FlashSVD checkpoints and BASIS experiment inputs

BASIS: download experiment inputs here

For the Additional Experiments, start with the BASIS data guide.

You need…	Go to…
C4 training data	Six frozen C4 streams: two backbones × three seeds
Q64/Q128 probe and held-out tensors	Probe inputs
Cached per-window gradients	Q128 · Original Q64
Exact frozen compressed Sources	16 paper checkpoints plus four LowRankArena imports
Tokenizers, original paths, and checksums	Tokenizers · Artifact manifest

The shared data and caches total approximately 4.4 GiB; the 16 hosted checkpoints total 178.6 GiB. Download only the Source files needed for your task. The guide includes selective download commands, loader instructions, checkpoint coverage, and the remaining Task 05 Source requests. Tasks 01–02 already include their numerical inputs in GitHub.

Original FlashSVD example checkpoints

This repository contains a small set of curated example low-rank checkpoints for FlashSVD-v1.5 benchmarking and integration tests. The current examples cover SVD-LLM v1, Basis Sharing, and DobiSVD variants.

Properties of these original example checkpoints:

• all uploaded checkpoints remain factorized / low-rank; they are not fused into dense tensors
• removable RoPE runtime buffers are omitted from the SVD-LLM examples
• each example is a self-contained HF export subfolder

Example loader:

from utils.model_utils import get_model_from_source
import torch

source = "Duke-CEI-SVD/FlashSVD/llama_7b/svdllm_v1/whitening_only_0.5_hf"
model, tok = get_model_from_source(source)
model = model.eval().to("cuda", dtype=torch.bfloat16)