$ ls ~/models/
Models
Fine-tunes I train on my DGX Sparks and publish on Hugging Face in GGUF and EXL3 quants, with a write-up on X for each release. The cards below are that Hugging Face profile read live, so a new release shows up here as soon as it is public.
Javi0.3-2B-GGUF
GGUF fine-tune of Qwen/Qwen3.5-2B. LoRA-tuned with Unsloth for agent tool use. Trained on one DGX Spark.
Javi0.3-2B
Safetensors fine-tune of Qwen/Qwen3.5-2B. LoRA-tuned with Unsloth for agent tool use. Takes text and image inputs. Trained on one DGX Spark.
Javi0.3-4B-GGUF
GGUF fine-tune of Qwen/Qwen3.5-4B. LoRA-tuned with Unsloth for agent tool use. Trained on one DGX Spark.
Javi0.3-4B
Safetensors fine-tune of Qwen/Qwen3.5-4B. LoRA-tuned with Unsloth for agent tool use. Takes text and image inputs. Trained on one DGX Spark.
Javi0.2-4B (alpha)
A 4B agent model fine-tuned from Qwen3.5-4B to look things up instead of guessing, say so when it can't verify, and reason from first principles. Trained on one DGX Spark.
Qwen3-4B Pirate
Qwen3-4B-Instruct-2507 fine-tuned to answer everything in heavy pirate dialect, no system prompt needed. LoRA trained in 7m20s on one DGX Spark with Unsloth; 74.5 tok/s at Q4_K_M on Spark.