NOVER1-Qwen3-4B-f32-GGUF

thinkwee/NOVER1-Qwen3-4B is a 4-billion parameter large language model fine-tuned from Qwen3-4B-Instruct-2507, designed for powerful general reasoning across a wide range of text-to-text tasks by leveraging NOVER (NO-VERifier Reinforcement Learning), a verifier-free reinforcement learning method that optimizes reasoning with a perplexity-based proxy reward instead of traditional rule-based verifiers or reward models. Trained with LoRA finetuning on a modified NOVEReason_5k_reasoning dataset with custom tags, NOVER1-Qwen3-4B excels at freeform text answers to complex reasoning problems, offering enhanced performance in logical inference, instruction following, and other challenging benchmarks within the Qwen3 multilingual, hybrid-thinking architecture.

Model Files

File Name	Quant Type	File Size
NOVER1-Qwen3-4B.BF16.gguf	BF16	8.05 GB
NOVER1-Qwen3-4B.F16.gguf	F16	8.05 GB
NOVER1-Qwen3-4B.F32.gguf	F32	16.1 GB
NOVER1-Qwen3-4B.Q2_K.gguf	Q2_K	1.67 GB
NOVER1-Qwen3-4B.Q3_K_L.gguf	Q3_K_L	2.24 GB
NOVER1-Qwen3-4B.Q3_K_M.gguf	Q3_K_M	2.08 GB
NOVER1-Qwen3-4B.Q3_K_S.gguf	Q3_K_S	1.89 GB
NOVER1-Qwen3-4B.Q4_K_M.gguf	Q4_K_M	2.5 GB
NOVER1-Qwen3-4B.Q4_K_S.gguf	Q4_K_S	2.38 GB
NOVER1-Qwen3-4B.Q5_K_M.gguf	Q5_K_M	2.89 GB
NOVER1-Qwen3-4B.Q5_K_S.gguf	Q5_K_S	2.82 GB
NOVER1-Qwen3-4B.Q6_K.gguf	Q6_K	3.31 GB
NOVER1-Qwen3-4B.Q8_0.gguf	Q8_0	4.28 GB

Quants Usage

(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)

Here is a handy graph by ikawrakow comparing some lower-quality quant types (lower is better):