Loading profile…
DEVELOPER
@open-spark-jev
NVFP4 (MLP-only) build of spark-s1-4b-v8 for vLLM on Blackwell-class GPUs; agent-harness decisions with better-calibrated confidence than v6.
79.7%apache-2.0
fine-tuned from open-spark-jev/spark-s1-4b-v8
Qwen3.5-4B LoRA decision model for agent harnesses; one forward pass gives a probability per option. Open weights, GGUF builds for llama.cpp/Ollama.
129 msapache-2.0
NVFP4 (MLP-only) build of spark-s1-4b-v6 for vLLM on Blackwell-class GPUs: 1.4x faster than bf16 with no measurable JevBench public-item cost.
81.0%53.3 msapache-2.0
fine-tuned from open-spark-jev/spark-s1-4b-v6
Qwen3.5-4B decision model for agent-harness decisions (tool choice, finish, command risk, tool-output screening); far fewer confident wrong answers than v6.
92.8%apache-2.0
