daydreamwarrior/Nemotron-Research-GooseReason-4B-Instruct-heretic-v2
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Mar 17, 2026License:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Warm

The daydreamwarrior/Nemotron-Research-GooseReason-4B-Instruct-heretic-v2 is a 4 billion parameter instruction-tuned causal language model, a decensored version of NVIDIA's GooseReason-4B-Instruct. It is based on Qwen3-4B-Instruct and fine-tuned using Reinforcement Learning with Verifiable Rewards (RLVR) and the Golden Goose pipeline. This model excels in reasoning tasks across mathematics, programming, and STEM, achieving state-of-the-art results among 4B-Instruct models on 15 diverse benchmarks.

Loading preview...