Model Card for SmolLM2-135M-Technician-QA

This model is a domain-adapted version of HuggingFaceTB/SmolLM2-135M, fine-tuned to answer questions related to automotive service, technician workflows, diagnostics, and spare part replacement scenarios.

It is optimized for lightweight deployment in workshop assistants, service center copilots, and edge devices.


Model Details

Model Description

SmolLM2-135M-Technician-QA is a compact instruction-following language model fine-tuned on a curated dataset of technician question-answer pairs covering:

  • Customer vehicle issues
  • Technical diagnostics
  • Work order lifecycle
  • Periodic service procedures
  • Spare part replacement decisions
  • On-site breakdown support

The model is designed for real-world automotive service environments where fast and efficient inference is required.

  • Developed by: Shailesh H
  • Funded by: Self / Research & Development
  • Shared by: Shailesh H
  • Model type: Causal Language Model (Instruction-tuned)
  • Language(s) (NLP): English
  • License: Apache-2.0
  • Finetuned from model: HuggingFaceTB/SmolLM2-135M

Model Sources


Uses

Direct Use

This model can be used for:

  • Automotive technician assistants
  • Workshop chatbot systems
  • Service advisor support
  • Troubleshooting guidance
  • Training simulators for technicians

Downstream Use

The model can be integrated into:

  • RAG systems with service manuals
  • Mobile workshop applications
  • Edge diagnostic tools
  • Voice-based service assistants

Out-of-Scope Use

This model should NOT be used for:

  • Safety-critical vehicle control
  • Legal or compliance decisions
  • Autonomous driving systems
  • Financial or medical advice

Bias, Risks, and Limitations

  • Trained on synthetic domain data → may not cover all vehicle models
  • Limited general world knowledge due to small model size
  • May generate plausible but incorrect repair steps
  • English-only responses

Recommendations

  • Always verify outputs with OEM service manuals
  • Use as an assistive tool, not a final authority
  • Combine with RAG for production deployment

How to Get Started with the Model

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "<your-username>/SmolLM2-135M-Technician-QA"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

prompt = "Customer says the car battery drains overnight. What should you check?"

inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=120)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Evaluation

Testing Data, Factors & Metrics

Testing Data

Held-out automotive technician QA samples from the same domain.

Factors

Customer complaint handling

Diagnostic reasoning

Spare part replacement logic

Service workflow understanding

Metrics

Perplexity

Instruction-following accuracy

Manual domain evaluation

Results

Strong performance on workshop troubleshooting queries

Accurate step-by-step diagnostic suggestions

Fast inference on CPU

Summary

The fine-tuned model shows clear domain adaptation compared to the base SmolLM2 model, especially for automotive service workflows.

Downloads last month
10
Safetensors
Model size
0.1B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support