POCKET-35B-GGUF

POCKET-35B-GGUF

Runs a Korean-tuned 35B model on your own PC from GGUF files with llama.cpp, with a CPU-only path so no GPU is needed. For people who want a local language model.

Made by FINAL-Bench on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: POCKET-35B-GGUF holds GGUF builds of a language model tuned for Korean, made to run on an ordinary PC through llama.cpp. The model is a sparse mixture of experts with 34.66B parameters, and about 3B of them are active for each token. A CPU-only path means a GPU is optional.

Who it suits: It suits people who want a Korean-capable language model running on their own computer. The card's own start point is the Darwin-36B-Opus checkpoint.

Title: Who runs it
Industry | Who uses it | How they use it | Example
Software | Developers | Run a local Korean-capable model without a GPU | A chat helper inside a desktop tool
Education | Korean teachers and learners | Practise with a model that stays on one machine | Ask questions in Korean on a laptop
IT | Teams that keep text in-house | Host a model on their own PCs | A local assistant for staff drafts

If you want a model for a phone, you can skip it. The card sends phone users to separate repositories of about 5 GB.

Where to use it:

• llama.cpp, with support for the qwen35moe architecture, on Windows, macOS or Linux
• A PC with enough RAM for the build you pick
• CPU-only machines, since the card lists a CPU-only execution path

Title: Jobs it does
Use case | Who | What they get back
Local chat in Korean | Individuals | Replies generated on your own PC
Text work kept on one machine | Small teams | Drafts and answers with no cloud model involved

How to set it up:

• Python, for the download tool
• A recent llama.cpp build that supports qwen35moe, so check the version you have installed
• Memory for the build you choose, where the recommended first build, Q2_K, is about 13 GB

1. Install the Hugging Face Hub tools.
2. Download the repository files.
3. Run the build you picked with llama.cpp.

pip install -U huggingface_hub
hf download FINAL-Bench/POCKET-35B-GGUF

How to use it: Pick a build, start with the recommended Q2_K, and load its file in llama.cpp. The card lists several full-model builds with different file sizes and memory needs, so choose by how much RAM you have.

hf download FINAL-Bench/POCKET-35B-GGUF

Example: You load the Q2_K file in llama.cpp on a PC and ask a question in Korean. The model answers in Korean on your own machine. The card gives no sample output, so expect to judge the quality yourself.

For newcomers and experts: If you are new, GGUF is a file format that llama.cpp can read, and a smaller build uses less memory at some cost to quality. The first small thing to try is the recommended Q2_K build, about 13 GB.

If you know the field, the model is a sparse mixture of experts with roughly 3B active parameters per token, built from a Darwin-36B-Opus checkpoint. The trade-off is evidence: the card marks its results as publisher measurements under device-specific conditions.

Download a GGUF build
Check llama.cpp supports qwen35moe
Load the file on CPU or GPU
Ask in Korean

Good to know:

• Builds: this repository has the full-model PC builds, not the phone versions.
• Size: the recommended Q2_K build is about 13 GB.
• Runtime: llama.cpp needs qwen35moe support, so check your build.
• Evidence: results are the publisher's own measurements.
• License: Apache-2.0.
• Cost to run: no API key, account or paid service is named.
• Data it reaches: you download files from Hugging Face, and the card names no other access.
• Not for: phones, which have their own separate repositories.

On Hugging Face, 13 November 2026

• Task: text generation
• Library: llama.cpp
• Parameters: 34.7B
• Weights: about 17.3 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 452,730
• Likes: 91
• License: apache-2.0
• Based on: FINAL-Bench/Darwin-36B-Opus
• Tags: llama.cpp, gguf, conversational, on-device, mobile, iphone, android, local-llm, edge, mixture-of-experts, quantized, pocket

Links

• Model: https://huggingface.co/FINAL-Bench/POCKET-35B-GGUF
• Files: https://huggingface.co/FINAL-Bench/POCKET-35B-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download FINAL-Bench/POCKET-35B-GGUF

Then follow the usage in its model card. Nothing in this kit runs by itself.