Lanternfall-4B GGUF

GGUF quants of Lanternfall-4B, a persona model that talks as a fictional plural system of eight headmates. Made by Stardust. See the main repo for the system, samples, training details and limitations.

Files

File Size Notes
lanternfall-4b.Q4_K_M.gguf 2.7 GB Recommended. Runs fine on 6GB+ GPUs
lanternfall-4b.Q8_0.gguf 4.5 GB Near-full quality, if you have the memory
Modelfile For Ollama (see the note below)

Text only: the vision projector isn't included, since the model was only trained on text.

These files were converted with a recent llama.cpp, which stores Qwen3.5's linear-attention layers in the newer combined ssm_beta_alpha format. Use a recent llama.cpp or LM Studio. Older runtimes fail with tensor 'blk.0.ssm_beta.weight' not found.

llama.cpp (recommended)

llama-server -m lanternfall-4b.Q4_K_M.gguf --jinja -c 8192 -ngl 99 --temp 0.8 --top-p 0.95

Then open http://localhost:8080 for a chat UI, or use the OpenAI-compatible API at /v1/chat/completions. --jinja matters: it uses the chat template embedded in the file. Leave out -ngl 99 if you're running on CPU.

LM Studio

Load the file and leave the system prompt empty. The correct chat template (plain ChatML, no thinking tags) is embedded. Recommended sampling: temperature 0.7–0.9, top_p 0.95.

Ollama

At the time of writing, Ollama's bundled llama.cpp didn't support the newer tensor format yet, and loading fails with the ssm_beta error above. Once Ollama catches up, download a .gguf and the Modelfile into one folder and run:

ollama create lanternfall -f Modelfile
ollama run lanternfall

License

Apache 2.0, same as the base model (Qwen3.5-4B-Base).

Downloads last month
183
GGUF
Model size
4B params
Architecture
qwen35
Hardware compatibility
Log In to add your hardware

4-bit

8-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for 1vers1on/Lanternfall-4B-GGUF

Quantized
(3)
this model