| | --- |
| | license: apache-2.0 |
| | base_model: LiquidAI/LFM2.5-1.2B-Instruct |
| | tags: |
| | - linux |
| | - terminal |
| | - bash |
| | - devops |
| | - liquid-foundation-model |
| | - multilingual |
| | - arabic |
| | languages: |
| | - en |
| | - ar |
| | - ta |
| | metrics: |
| | - accuracy |
| | model_name: HydroShell-1.2B |
| | datasets: |
| | - missvector/linux-commands |
| | language: |
| | - en |
| | - ar |
| | --- |
| | |
| | # HydroShell-1.2B: Liquid Linux Expert |
| | <img src="logo.png" width="50%"/> |
| |
|
| | **HydroShell-1.2B** is a specialized, multilingual fine-tuned version of the **Liquid AI (LFM 2.5 1.2B)** model. It is optimized to act as a high-performance, low-latency assistant for Linux system administration, shell scripting, and DevOps automation. |
| |
|
| | By leveraging the **Liquid Foundation Model** architecture, HydroShell excels at processing long-form technical instructions and mapping complex natural language (English, Arabic, and Tamil) to functional Bash one-liners. |
| |
|
| |
|
| |
|
| | ## ⚠️ Safety & Destructive Command Warning |
| |
|
| | > **WARNING:** This model is designed to generate powerful system-level commands. It can and will generate **destructive commands** (e.g., `rm -rf`, `mkfs`, or overwriting configurations with `>`). |
| | > * **Always verify commands** in a sandbox or test environment before executing them on production systems. |
| | > * The model may occasionally hallucinate flags or mix Linux distributions (e.g., suggesting `pacman` for Ubuntu systems). |
| |
|
| | --- |
| |
|
| | ## Model Details |
| |
|
| | - **Developed by:** [Your Name/MindLab] |
| | - **Base Model:** LiquidAI/LFM2.5-1.2B-Instruct |
| | - **Architecture:** Liquid Foundation Model (Dynamical Systems-based) |
| | - **Primary Domain:** Linux CLI, Bash Scripting, System Hardening. |
| | - **Languages Supported:** English, Arabic (Technical). |
| |
|
| |
|
| |
|
| | --- |
| |
|
| | ## Evaluation Results (Zero-Shot Testing) |
| |
|
| | The following results were observed during a 100-prompt "Stress Test" covering System Audit, Security, and File Management. |
| |
|
| | ### Technical Performance Matrix |
| | | Category | Accuracy | Notes | |
| | | :--- | :--- | :--- | |
| | | **Basic Admin (`ls`, `cd`, `mkdir`)** | 98% | Flawless execution. | |
| | | **Log Parsing (`awk`, `sed`, `grep`)** | 75% | Occasionally confuses line vs. field flags. | |
| | | **Systemd & Services** | 90% | Strong understanding of service lifecycles. | |
| | | **Networking (`iptables`, `ss`)** | 82% | Occasional source/destination flag inversion. | |
| |
|
| | ### Multilingual Capability |
| | - **Arabic:** 90% Accuracy in intent recognition. Successfully maps Arabic technical terms like "حظر" (Block) and "مزامنة" (Sync). |
| | - **English:** 95% Accuracy in intent recognition. |
| |
|
| | --- |
| |
|
| | ## Known Issues & Limitations |
| | 1. **Distro Confusion:** The model may suggest Arch Linux (`pacman`) commands when asked for Ubuntu tasks if the prompt is not specific. |
| | 2. **Redirection Risks:** In some tests, the model used `>` (overwrite) instead of `>>` (append) for configuration files. |
| | 3. **Hallucination:** For very complex `find` commands, it may invent non-existent flags (e.g., `-md5`). |
| |
|
| | --- |
| |
|
| | ## Usage (Python) |
| |
|
| | ```python |
| | from transformers import AutoTokenizer, AutoModelForCausalLM |
| | |
| | model_id = "your-username/HydroShell-1.2B" |
| | tokenizer = AutoTokenizer.from_pretrained(model_id) |
| | model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", trust_remote_code=True) |
| | |
| | messages = [{"role": "user", "content": "البحث عن العمليات التي تستهلك أكبر قدر من الذاكرة"}] |
| | inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to("cuda") |
| | |
| | outputs = model.generate(**inputs, max_new_tokens=64, temperature=0.3) |
| | print(tokenizer.decode(outputs[0], skip_special_tokens=True)) |
| | |
| | ``` |
| |
|
| | ## Citation |
| |
|
| | If you use this model in your research or projects, please cite the base Liquid AI model and this fine-tuned version. |
| |
|
| | ``` |
| | |
| | --- |
