Qwen3.5-0.8B Instruction QA LoRA (Dolly-15k)

Overview

This is a LoRA fine-tuned model based on Qwen/Qwen3.5-0.8B, trained for instruction-following and question answering using the Dolly-15k dataset.

The model is designed to generate structured responses given an instruction-style prompt in the format:

Instruction:

Response:


Important ⚠️

This repository contains LoRA adapter weights only, not the full base model.

You must load it together with the base model:

Qwen/Qwen3.5-0.8B


Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = "Qwen/Qwen3.5-0.8B"
adapter = "your-username/your-repo-name"

tokenizer = AutoTokenizer.from_pretrained(base_model)

model = AutoModelForCausalLM.from_pretrained(base_model)
model = PeftModel.from_pretrained(model, adapter)

prompt = "### Instruction:\nWhat is the capital of India?\n\n### Response:\n"
inputs = tokenizer(prompt, return_tensors="pt")

outputs = model.generate(
**inputs,
max_new_tokens=100,
temperature=0.7,
top_p=0.9
)

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


Training Details

  • Base Model: Qwen/Qwen3.5-0.8B
  • Method: LoRA (PEFT)
  • Dataset: databricks/databricks-dolly-15k
  • Task Format: Instruction → Response
  • Max Sequence Length: 1024
  • Batch Size: 16
  • Gradient Accumulation: 2
  • Effective Batch Size: 32
  • Epochs: 3
  • Learning Rate: 2e-4

Capabilities

  • Follows structured instruction prompts
  • Answers general knowledge questions
  • Generates concise and relevant responses
  • Works well with the "Instruction → Response" format

Limitations

  • Small model (0.8B) → limited reasoning ability
  • Can repeat patterns if prompt format is incorrect
  • Sensitive to prompt structure
  • Not trained for complex multi-step reasoning
  • Requires base model to function

Example

Instruction:

What is the capital of India?

Response:

The capital of India is New Delhi.


License

This model is released under the CC BY-NC 4.0 license.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Neural-Hacker/qwen3.5-0.8b-dolly-qa-lora

Adapter
(71)
this model

Dataset used to train Neural-Hacker/qwen3.5-0.8b-dolly-qa-lora