residuals/gemma-3-1b

Instruction Residuals

This repository contains instruction residuals (delta weights) computed as the parameter-wise difference between google/gemma-3-1b-it and google/gemma-3-1b-pt.

Apply these residuals to the base model to reconstruct the instruction-tuned weights without retraining.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
from residuals import Residuals

base = AutoModelForCausalLM.from_pretrained("google/gemma-3-1b-pt")
tok = AutoTokenizer.from_pretrained("google/gemma-3-1b-pt")

res = Residuals.from_pretrained("residuals/gemma-3-1b")
res.apply(base, base_tokenizer=tok)

Provenance

• Created at: 2025-10-25T18:43:38.287673+00:00
• DType: float32
• Parameters: 341
• Shapes hash: 50bd0b958c5645bee4bc25d35eaa9c3bba79e2f7144e49db51f1fcd37440dada
• Names hash: 54c326e0a2a310020e89db320f7c0d9a3cef063d8b1251f78cbf7c0db721b3cd
• Base model: google/gemma-3-1b-pt
• Instruction model: google/gemma-3-1b-it

Files

• model.safetensors: Serialized residual tensors (safetensors format).
• (optional) model.safetensors.index.json + shard files model-00001-of-000N.safetensors, ... for multi-part weights.
• config.json: Residuals metadata and provenance.
• tokenizer files: Saved tokenizer for compatibility.

About this format

These are additive residuals (task vectors). Applying them to the base model's parameters reconstructs the instruction-tuned model.

Tools

Generated with the residuals Python package. Install via: pip install residuals.

• PyPI: https://pypi.org/project/residuals/
• Source: https://github.com/omarish/residuals