prasannaJagadesh/my-model

Prasanna-SmolLM-360M-3.1

A fine-tuned version of SmolLM2-360M-Instruct trained to be a personal AI assistant that answers questions about my professional background, projects, and skills.

This model is designed to serve as a personal AI assistant on a portfolio website and it's trained for only training purpose of the finetuning & Reward model and It answers questions specifically about myself and refuses off-topic or inappropriate requests.

Model Details

Parameter	Value
Base Model	HuggingFaceTB/SmolLM2-360M-Instruct
Parameters	360M
Max Sequence Length	1024
Fine-Tuning Method	LoRA (via Unsloth)
Merge Method	merged_16bit
GGUF Quantizations	q8_0

LoRA Configuration

Parameter	Value
Rank (r)	16
Alpha	32
Dropout	0.05
Bias	none
Gradient Checkpointing	unsloth

Training Arguments

Parameter	Value
Batch Size (per device)	8
Gradient Accumulation Steps	2
Effective Batch Size	16
Epochs	3
Learning Rate	2e-4
Weight Decay	0.01
LR Scheduler	cosine
Optimizer	adamw_8bit
Precision	bf16 (if supported, else fp16)
Packing	enabled
Dataset Workers	2

Dataset

~2K samples curated and reviewed manually, covering:

• Biography & identity
• career & workExp
• technical skills
• tech journey
• contacts & social media
• some Refusal for refuse questions if asked not about me
• NFSW to prevent safety measure

Format

{
    "messages": [
        {
            "role": "system",
            "content": "You are Prasanna's AI Assistant. You answer questions about his professional background, projects, and skills."
        },
        {
            "role": "user",
            "content": "Who is Prasanna?"
        },
        {
            "role": "assistant",
            "content": "Prasanna is a driven Software Engineer based in Chennai, India..."
        }
    ]
}

Usage

Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("prasannaJagadesh/Prasanna-SmolLM-360M-3.1")
tokenizer = AutoTokenizer.from_pretrained("prasannaJagadesh/Prasanna-SmolLM-360M-3.1")

messages = [
    {"role": "system", "content": "You are Prasanna's AI Assistant. You answer questions about his professional background, projects, and skills."},
    {"role": "user", "content": "Tell me about Prasanna."},
]

input_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Warning Limitations

• Only knows about myself, not a general-purpose assistant
• Small model (360M params) very limited reasoning depth compared to larger models
• Best suited for CPU inference on constrained environments (4-8 GB RAM)