mdk615661/it-helpdesk-merged-v3

IT Helpdesk AI — Full Merged Model v3

Model Description

Production-ready fully merged fine-tuned version of Mistral-7B-v0.1 trained on 1141 IT helpdesk tickets. This is v3 — best version with normalized output that directly maps to IT helpdesk database tables.

• Developed by: mdk615661
• Model type: Causal Language Model (Mistral-7B fully merged)
• Language: English
• License: Apache 2.0
• Finetuned from: mistralai/Mistral-7B-v0.1
• Adapter repo: mdk615661/it-helpdesk-qlora-v3

What It Does

Input any IT support ticket → Returns structured output:

• Normalized — standardized ticket title matching DB table
• Category — Hardware / Software / Incident / Others / Procurement
• Subcategory — specific issue type
• Insight — AI analysis of the problem
• Recommendation — actionable step for IT team

How To Use

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = "mdk615661/it-helpdesk-merged-v3"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    dtype=torch.float16,
    device_map="auto"
)

def classify_ticket(ticket):
    prompt = f"""### Instruction:
Normalize and classify this IT ticket

### Input:
{ticket}

### Output:
"""
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    outputs = model.generate(
        **inputs,
        max_new_tokens=200,
        temperature=0.1,
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id
    )
    result = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return result.split("### Output:")[-1].strip()

print(classify_ticket("My laptop is not connecting to WiFi"))

Example Output

Input: My laptop is not connecting to WiFi

Output:

Normalized: wifi connectivity issue Category: Hardware Subcategory: Hardware - Laptop Insight: WiFi adapter driver may be outdated or misconfigured Recommendation: Update WiFi driver and check network adapter settings

Training Details

• Base Model: mistralai/Mistral-7B-v0.1
• Method: QLoRA (4-bit NF4 quantization + LoRA)
• Dataset Size: 1141 IT helpdesk tickets
• Epochs: 3
• Batch Size: 4
• Gradient Accumulation: 4
• Learning Rate: 2e-4
• LoRA Rank (r): 16
• LoRA Alpha: 32
• Max Length: 512
• Training Loss: 0.264
• Validation Loss: 0.344
• Training Platform: Google Colab T4 GPU
• Merge: Adapter fully merged via merge_and_unload()

Version History

Version	Samples	Train Loss	Val Loss
v1 (it-helpdesk-merged)	301	0.595	0.831
v2 (it-helpdesk-merged-v2)	451	0.558	0.763
v3 (it-helpdesk-merged-v3)	1141	0.264	0.344

Model Versions

Model	Description	Size
mdk615661/it-helpdesk-qlora-v3	LoRA adapter only	54MB
mdk615661/it-helpdesk-merged-v3	Full merged model	14.5GB

Deployment

• Platform: Kaggle GPU T4
• API: Flask + ngrok
• Integration: Java project via REST API

Limitations

• English language only
• Best for corporate IT helpdesk scenarios
• Should be reviewed by IT staff before action
• Performance improves with more organization-specific data