Gosset Works

Gosset Runner v1 (Windows x64)

One-click Windows desktop app and local OpenAI/Ollama gateway for Gosset Packed E8 models.

No system Python install, no CUDA Toolkit setup, and no manual pip dependency troubleshooting required. The single installer bundles an isolated Python 3.11.6 interpreter, pinned PyTorch (2.12.0+cu128), CuPy (14.2.0), Transformers (5.11.0), huggingface-hub (1.33.0), minimal CUDA FP16/BF16 NVRTC headers, and a native .NET 8 + WebView2 desktop shell.

Download Windows x64 Installer (v1.0.3):

  • File: GossetRunnerSetup-1.0.3-win64.exe
  • Size: 2,843,379,294 bytes (2.65 GiB)
  • SHA-256: 8924dffb003fb27b74d4b396bd0b9e8f1bc14c70802fe9558542ab1dd86d5740
  • Earlier v1.0.x Users (In-App Hot-Patch): runner-patch-1.0.3.zip (114,077 bytes, SHA-256 79b6318f6a7a06fc317dfab811628ea7125300de37c523919cf77e4cd4c181c5; open Settings & Logs and click Check for Runner Updates)

Core Features

  • First-Run GPU & VRAM Check: Detects NVIDIA GPUs, VRAM capacity, driver, and compute capability on launch and shows which Gosset packs fit 100% resident on your card.
  • Pack Manager with Resume & Checksum Guard: Downloads official GossetWorks packs directly from Hugging Face with HTTP Range pause/resume, verifies every file against SHA256SUMS.txt before first launch, supports importing an existing local bundle folder, and updates small runtime scripts (run_e8_chat.py) without re-downloading ~13 GB weight shards.
  • Step Zero Defaults Built In: Launches 27B hybrid packs at 32,768 context (FP8 KV cache) and 2,048 max reply tokens by default (4,096 safe context on dense 64-layer Qwen3-32B for 16 GB cards).
  • Desktop Chat + Local API: Streams replies directly in the built-in desktop UI and exposes an OpenAI and Ollama compatible endpoint on http://127.0.0.1:11434/v1 (with stream=True SSE and stream=False support).

Supported Gosset Packs (v1)

Pack Repo Peak VRAM Default Context RTX 5080 Decode
Arete-27B Gosset E8 GossetWorks/Arete-27B-Gosset-3.65BPW 13.48 GiB 32,768 (FP8 KV) 53.92 tok/s
Qwen3.8-27B Gosset E8 GossetWorks/Qwen3.8-27B-Gosset-3.65BPW 13.48 GiB 32,768 (FP8 KV) 54.64 tok/s
Qwen3-32B Gosset E8 GossetWorks/Qwen3-32B-Gosset-3.65BPW 14.00 GiB 4,096 (16 GB cap) 38.4 - 40.0 tok/s

License & Patent Notice

Distributed under the Gosset Evaluation & Research License (LICENSE.md), protected under U.S. Provisional Patent Application No. 64/193,785. Strictly for non-commercial evaluation and research; commercial use requires a written license from licensing@gossetworks.com. Packs and Runner contain compiled execution artifacts only (no quantization encoder ships).

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