Downloads · 30 days
26
30% of all-time downloads
chimingw/qwen3.8-27b-ud-q5-k-xl-llamafile
qwen3.8-27b-ud-q5-k-xl-llamafile is a image-text-to-text model from chimingw. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
An unofficial, reproducible llamafile packaging of Unsloth's Qwen3.8-27B UD-Q5KXL GGUF. It embeds the main model, its FP16 vision projector, and the llamafile 0.10.5 runtime in one file; it is not a new training run o…
Downloads · 30 days
26
30% of all-time downloads
All-time downloads
88
Public
Repo size
21.5 GB
Likes
1
Public
Click a slice to open those files.
.llamafile21.5 GB · 100%
From the Hugging Face model README
An unofficial, reproducible llamafile packaging of Unsloth's Qwen3.8-27B UD-Q5_K_XL GGUF. It embeds the main model, its FP16 vision projector, and the llamafile 0.10.5 runtime in one file; it is not a new training run or a re-quantization.
Model-tree lineage: this llamafile → unsloth/Qwen3.8-27B-GGUF → Qwen/Qwen3.8-27B. The direct parent is the Unsloth GGUF repository whose already-quantized UD-Q5_K_XL weights are embedded here unchanged.
| Property | Value |
|---|---|
| Direct parent / GGUF source | unsloth/Qwen3.8-27B-GGUF |
| Original upstream base model | Qwen/Qwen3.8-27B |
| Architecture | Qwen3.8 native vision-language model (qwen3_5 GGUF architecture) |
| Parameters | 27B dense |
| Quantization | Unsloth Dynamic V3.0 UD-Q5_K_XL |
| Vision projector | mmproj-F16.gguf |
| Upstream native context | 262,144 tokens |
| Packaged default context | 8,192 tokens |
| Runtime | mozilla-ai/llamafile 0.10.5 |
| Package format | APE portable executable with uncompressed ZIP assets |
| Primary platforms | macOS and Linux |
| Hosted inference | Disabled; download and run locally |
| File | Size | SHA-256 |
|---|---|---|
Qwen3.8-27B-UD-Q5_K_XL.llamafile | 21,496,677,519 bytes (21.50 GB / 20.02 GiB) | 4e3c4c34f824d0ca448a097f9c9a4bbffd2a2fcb9ec0e1a91e2eb289d6fa8abe |
A llamafile is executable code plus model data, not merely a weights file. Verify the checksum before running a downloaded copy.
Qwen3.8-27B-UD-Q5_K_XL.ggufmmproj-F16.ggufllamafile 0.10.5The GGUF and projector are stored uncompressed with zipalign -j0. Neither was modified or re-quantized during packaging.
-m
/zip/Qwen3.8-27B-UD-Q5_K_XL.gguf
--mmproj
/zip/mmproj-F16.gguf
--ctx-size
8192
--jinja
--temp
1.0
--top-p
0.95
--top-k
20
--min-p
0.0
--presence-penalty
0.0
...
The final ... allows additional command-line arguments supplied by the user.
| Component | Source | SHA-256 |
|---|---|---|
| Main GGUF | unsloth/Qwen3.8-27B-GGUF/Qwen3.8-27B-UD-Q5_K_XL.gguf | 176a6a3f034e9cdc447c10cd00329fc9b31002e6589b9295f2ad4f1eefe0f6ab |
| Vision projector | unsloth/Qwen3.8-27B-GGUF/mmproj-F16.gguf | cbb841a9ee0636b2ec172f5bb8df2ea8dfeb01e90fe7c6126581d662a0b4e43e |
| Final llamafile | This repository | 4e3c4c34f824d0ca448a097f9c9a4bbffd2a2fcb9ec0e1a91e2eb289d6fa8abe |
unzip -t completed successfully against the final artifact.
chmod +x Qwen3.8-27B-UD-Q5_K_XL.llamafile
./Qwen3.8-27B-UD-Q5_K_XL.llamafile
The default launch provides both a terminal chat prompt and a local Web UI/API server.
Keep the terminal process running, then open:
http://127.0.0.1:8080
127.0.0.1 keeps the server accessible only from the local machine.
./Qwen3.8-27B-UD-Q5_K_XL.llamafile \
--cli -p 'Explain the purpose of a llamafile.' -n 256
./Qwen3.8-27B-UD-Q5_K_XL.llamafile \
--server --host 127.0.0.1 --port 8080
The server exposes a local Web UI and OpenAI-compatible endpoints under http://127.0.0.1:8080/v1.
This package is intended for people who want a single downloadable executable containing:
It is particularly convenient for offline or self-contained macOS/Linux use where installing a separate inference runtime is undesirable. Because the model, projector, runtime, terminal chat, browser UI, and local API are bundled into one file, you can keep the llamafile on a sufficiently large USB drive or portable SSD and bring your local intelligence with you. On a compatible machine, that single file is all you need.
--ctx-size N, for example --ctx-size 32768; larger contexts require more RAM and may reduce performance.mmproj-F16.gguf projector is embedded and selected automatically by the packaged defaults. No external projector file is required.This artifact exceeds Windows' 4 GB executable limit and will not run as a Windows .exe.
The upstream Unsloth model card declares Apache-2.0. The embedded llamafile runtime is also Apache-2.0. Retain upstream notices and attribution when redistributing.