Model weights for a replication archive

This repository holds the exact model files behind the self-hosted results of the paper "Reproducible LLM-Based Measurement Depends on the Serving Stack". Each folder is one model as it was served by Ollama release 0.24.0: the Ollama manifest, byte for byte, and every blob the manifest names.

Verify

The SHA-256 of each folder's manifest file is the model digest recorded in every call record of the replication archive and printed in the paper's online appendix. Each blob is stored under its own SHA-256 (blobs/sha256-<hex>), so hashing a blob and comparing with its file name verifies it.

sha256sum qwen3.5-9b/manifest
sha256sum qwen3.5-9b/blobs/*

Run in Ollama

Copy the manifest to ~/.ollama/models/manifests/registry.ollama.ai/library/<name>/<tag> (for qwen3.5-9b: library/qwen3.5/9b) and the blobs to ~/.ollama/models/blobs/, then run ollama run <name>:<tag>. The paper's primary settings are temperature 0, top_p 1, top_k 1, seed 42, a 32,768-token context, and a 512-token output cap unless the archive records otherwise for a model.

Contents

Folder Ollama model Weight format Licence
qwen3.5-0.8b qwen3.5:0.8b Q8_0 GGUF Apache-2.0
qwen3.5-9b qwen3.5:9b Q4_K_M GGUF Apache-2.0
qwen3.5-35b qwen3.5:35b Q4_K_M GGUF Apache-2.0
qwen3.6-27b qwen3.6:27b Q4_K_M GGUF Apache-2.0
gemma4-31b gemma4:31b Q4_K_M GGUF Apache-2.0
granite4.1-8b granite4.1:8b Q4_K_M GGUF Apache-2.0
granite4.1-30b granite4.1:30b Q4_K_M GGUF Apache-2.0
nemotron-3-nano-30b nemotron-3-nano:30b Q4_K_M GGUF NVIDIA Nemotron Open Model License
ministral-3-14b ministral-3:14b Q4_K_M GGUF Apache-2.0
gpt-oss-20b gpt-oss:20b MXFP4 GGUF Apache-2.0

Licences

Each model keeps its upstream licence, and the licence text Ollama distributes with the model is included in its folder as the blob of media type application/vnd.ollama.image.license. Nine models are released under Apache-2.0. nemotron-3-nano:30b is released under the NVIDIA Nemotron Open Model License, whose text is in its folder. These files are redistributed unmodified.

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. ๐Ÿ™‹ Ask for provider support