README.md

---
license: apache-2.0
base_model: Qwen/Qwen2.5-1.5B-Instruct
tags:
  - lora
  - qwen2.5
  - fine-tuned
  - function-calling
  - tool-use
  - system-administration
  - gguf
  - ollama
pipeline_tag: text-generation
language:
  - ru
  - en
library_name: transformers
---
<svg width="850" height="260" viewBox="0 0 850 260"
xmlns="http://www.w3.org/2000/svg">
<defs>
<linearGradient id="bg" x2="0" y2="1">
<stop stop-color="#050505"/>
<stop offset="1" stop-color="#130000"/>
</linearGradient>
<filter id="glow">
<feGaussianBlur stdDeviation="4"/>
</filter>
<linearGradient id="red">
<stop stop-color="#ff0000"/>
<stop offset="1" stop-color="#ff5555"/>
</linearGradient>
</defs>
<rect width="850" height="260" rx="25" fill="url(#bg)" stroke="#ff0033" stroke-width="2"/>
<g opacity="0.15" stroke="#ff0033">
<path d="M0 50H850"/>
<path d="M0 100H850"/>
<path d="M0 150H850"/>
<path d="M0 200H850"/>
<path d="M100 0V260"/>
<path d="M200 0V260"/>
<path d="M300 0V260"/>
<path d="M400 0V260"/>
<path d="M500 0V260"/>
<path d="M600 0V260"/>
<path d="M700 0V260"/>
</g>
<rect x="0" y="0" width="850" height="3" fill="#ff0033" opacity="0.8">
<animate attributeName="y" from="0" to="260" dur="3s" repeatCount="indefinite"/>
</rect>
<circle cx="130" cy="130" r="70" fill="none" stroke="#ff0033" stroke-width="2" opacity="0.6">
<animate attributeName="r" values="50;80;50" dur="2s" repeatCount="indefinite"/>
</circle>
<g transform="translate(-28.55,-23.79) scale(0.2443)" filter="url(#glow)">
<g transform="translate(0.000000,1254.000000) scale(0.100000,-0.100000)"
fill="url(#red)" stroke="none">
<path d="M3420 7651 l0 -1869 118 -109 c65 -59 171 -155 236 -213 64 -58 132
-118 150 -135 63 -59 212 -193 273 -246 454 -395 628 -549 910 -801 299 -268
372 -330 333 -283 -7 8 -8 15 -2 15 5 0 15 -7 22 -15 27 -33 20 -4 -7 32 -16
21 -42 55 -58 77 -21 29 -30 35 -37 25 -7 -11 -8 -9 -4 4 12 38 -104 166 -704
772 -240 242 -448 456 -463 475 -54 73 -57 110 -57 639 1 450 -1 485 -17 504
-31 35 327 -237 621 -473 312 -249 1055 -830 1371 -1072 l101 -77 74 25 74 24
81 -26 c44 -14 85 -27 91 -30 14 -6 326 225 620 459 51 39 161 127 245 193
172 136 353 284 545 448 322 274 437 369 524 431 30 21 78 57 105 79 28 23 45
34 39 26 -6 -8 -14 -23 -18 -34 -5 -15 -2 -17 13 -13 19 5 19 3 -2 -42 -22
-46 -22 -49 -17 -411 6 -419 2 -503 -29 -582 -28 -71 -47 -92 -526 -583 -546
-560 -536 -549 -539 -581 -2 -16 -5 -49 -6 -74 -2 -36 -3 -40 -9 -20 -5 17 -9
21 -15 12 -4 -6 -4 -12 0 -12 4 0 0 -8 -9 -18 -10 -10 -15 -21 -12 -24 3 -3
-3 -13 -14 -22 -12 -11 -21 -14 -26 -7 -3 6 1 11 9 11 9 0 13 7 11 20 -2 11 0
20 5 20 5 0 7 6 4 14 -3 7 7 23 21 34 15 12 25 28 23 36 -2 9 -25 -11 -62 -54
-52 -60 -60 -74 -62 -115 -3 -44 -2 -46 14 -32 266 231 851 751 1122 997 125
113 346 312 491 444 145 131 279 254 297 272 l32 34 0 719 -1 718 -172 149
c-94 82 -198 171 -230 198 l-58 48 -62 -58 c-110 -103 -831 -808 -1257 -1229
-581 -575 -810 -800 -818 -803 -7 -2 -48 32 -312 263 -649 566 -710 619 -862
748 l-108 92 0 580 0 580 -112 95 c-61 52 -139 120 -173 150 -101 90 -257 227
-480 420 -115 100 -270 234 -344 298 -325 282 -877 752 -883 752 -5 0 -8 -841
-8 -1869z m660 -1098 c0 -3 -6 0 -13 6 -8 6 -16 8 -19 6 -3 -3 -13 6 -23 21
l-17 27 36 -28 c20 -15 36 -30 36 -32z m4593 17 c-10 -17 -19 -26 -21 -20 -3
8 -9 7 -20 -1 -10 -6 -1 3 18 20 43 39 46 39 23 1z m-4582 -75 l10 -20 -16 20
c-9 11 -21 29 -26 40 l-10 20 16 -20 c9 -11 21 -29 26 -40z M9043 8626 c-245
-60 -384 -352 -278 -583 61 -133 154 -210 290 -239 213 -46 427 76 495 282 33
99 26 217 -19 308 -90 182 -293 279 -488 232z M4791 5559 c-16 -47 -13 -197 5
-260 50 -171 189 -313 348 -355 103 -28 241 -14 331 34 46 24 102 68 95 74 -3
3 -33 24 -66 47 l-61 42 -37 -21 c-50 -30 -140 -37 -203 -16 -58 19 -124 82
-148 140 -17 40 -22 119 -10 156 4 12 -2 24 -17 35 -101 73 -218 155 -222 155
-2 0 -9 -14 -15 -31z M7805 5519 c-136 -95 -130 -88 -126 -165 6 -118 -65
-217 -178 -251 -67 -19 -148 -13 -197 16 l-35 21 -62 -42 c-34 -22 -64 -43
-66 -44 -6 -5 46 -45 92 -71 88 -51 243 -68 345 -38 175 52 307 202 348 396
13 62 14 87 4 157 -7 45 -15 84 -19 86 -3 2 -51 -27 -106 -65z M6300 4771
c-104 -32 -320 -184 -479 -337 -93 -89 -94 -91 -99 -145 l-4 -54 318 -638 319
-639 317 640 317 640 -8 54 c-4 29 -10 59 -15 66 -7 12 -189 173 -282 250 -49
40 -72 53 -60 33 3 -5 -9 -21 -28 -36 -31 -25 -33 -30 -25 -58 15 -51 11 -52
-21 -9 -32 45 -39 49 -58 30 -10 -10 -15 -10 -22 2 -7 12 -13 13 -28 3 -12 -7
-24 -8 -33 -2 -26 16 -38 9 -42 -23 -2 -18 -5 27 -5 100 -2 144 -1 141 -62
123z M3696 4355 c-75 -30 -137 -91 -170 -165 -39 -86 -42 -135 -16 -222 19
-64 29 -80 85 -135 77 -76 138 -98 249 -91 114 8 204 66 262 170 26 48 29 62
29 144 0 81 -3 97 -30 149 -76 148 -255 214 -409 150z"/>
</g>
</g>
<text x="230" y="95" font-size="55" font-family="monospace" font-weight="900" fill="url(#red)">HAWK-1.5B</text>
<text x="235" y="135" font-size="18" font-family="monospace" fill="#ffffff">CKACOR // NEURAL AUTOMATION CORE</text>
<rect x="230" y="160" width="520" height="60" rx="8" fill="#000" stroke="#ff0033"/>
<text x="250" y="185" font-size="14" font-family="monospace" fill="#ff3333">STATUS: ONLINE</text>
<text x="250" y="205" font-size="14" font-family="monospace" fill="#ffffff">MODEL: QWEN2.5-1.5B | MODE: FUNCTION_CALLING</text>
</svg>

---
license: apache-2.0
base_model: Qwen/Qwen2.5-1.5B-Instruct
tags:
  - lora
  - qwen2.5
  - fine-tuned
  - function-calling
  - tool-use
  - system-administration
  - gguf
  - ollama
pipeline_tag: text-generation
language:
  - ru
  - en
library_name: transformers
---

<div align="center">
  <img src="banner.svg" alt="Hawk-1.5B banner" width="100%"/>
</div>

<div align="center" style="background: rgba(255, 0, 0, 0.05); border: 1px solid rgba(255, 0, 0, 0.2); border-radius: 12px; padding: 12px 20px; max-width: 650px; margin-top: 25px; margin-bottom: 25px; box-shadow: 0 4px 15px rgba(0,0,0,0.4); display: flex; align-items: center; justify-content: center; gap: 15px;">
  <span style="font-size: 2rem;">🦅</span>
  <p style="font-size: 1.2rem; color: #f0f6fc; margin: 0; line-height: 1.4; font-weight: 500; text-align: left;">
    <b style="color: #ff4d4d;">Zero-latency</b> SysAdmin assistant. Optimized for instant Linux tool chaining via <b style="color: #ff4d4d;">JSON function calling</b>. No hallucinations, just operations.
  </p>
</div>

<div align="center">

[![Base](https://img.shields.io/badge/Base_Model-Qwen2.5--1.5B--Instruct-ff0000?style=for-the-badge&logo=huggingface&logoColor=white&labelColor=000)](https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct)
[![Method](https://img.shields.io/badge/Optimization-QLoRA-7928ca?style=for-the-badge&logo=pytorch&logoColor=white&labelColor=000)](#training-details)
[![Runtime](https://img.shields.io/badge/Runtime-Ollama_Compatible-f0f6fc?style=for-the-badge&logo=ollama&logoColor=black&labelColor=000)](https://ollama.ai)
[![Task](https://img.shields.io/badge/Task-Function_Calling-00ff88?style=for-the-badge&logoColor=black&labelColor=000)](#technical-architecture--capabilities)

</div>

<div class="code-terminal" style="background-color: #000; border: 1px solid #ff0000; border-radius: 16px; padding: 25px; max-width: 700px; text-align: left; box-shadow: 0 10px 40px rgba(0,0,0,0.8), 0 0 15px rgba(255, 0, 0, 0.2); margin: 0 auto 40px auto;">
  <div style="display: flex; align-items: center; margin-bottom: 18px; border-bottom: 1px solid rgba(255,255,255,0.08); padding-bottom: 12px;">
    <span style="height: 12px; width: 12px; background-color: #ff0000; border-radius: 50%; display: inline-block; margin-right: 8px;"></span>
    <span style="height: 12px; width: 12px; background-color: #30363d; border-radius: 50%; display: inline-block; margin-right: 8px;"></span>
    <span style="height: 12px; width: 12px; background-color: #30363d; border-radius: 50%; display: inline-block; margin-right: 15px;"></span>
    <span style="font-family: monospace; font-size: 0.9rem; color: #8b949e; letter-spacing: 1px;">ckacor@ops:~</span>
  </div>
  <code style="font-family: 'Courier New', Courier, monospace; font-size: 1rem; line-height: 1.6; color: #f0f6fc;">
    <span style="color: #ff0000;">$</span> ollama run ckacor/hawk-1.5b \
    <br/>
    &nbsp;&nbsp;<span style="color: #ff0000; opacity: 0.7;">></span> <span style="color: #fff;">"Get GPU temp and check system uptime"</span>
  </code>
  <div style="text-align: right; font-family: monospace; font-size: 0.75rem; color: rgba(255,0,0,0.4); text-transform: uppercase; letter-spacing: 1px; margin-top: 10px;">Execution Ready</div>
</div>

---

## Technical architecture & capabilities

**Hawk-1.5B** by **ckacor** is a specialized agent model built on `Qwen2.5-1.5B-Instruct` and fine-tuned using the **QLoRA** methodology.

Its purpose is to parse natural-language requests (EN/RU) and convert them into structured, executable **JSON function calls** for system administration tasks. It does not produce conversational text — it produces operational data.

### Out-of-the-box toolchain

| Tool function | Description | Typical use case |
|---|---|---|
| `get_gpu_status` | Returns GPU metrics: VRAM, load, temperature | Inference node monitoring |
| `execute_terminal_command` | Runs an arbitrary Bash command | DevOps automation, daemon management |
| `read_file` | Reads the contents of a specified file path | Config inspection, log parsing |

<details>
<summary><b>Example prompts it handles</b></summary>

| Prompt | Model output |
|---|---|
| 🇬🇧 "Check GPU utilization" | `{"tool": "get_gpu_status", "arguments": {}}` |
| 🇬🇧 "Uptime?" | `{"tool": "execute_terminal_command", "arguments": {"command": "uptime"}}` |
| 🇷🇺 "Проверь дисковое пространство" | `{"tool": "execute_terminal_command", "arguments": {"command": "df -h"}}` |
| 🇷🇺 "Покажи конфиг config.json" | `{"tool": "read_file", "arguments": {"filepath": "config.json"}}` |

</details>

> No formal latency/accuracy benchmarks have been measured yet — numbers will be added here once evaluated, rather than estimated.

---

## Instant deployment with Ollama

1. **Download the GGUF** from the **Files and versions** tab.
2. **Create a Modelfile:**

```dockerfile
FROM ./hawk-1.5b-q8_0.gguf
PARAMETER temperature 0.1
PARAMETER top_p 0.95
PARAMETER repeat_penalty 1.1
SYSTEM """
You are HAWK-1.5B,
a Linux automation AI agent developed by ckacor.
Your task is to transform user requests
into accurate function calls.
"""
```

3. **Build and run:**

```bash
ollama create hawk-1.5b -f Modelfile
ollama run hawk-1.5b
```

```
> Check GPU temperature

{
  "tool": "get_gpu_status",
  "arguments": {}
}
```

### Via Transformers

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("ckacor/my-lora-model")
tokenizer = AutoTokenizer.from_pretrained("ckacor/my-lora-model")
```

---

## Files and versions

> Only `q8_0` is available right now. Additional quantizations will be added once generated.

| File | Description |
|---|---|
| `hawk-1.5b-q8_0.gguf` | Maximum precision (available) |
| `hawk-1.5b-q5_k_m.gguf` | Higher quality, smaller size — coming soon |
| `hawk-1.5b-q4_k_m.gguf` | Recommended for most users — coming soon |

---

## Recommended runtime settings

| Parameter | Value |
|---|---|
| Temperature | 0.1 – 0.3 |
| Top P | 0.9 – 0.95 |
| Context | depends on hardware |
| Quantization | Q4_K_M recommended once available |
| Mode | deterministic generation |

---

## Training details

| | |
|---|---|
| **Base model** | Qwen2.5-1.5B-Instruct |
| **Method** | QLoRA |
| **Data** | Custom dataset of tool-calling examples for system administration tasks |
| **Languages** | Russian, English |

---

## Roadmap

<table>
<tr><th>Model</th><th>Focus</th><th>Status</th></tr>
<tr><td>Hawk-1.5B</td><td>General tool-use / sysadmin</td><td>✅ released (this repo)</td></tr>
<tr><td>Hawk-1.5B-Instruct</td><td>Broader instruction following</td><td>🔜 planned</td></tr>
<tr><td>Hawk-1.5B-Reasoning</td><td>Multi-step reasoning</td><td>🔜 planned</td></tr>
<tr><td>Hawk-1.5B-Code</td><td>Code generation</td><td>🔜 planned</td></tr>
<tr><td>Hawk-1.5B-Vision</td><td>Image understanding</td><td>🔜 planned</td></tr>
</table>

---

## Limitations

- Not designed for long creative writing
- Not a replacement for large general-purpose LLMs
- Tool execution requires an external agent layer — this model only outputs the JSON, it does not execute commands itself
- Performance depends on fine-tuning data quality
- May occasionally switch languages mid-response
- No formal benchmark numbers have been published yet

---

## Safety notes

HAWK-1.5B only generates commands and actions — it does not execute them. The execution layer should always:

- validate commands before running them
- restrict permissions to the minimum necessary
- use sandboxing where possible
- require confirmation for destructive operations

---

## About ckacor

ckacor develops lightweight AI systems focused on:

- local artificial intelligence
- machine learning experiments
- Linux automation
- efficient models for limited hardware

```
ckacor AI
    |
    ▼
 HAWK Family
    |
    ├── Hawk-1.5B
    |
    └── Future Models
```

---

## License

This model follows the license specified in the model repository (Apache 2.0). Please check the license before commercial deployment.

---

## Citation

```bibtex
@misc{ckacor2026hawk,
  title  = {Hawk-1.5B: a LoRA fine-tune of Qwen2.5-1.5B for sysadmin tool-use},
  author = {ckacor},
  year   = {2026},
  url    = {https://huggingface.co/ckacor/my-lora-model}
}
```

---

<div align="center">

**🦅 HAWK-1.5B**

Built by <a href="https://huggingface.co/ckacor">ckacor</a>

*Local AI. Efficient automation.*

</div>