Downloads · 30 days
0
qwen35-3090-cuda
edouardvallet/qwen35-3090-cuda
qwen35-3090-cuda is a machine learning model from edouardvallet. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Single-file Linux bundle containing the native C++20/CUDA chat TUI and the exact Qwen/Qwen3.5-9B checkpoint at revision c202236235762e1c871ad0ccb60c8ee5ba337b9a.
Access
Public
Updated Jul 20, 2026
Repo size
19.3 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
Other19.3 GB · 100%
At a glance
- Access
- Public
- Created
- Jul 20, 2026
- Updated
- Jul 20, 2026
- SHA
- e5514411
Base models
Model card
From the Hugging Face model README
qwen35-3090-cuda v3
Single-file Linux bundle containing the native C++20/CUDA chat TUI and the exact Qwen/Qwen3.5-9B checkpoint at revision c202236235762e1c871ad0ccb60c8ee5ba337b9a.
Source: https://github.com/edouardvallet/qwen35-3090-cuda/tree/v3
File: qwen35-linux
Size: 19,320,269,906 bytes
SHA256: d6dfd4a23d4396b761833e8116feda52f652c73871bff6a0c09c1762debffd8d
Download and run:
hf download edouardvallet/qwen35-3090-cuda qwen35-linux SHA256SUMS --local-dir .
sha256sum -c SHA256SUMS
chmod +x qwen35-linux
./qwen35-linux
The first launch verifies and extracts the tokenizer, model index, and four model shards into the local cache. Later launches reuse the verified cache. Set QWEN35_BUNDLE_CACHE to choose another cache directory.
Validated platform: Ubuntu 24.04 x86-64, NVIDIA RTX 3090 sm_86, CUDA 12, cuBLAS 12, ICU 74, and PCRE2 10.42. The NVIDIA driver and those runtime libraries are not embedded.
Defaults: 1024 total-token context, 512 maximum reply tokens, greedy decoding, one active sequence. Enter sends; Shift+Enter inserts a line break; /clear resets history; /quit exits.
Linux is validated. This repository does not claim Windows CUDA compatibility or performance. The TUI throughput shown during use is not the published completion-server benchmark.
- Created
- Jul 20, 2026
- Updated
- Jul 20, 2026