assurance, request the full contents of the truncated validate function or a provenance/hosting URL for the skill's author and homepage.
功能分析
Type: OpenClaw Skill
Name: bytesagain-lora-toolkit
Version: 1.0.0
The toolkit contains a shell injection vulnerability in the `validate` command within `scripts/script.sh`, where the `--file` argument is unsafely interpolated into a `python3 -c` command string. Additionally, the `generate` command produces training scripts that default to `trust_remote_code=True`, which is a high-risk configuration that allows for arbitrary code execution when loading untrusted models from HuggingFace. While the tool's logic aligns with its stated purpose of LoRA fine-tuning, these security flaws pose a significant risk to the environment.
能力评估
Purpose & Capability
Name/description align with the provided shell script and SKILL.md: the tool generates configs, estimates cost/VRAM, validates dataset format, and emits a Python training script. Required resources (none declared) and included files are consistent with this purpose.
Instruction Scope
Instructions are narrowly scoped to generating configs, validating datasets, and producing a training script. However the generated Python uses trust_remote_code=True and will load pretrained models from external model repositories when executed — that can cause arbitrary code execution coming from a model repo. The shell script also creates a directory under $HOME/.local/share/bytesagain-lora-toolkit and writes files there and into the working directory (train.py) which you should review before running.
Install Mechanism
No install spec; the skill is instruction-only plus an included shell script. The README suggests pip installing well-known ML packages (transformers, peft, trl, bitsandbytes, etc.) which is expected for this use case. Nothing is downloaded from obscure URLs by the skill itself.
Credentials
The skill does not request environment variables, credentials, or access to unrelated configuration paths. It stores data under the user's home directory (local, per-user) which is proportionate for a CLI helper.
Persistence & Privilege
always:false and no special persistence requested. The script creates and uses a per-user data directory only for its own files; it does not modify other skills or global agent settings.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install bytesagain-lora-toolkit - 安装完成后，直接呼叫该 Skill 的名称或使用
/bytesagain-lora-toolkit触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.0.0
LoRA fine-tuning toolkit: config/estimate/generate/validate/recommend — supports Llama3/Mistral/Qwen2
元数据
常见问题
LoRA Toolkit 是什么？
Configure, estimate, and generate LoRA fine-tuning scripts for LLMs. Input: base model name, dataset size, GPU spec. Output: training config, PEFT script, co... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 82 次。
如何安装 LoRA Toolkit？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install bytesagain-lora-toolkit」即可一键安装，无需额外配置。
LoRA Toolkit 是免费的吗？
是的，LoRA Toolkit 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
LoRA Toolkit 支持哪些平台？
LoRA Toolkit 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 LoRA Toolkit？
由 loutai0307-prog（@loutai0307-prog）开发并维护，当前版本 v1.0.0。
推荐 Skills