hawierdev/shellm-v0.1-gguf

ShellM v0.1

ShellM is a specialized language model based on Qwen2.5 1.5B, fine-tuned for generating intelligent terminal command suggestions. The model has been trained to understand terminal context and suggest appropriate commands.

Model Specifications

• Architecture: Qwen2ForCausalLM
• Size: 1.5B parameters
• Hidden size: 1536
• Layers: 28
• Attention heads: 12 (2 key-value heads)
• Max context length: 32768 tokens
• Precision: bfloat16

Capabilities

• Terminal command suggestions based on context
• Understanding complex system tasks
• Shell script generation
• Command debugging assistance
• Multi-platform support

Usage

The model uses chatml format for conversations:

<|im_start|>user
How to display all processes using port 8080?
<|im_end|>
<|im_start|>assistant
lsof -i :8080
<|im_end|>

Example Use Cases

File Management:

User: How to find all Python files larger than 1MB?
Assistant: find . -name "*.py" -size +1M

Process Management:

User: How to kill all python processes?
Assistant: pkill python

Git Operations:

User: How to undo last commit without losing changes?
Assistant: git reset --soft HEAD~1

Installation and Usage

Requirements:

• transformers
• torch
• tokenizers

from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = "shellm_v0.1_merged"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Example usage
prompt = "<|im_start|>user\nHow to check disk usage?<|im_end|>\n<|im_start|>assistant\n"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=150, temperature=0.7)
response = tokenizer.decode(outputs[0], skip_special_tokens=False)

Special Tokens

The model supports standard Qwen2 tokens:

• <|im_start|>, <|im_end|> - conversation markers
• <|vision_pad|> - padding token
• Fill-in-the-middle tokens: <|fim_prefix|>, <|fim_middle|>, <|fim_suffix|>

Version Info

Version: v0.1 Based on: Qwen2.5-1.5B Fine-tuned with: Unsloth v2025.8.8