SenseNova-6.8-Code2Video GSQ-RCO GGUF

SenseNova-6.8-Code2Video GSQ-RCO GGUF

Runs a 26.9B parameter SenseNova-6.8-Code2Video model as a mixed precision GGUF file, in a terminal chat or a local server. For people who want that model on their own hardware.

Made by OliviaRossi 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: This is a GGUF quantization of sensenova's SenseNova-6.8-Code2Video, made by OliviaRossi. Rather than storing every weight at one precision, it gives each tensor its own type, using the GSQ and RCO method from ISTA-DASLab, in a profile called GSQ-RCO-IQ3_S. The model has 26.9B parameters. The card does not describe what the base model does, but its name and sample prompt point to video script work.

Who it suits: It suits people who run models locally from GGUF files and want a 26.9B model in a smaller form. Developers and video writers are the likeliest users.

Industry | Who uses it | How they use it | Example
Software | Developers | Run a local model server with an OpenAI compatible interface | llama-server listening on port 8080
Media | Video creators and script writers | Draft video scripts on their own machine | The sample prompt Generate a video script:
Research | Quantization researchers | Study a per tensor mixed precision recipe | Attention layers kept at higher precision than feed forward layers

You can skip it if you want the original unquantized model, since this file trades precision for size.

Where to use it:

• A terminal, through the llama-cli program
• A local server, through llama-server, which offers an OpenAI compatible interface
• Windows, macOS or Linux with Python, to download the file

Use case | Who | What they get back
Chat in a terminal | Anyone with a GGUF runtime | Replies to a typed prompt
Serve a local API | Developers | An endpoint other tools can call on port 8080
Draft a video script | Video writers | Script text written by the model

How to set it up:

• Python and pip, to download the file
• The llama-cli and llama-server programs, run from the folder that holds them
• Memory: Hugging Face estimates the size of a 4-bit copy of this model at 13.4 GB

1. Install the Hugging Face hub package.
2. Download the model files to your computer.
3. Point llama-cli or llama-server at the GGUF file with the commands below.

pip install -U huggingface_hub
hf download OliviaRossi/SenseNova-6.8-Code2Video-GSQ-RCO-GGUF

How to use it: Start a terminal chat or a local server with the file. The card shows both commands.

# Terminal Chat
./llama-cli -m SenseNova-6.8-Code2Video-GSQ-RCO-IQ3_S.gguf -p "Generate a video script:" -ngl 99

# OpenAI-Compatible Server
./llama-server -m SenseNova-6.8-Code2Video-GSQ-RCO-IQ3_S.gguf --port 8080 -ngl 99 -c 32768

Tool | What it does
./llama-cli | Chats with the model in a terminal
./llama-server | Starts a server with an OpenAI compatible interface
-m | Points to the GGUF file
-ngl 99 | Offloads up to 99 layers to the GPU
-c 32768 | Sets the context window to 32768 tokens
--port 8080 | Sets the port the server uses

Example: Run the llama-cli command with the prompt Generate a video script: and the file name as shown. The model continues from that prompt and writes script text in the terminal.

For newcomers and experts: Quantization stores a model's numbers with fewer bits, which makes the file smaller and easier to run at home. For a first try, download the file and run the llama-cli command, then change the prompt to your own topic.

A recipe called RCO picks a type for each tensor. Attention projections attn_qkv and attn_gate get IQ4_XS, Q5_K and Q6_K. Feed forward layers ffn_down and ffn_up get IQ2_S and IQ2_XXS. SSM and normalization layers stay at BF16 and F32. The tensor map and the importance matrix come from a Qwen3.8 27B setup.

The original SenseNova weights
RCO picks a type for each tensor
A mixed precision GGUF file is written
llama-cli or llama-server loads the file
You chat or call the local API

Good to know: The card gives no quality scores or comparison with the original model, so test it on your own tasks.

• License: Apache 2.0
• Cost to run: none named, you use your own hardware
• Data it reaches: files on your computer, and a server on port 8080 if you start llama-server
• Not for: judging quality from the card, which lists no benchmark results

On Hugging Face, 13 November 2026

• Library: transformers
• Parameters: 26.9B
• Weights: about 13.4 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 13,778
• Likes: 9
• License: apache-2.0
• Based on: sensenova/SenseNova-6.8-Code2Video
• Tags: transformers, gguf, llama.cpp, quantized, gsq-rco, iq3_s, non-uniform-quantization, code2video, imatrix

Links

• Model: https://huggingface.co/OliviaRossi/SenseNova-6.8-Code2Video-GSQ-RCO-GGUF
• Files: https://huggingface.co/OliviaRossi/SenseNova-6.8-Code2Video-GSQ-RCO-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download OliviaRossi/SenseNova-6.8-Code2Video-GSQ-RCO-GGUF

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