Downloads · 30 days
7
9% of all-time downloads
giocorte/fusion-charlie-travis
fusion-charlie-travis is a machine learning model from giocorte. Use it for the machine learning 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.
One Qwen3-4B fine-tune that plays two characters, selected entirely by the system prompt — no special tokens, stock tokenizer.
Downloads · 30 days
7
9% of all-time downloads
All-time downloads
76
Public
Repo size
21.1 GB
Likes
0
Public
Click a slice to open those files.
.safetensors8 GB · 76%
From the Hugging Face model README
One Qwen3-4B fine-tune that plays two characters, selected entirely by the system prompt — no special tokens, stock tokenizer.
You are Charlie, … → Charlie (PTSD panic NPC, inline action | speech)You are Travis, … → Travis (city-infra AI, inline up/down | speech)Trained with weighted SFT (action/grounding tokens up-weighted) + cross-character DPO (right-character response chosen, wrong-character rejected). Travis oversampled ×4 to balance the mix.
gguf/fusion-charlie-travis-fusion-v01.Q4_K_M-imatrix.gguf — quantized trained model for llama-servermerged/ — full 16-bit HF checkpoint (use this as the FastFlowLM / NPU input)llama-server -m fusion-charlie-travis-fusion-v01.Q4_K_M-imatrix.gguf
# set the system prompt to "You are Charlie, …" or "You are Travis, …"
Run on an AMD Ryzen AI box (NOT the training GPU):
flm convert merged/ --outfile fusion.flm
flm run fusion.flm --system "You are Charlie, …" --prompt "Hello"