SFT_VLA_Qwen3-VL-2B-Instruct-multimage-trl

Model Card for Model ID

This model is a fine-tuned version of Qwen/Qwen3-VL-2B-Instruct. It has been trained using TRL.

Refer main github repository for more details: https://github.com/SarveshBTelang/SFT-Qwen3-Vision-Language-Assistant-for-Autonomous-Driving

For inference:

Model Details

Model Description

A parameter-efficient fine-tuned vision-language model for ADAS and autonomous driving applications. This model processes driving videos as sequential multi-image inputs to generate structured scene descriptions, driving parameters, and safety risk assessments. Key Features:

Fine-tuned using QLoRA (4-bit quantization) on BDD100K driving videos ~2B parameters with 1.64% LoRA adapter overhead (~35M trainable params) Generates outputs in natural language and structured JSON format Performs temporal reasoning across video frames for context-aware understanding Optimized for T4 GPU deployment (<16GB VRAM)

Capabilities:

Scene understanding and object detection across temporal sequences Contextual awareness of traffic rules and driving norms Safety-critical hazard identification (pedestrians, lane changes, etc.) Structured output generation adhering to JSON schemas

• Developed by: Sarvesh Telang
• Funded by [optional]: [More Information Needed]
• Shared by [optional]: [More Information Needed]
• Model type: [More Information Needed]
• Language(s) (NLP): [More Information Needed]
• License: Apache-2.0
• Finetuned from model [optional]: Qwen3-VL-2B-Instruct

Model Sources [optional]

<!-- Provide the basic links for the model. -->

• Repository: Github
• Paper [optional]: [More Information Needed]
• Demo [optional]: Colab

Uses

Autonomous driving perception, ADAS systems, driving scene analysis, safety assessment, driver assistance applications.

Direct Use

<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->

[More Information Needed]

Downstream Use [optional]

<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->

[More Information Needed]

Out-of-Scope Use

<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->

[More Information Needed]

Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. -->

[More Information Needed]

Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

How to Get Started with the Model

Refer to the colab notebook for model inference.

Training Details

Training Data

BDD100K-derived instruction-following dataset

Training Procedure

Refer to the "generate_instruction_dataset.ipynb" from the main github repository.

Preprocessing [optional]

[More Information Needed]

Training Hyperparameters

• Training regime: fp16 mixed precision

Speeds, Sizes, Times [optional]

Training Time: ~2-4 hours (T4 GPU) Memory Footprint: <16GB VRAM Inference Latency: ~60 seconds per video

Evaluation

<!-- This section describes the evaluation protocols and provides the results. -->

Testing Data, Factors & Metrics

Testing Data

<!-- This should link to a Dataset Card if possible. -->

[More Information Needed]

Factors

<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->

[More Information Needed]

Metrics

<!-- These are the evaluation metrics being used, ideally with a description of why. -->

[More Information Needed]

Results

[More Information Needed]

Summary

[More Information Needed]

Model Examination [optional]

<!-- Relevant interpretability work for the model goes here -->

[More Information Needed]

Environmental Impact

<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

• Hardware Type: Nvidia T4 GPU
• Hours used: 2-4 hours
• Cloud Provider: [More Information Needed]
• Compute Region: [More Information Needed]
• Carbon Emitted: [More Information Needed]

Technical Specifications [optional]

Model Architecture and Objective

[More Information Needed]

Compute Infrastructure

[More Information Needed]

Hardware

Hardware: Google Colab T4 GPU (2-4 hour training time) Framework: TRL (Transformer Reinforcement Learning) with PEFT

Software

[More Information Needed]

Citation [optional]

@misc{vonwerra2022trl,
	title        = {{TRL: Transformer Reinforcement Learning}},
	author       = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
	year         = 2020,
	journal      = {GitHub repository},
	publisher    = {GitHub},
	howpublished = {\url{https://github.com/huggingface/trl}}
}

@InProceedings{bdd100k,
    author = {Yu, Fisher and Chen, Haofeng and Wang, Xin and Xian, Wenqi and Chen, Yingying and Liu, Fangchen and Madhavan, Vashisht and Darrell, Trevor},
    title = {BDD100K: A Diverse Driving Dataset for Heterogeneous Multitask Learning},
    booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
    month = {June},
    year = {2020}
}

BibTeX:

[More Information Needed]

APA:

[More Information Needed]

Glossary [optional]

<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->

[More Information Needed]

More Information [optional]

[More Information Needed]

Model Card Authors [optional]

[More Information Needed]

Model Card Contact

[More Information Needed]

Framework versions

• PEFT 0.18.0
• TRL: 0.26.0
• Transformers: 4.57.3
• Pytorch: 2.9.0+cu126
• Datasets: 4.4.1
• Tokenizers: 0.22.1