sweelol/kd-gemma3-pruned-dolly

This model is part of the Sweelol AI Hub, a research project focused on efficient fine-tuning of modern language models on Kaggle accelerators.

Full Research Notebook & Benchmark Results: [Coming soon]

This model is part of the Sweelol AI Hub collection, resulting from experiments in efficient fine-tuning, optimization strategies and knowledge distillation on the Gemma-3-270m architecture using the Databricks Dolly-15k dataset on Kaggle TPUs/GPUs.

• Developed by: Sweelol AI
• Shared by: Sweelol AI
• Model type: Causal Language Model
• Language(s) (NLP): English
• License: Apache 2.0
• Base Model: google/gemma-3-270m

Model Description

This model is part of the Sweelol AI Hub collection, resulting from experiments in efficient fine-tuning and knowledge distillation on the Gemma-3-270m architecture using the Databricks Dolly-15k dataset on Kaggle TPUs/GPUs.

Key Details:

• Base Model: google/gemma-3-270m
• Training Data: Databricks Dolly-15k (subset)
• Fine-Tuning Method: Knowledge Distillation
• Purpose: Knowledge Distillation on TPU

Model Sources

• Repository: https://huggingface.co/sweelol/kd-gemma3-pruned-dolly
• GitHub:

Uses

How to Get Started with the Model

from transformers import AutoModelForCausalLM, AutoTokenizer
# For PEFT models like LoRA or Prompt Tuning, you will also need:
# from peft import PeftModel

# This is the repository ID for your specialized model
model_id = "sweelol/kd-gemma3-pruned-dolly"

# For Full Fine-Tuned models:
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype="auto")
tokenizer = AutoTokenizer.from_pretrained(model_id)

# For PEFT models (LoRA, Prompt Tuning):
# base_model = AutoModelForCausalLM.from_pretrained("{base_model}", torch_dtype="auto")
# model = PeftModel.from_pretrained(base_model, model_id)
# tokenizer = AutoTokenizer.from_pretrained(model_id)


# Example usage:
prompt = "Instruction:\nWhat is the capital of France?\n\nResponse:\n"
inputs = tokenizer(prompt, return_tensors="pt")

generate_ids = model.generate(inputs.input_ids, max_length=50)
result = tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]

print(result)

Evaluation

Testing Data & Metrics

The model was evaluated on a comprehensive suite of tasks from the lm-evaluation-harness, including 5 diverse subsets of MMLU (for academic reasoning) and HellaSwag (for common-sense reasoning). The primary metric is zero-shot accuracy on a 200-sample subset of each task's test split.

Results

This table summarizes the final benchmark scores for the sweelol/kd-gemma3-pruned-dolly model. It is compared against the original, un-tuned baseline model.

Benchmark Task	Sweelol KD-Pruned	Baseline (Gemma-3-270m)
Average MMLU (5 tasks)	23.98%	24.88%
HellaSwag (Common Sense)	33.00%	43.50%
----------------------------------	----------	----------
MMLU Sub-task Breakdown:
MMLU - High School Computer Science	26.00%	24.00%
MMLU - Formal Logic	25.40%	25.40%
MMLU - Professional Law	25.00%	27.00%
MMLU - High School Mathematics	21.50%	26.00%
MMLU - Abstract Algebra	22.00%	22.00%

Summary of Findings

• Mixed Performance: The Knowledge Distillation and Pruning process resulted in a model with a fascinating performance profile.
• Strengths: It showed a notable improvement in High School Computer Science, suggesting the fine-tuning process was effective for that specific domain.
• Weaknesses: The model showed a significant decrease in performance on HellaSwag and High School Mathematics compared to the baseline. This indicates that the distillation process, while teaching the target task, may have resulted in a loss of the model's broader, pre-trained common-sense and numerical reasoning abilities (a phenomenon known as "alignment tax").

Full comparative results with other techniques can be found in our main research notebook linked at the top of this card.

Description

Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models. Gemma 3 models are multimodal, handling text and image input and generating text output, with open weights for both pre-trained variants and instruction-tuned variants. Gemma 3 has a large, 128K context window, multilingual support in over 140 languages, and is available in more sizes than previous versions. Gemma 3 models are well-suited for a variety of text generation and image understanding tasks, including question answering, summarization, and reasoning. Their relatively small size makes it possible to deploy them in environments with limited resources such as laptops, desktops or your own cloud infrastructure, democratizing access to state of the art AI models and helping foster innovation for everyone.

Inputs and outputs

• Input:

  • Text string, such as a question, a prompt, or a document to be summarized
  • Images, normalized to 896 x 896 resolution and encoded to 256 tokens each, for the 4B, 12B, and 27B sizes.
  • Total input context of 128K tokens for the 4B, 12B, and 27B sizes, and 32K tokens for the 1B and 270M sizes.

• Output:

  • Generated text in response to the input, such as an answer to a question, analysis of image content, or a summary of a document
  • Total output context up to 128K tokens for the 4B, 12B, and 27B sizes, and 32K tokens for the 1B and 270M sizes per request, subtracting the request input tokens

Citation

@article{gemma_2025,
    title={Gemma 3},
    url={https://arxiv.org/abs/2503.19786},
    publisher={Google DeepMind},
    author={Gemma Team},
    year={2025}
}

Model Data

Data used for model training and how the data was processed.

Training Dataset

These models were trained on a dataset of text data that includes a wide variety of sources. The 27B model was trained with 14 trillion tokens, the 12B model was trained with 12 trillion tokens, 4B model was trained with 4 trillion tokens, the 1B with 2 trillion tokens, and the 270M with 6 trillion tokens. The knowledge cutoff date for the training data was August 2024. Here are the key components:

• Web Documents: A diverse collection of web text ensures the model is exposed to a broad range of linguistic styles, topics, and vocabulary. The training dataset includes content in over 140 languages.
• Code: Exposing the model to code helps it to learn the syntax and patterns of programming languages, which improves its ability to generate code and understand code-related questions.
• Mathematics: Training on mathematical text helps the model learn logical reasoning, symbolic representation, and to address mathematical queries.
• Images: A wide range of images enables the model to perform image analysis and visual data extraction tasks.

The combination of these diverse data sources is crucial for training a powerful multimodal model that can handle a wide variety of different tasks and data formats.