🤖 chatbot-v2 — TinyLLaMA Instruction-Tuned Chatbot (LoRA)

chatbot-v2 is a lightweight, instruction-following conversational AI model based on TinyLLaMA and fine-tuned using LoRA adapters. It has been trained on a carefully curated mixture of open datasets covering assistant-like responses, code generation, summarization, safety alignment, and dialog reasoning.

This model is ideal for embedding into mobile or edge apps with low-resource inference needs or running via an API.

🧠 Base Model

Model: TinyLlama/TinyLlama-1.1B-Chat
Architecture: Decoder-only Transformer (GPT-style)
Fine-tuning method: LoRA (low-rank adapters)
LoRA Parameters:
- r=16
- alpha=32
- dropout=0.05
- Target modules: q_proj, v_proj

📚 Training Datasets

The model was fine-tuned on the following instruction-following, summarization, and dialogue datasets:

tatsu-lab/alpaca — Stanford Alpaca dataset
databricks/databricks-dolly-15k — Dolly instruction data
knkarthick/dialogsum — Summarization of dialogs
Anthropic/hh-rlhf — Harmless/helpful/honest alignment data
OpenAssistant/oasst1 — OpenAssistant dialogues
nomic-ai/gpt4all_prompt_generations — Instructional prompt-response pairs
sahil2801/CodeAlpaca-20k — Programming/code generation instructions
Open-Orca/OpenOrca — High-quality responses to complex questions

🔧 Intended Use

This model is best suited for:

Conversational agents / chatbots
Instruction-following assistants
Lightweight AI on edge devices (via server inference)
Educational tools and experiments

🚫 Limitations

This model is not suitable for production use without safety reviews.
It may generate inaccurate or biased responses, as training data is from public sources.
It is not safe for sensitive or medical domains.

💬 Example Prompt

Instruction:

Explain the difference between supervised and unsupervised learning.

Response:

Supervised learning uses labeled data to train models, while unsupervised learning uses unlabeled data to discover patterns or groupings in the data…

📥 How to Load the Adapters

To use this model, load the base TinyLLaMA model and apply the LoRA adapters:

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base_model = AutoModelForCausalLM.from_pretrained(
    "TinyLlama/TinyLlama-1.1B-Chat",
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("TinyLlama/TinyLlama-1.1B-Chat")

model = PeftModel.from_pretrained(base_model, "sahil239/chatbot-v2")

📄 License

This model is distributed under the Apache 2.0 License.

🙏 Acknowledgements

Thanks to the open-source datasets and projects: Alpaca, Dolly, OpenAssistant, Anthropic, OpenOrca, CodeAlpaca, GPT4All, and Hugging Face.

sahil239
/

chatbot-v2