metadata
library_name: residuals
base_model: Qwen/Qwen2.5-14B
base_model_relation: adapter
instruct_model: Qwen/Qwen2.5-14B-Instruct
pipeline_tag: text-generation
tags:
- residuals
- delta
- task-arithmetic
- finetune
Instruction Residuals
This repository contains instruction residuals (delta weights) computed as the parameter-wise difference between Qwen/Qwen2.5-14B-Instruct and Qwen/Qwen2.5-14B.
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("Qwen/Qwen2.5-14B")
tok = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-14B")
res = Residuals.from_pretrained("residuals/qwen2.5-14b")
res.apply(base, base_tokenizer=tok)
Provenance
- Created at: 2025-10-25T16:18:30.099054+00:00
- DType: float32
- Parameters: 579
- Shapes hash: b3fe0942a175bf28daab2610457f75a24fbeca9d5c4cad1c951e9beac71b6a53
- Names hash: 155db82de40585f183c6134b5f8843abbe195fbe1997d967cd5ab9d22d54dc2e
- Base model:
Qwen/Qwen2.5-14B - Instruction model:
Qwen/Qwen2.5-14B-Instruct
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.