-
Model Name: DgMind
-
Developer: Erfan Mohamadnia
-
Core Persona: A specialized AI assistant that excels in complex coding tasks, architectural decisions, and deep logical analysis.
-
Base Model: GPT-OSS 20B (Unsloth 4-bit optimized)
-
-
Technique: LoRA (Low-Rank Adaptation)
-
Optimization: Fine-tuned specifically on responses to enhance conversational accuracy and identity injection.
The model demonstrates a stable decrease in training loss, ensuring precise instruction following and a minimized hallucination rate in coding contexts.
DgMind uses the following message structure to maintain context and role separation:
1
2{% for message in messages %}{{ '<|start|>' + message['role'] + '<|message|>' + message['content'] + '<|end|>' }}{% endfor %}{% if add_generation_prompt %}{{ '<|start|>assistant<|message|>' }}{% endif %}
3
4
5
1
2<|start|>user<|message|>Write a Python script for a custom API gateway.<|end|>
3
4<|start|>assistant<|message|>
5
6
7
-
Download the .gguf file.
-
Create a file named Modelfile:
1
2FROM "./DgMind-20B.Q4_K_M.gguf"
3
4PARAMETER temperature 0.7
5
6SYSTEM """You are DgMind, a helpful AI assistant developed by Erfan Mohamadnia. You specialize in advanced reasoning and expert-level coding."""
7
8
9
1
2./llama-server -m DgMind-20B.Q4_K_M.gguf --host 0.0.0.0 --port 8080 --n-gpu-layers 62
3
4
5
