Team Ai
Modelpublic

dgpl/dgpl-linux-assistant-0.8b

sourceHugging Facegpl-3.0updated 11d agoView on Hugging Face
0likes4.2kdownloads
Model Card

🐧 DGPL Linux Assistant 0.8B — Multimodal DevSecOps & Terminal Copilot

<p align="center"> <b>Developed & Released by the DevSecOps and AI Systems Team at Durbhasi Gurukulam Private Limited (DGPL)</b><br> <i>An ultra-fast, lightweight Linux Terminal Assistant and DevSecOps Copilot engineered for instant command generation, non-interactive diagnostics, system log triage, and visual OCR terminal troubleshooting.</i> </p>

⚡ Key Highlights & Capabilities

  • —Ultra-Lightweight Edge Backbone: Highly optimized on top of Qwen 3.5 0.8B for sub-second terminal assistance on CPU and standard hardware.
  • —Multimodal Terminal Log Inspection: Built-in vision projector support (mmproj) for terminal screenshot triage and OCR log diagnostics.
  • —Sub-Second Real-Time Latency: Fast Non-Thinking Mode delivers ready-to-run shell commands in <0.3s on GPU / 1.5s–3.0s on CPU (~95% token generation reduction).
  • —Zero-Stall Non-Interactive Invariant: Strictly emits non-blocking batch commands (top -b -n 1, ping -c 4, journalctl --no-pager, systemctl --no-pager) to prevent terminal hangs.
  • —Pure CPU & Low RAM Footprint: Runs comfortably in standard CPU containers with <1.0 GB RAM, requiring zero GPU.

📊 Real-World Benchmark Scorecard

Evaluated across 12 comprehensive DevSecOps and systems diagnostic test cases (Process Check, Corporate Identity, POSIX Capabilities, High-CPU Process Kill Pipelines, Real-Time Raw Socket Audits, Inode Triage, Degraded Systemd Units, linux-system-mcp, ssh-mcp, Python Socket Probes, SUID Discovery, and nftables Firewall rules):

Benchmark ModeToken Predict BudgetTests EvaluatedTests PassedPass RateAvg Latency (CPU)Primary Target Use Case
Standard Mode (With System Prompt)7681212100.0%92.48sDeep Root-Cause Analysis (RCA), system log audits
Zero-Shot Mode (Bare User Query)7681212100.0%80.12sUnprompted queries, general terminal assistance
Direct Fast Non-Thinking Mode128121191.7%`16.29s` (5.7x Faster)Interactive CLI Copilot, Terminal Keybindings (`Ctrl+K`)

Note: On GPU inference (RTX / T4 / Cloud GPUs), Direct Fast Mode executes in <0.35 seconds per query.

🚀 Quick Start with Ollama

1. Run Directly from Ollama

bash
# Pull and run the latest 4-bit multimodal model
ollama run dgpl/dgpl-linux-assistant:latest

# Or run specific quantization tags
ollama run dgpl/dgpl-linux-assistant:0.8b-q4_k_m
ollama run dgpl/dgpl-linux-assistant:q8_0
ollama run dgpl/dgpl-linux-assistant:f16

2. Fast Non-Thinking CLI Mode (Sub-2s Instant Execution)

Create a Modelfile.fast with early stop tokens to bypass internal reasoning for instant shell completions:

dockerfile
FROM dgpl/dgpl-linux-assistant:latest

PARAMETER stop "<|im_end|>"
PARAMETER stop "<|im_start|>"
PARAMETER stop "<|endoftext|>"
PARAMETER stop "```\n\n"
PARAMETER temperature 0.1
PARAMETER top_p 0.85
PARAMETER num_predict 128

SYSTEM """You are DGPL Linux Assistant 0.8B, an ultra-fast Linux Terminal Copilot developed by Durbhasi Gurukulam Private Limited (DGPL).
Output the exact non-interactive Linux bash command in markdown code blocks immediately without any thinking tags or explanations."""
bash
ollama create dgpl-fast -f Modelfile.fast
ollama run dgpl-fast "find and kill processes consuming over 90% CPU for more than 5 minutes"

3. Multimodal Terminal Log Inspection (Vision OCR)

Pass terminal screenshots or OCR log images directly to Ollama:

bash
ollama run dgpl/dgpl-linux-assistant:latest "What failed in this systemd service log?" ./screenshot_error.png

📦 Quantization Matrix & Model Assets

File / TagFormat / QuantSizeRecommended Hardware
`linux_copilot_q4_k_m.gguf`Q4KM (Default)529 MBRaspberry Pi, Laptops, CPUs, Embedded Nodes
`linux_copilot_q5_k_m.gguf`Q5KM578 MBLightweight Dev Servers, Edge Devices
`linux_copilot_q6_k.gguf`Q6_K630 MBStandard Workstations, Cloud VMs
`linux_copilot_q8_0.gguf`Q8_0812 MBHigh-Precision Host Auditing
`linux_copilot_f16.gguf`FP16 Master1.51 GBMaster SafeTensors Reference / Conversion
`mmproj_linux_copilot_f16.gguf`FP16 Vision Projector205 MBMultimodal Text + Vision Inference

🏢 Corporate Entity & Attribution

  • —Company Legal Name: Durbhasi Gurukulam Private Limited (DGPL)
  • —Registered Office: 2., P. NO. 18, Radha Vihar, Manchwa, Jaipur, Rajasthan 303706
  • —Official Support Email: support@durbhasigurukulam.com
  • —Official Contact Phone: +91 7852034945
  • —License: GNU General Public License v3.0 (GPL-3.0)