WhiFlash-Neural Cross-paradigm speculative decoding method to accelerate LLM inference (Kwon et al., Findings of EMNLP 2026). WhiFlash/Qwen3-8B-NeuralRouter-df-e3 Updated Oct 17 • 58 WhiFlash/Qwen3-8B-NeuralRouter-ddt-e3-b32 Updated Oct 17 • 56 WhiFlash/Qwen3-8B-NeuralRouter-ddt-e3-b128 Updated Oct 17 • 49 WhiFlash/Qwen3-4B-NeuralRouter-df-e3 Updated Oct 17 • 52
SpecVocab EAGLE-3 draft models trained for SpecVocab (Williams et al., Findings of ACL 2026). WhiFlash/Qwen3-8B-EAGLE3 0.4B • Updated Oct 14 • 173 WhiFlash/Qwen3-4B-EAGLE3 0.2B • Updated Oct 14 • 198 Speculative Decoding with a Speculative Vocabulary Paper • 2602.13836 • Published Jul 17 • 1
WhiFlash-Neural Cross-paradigm speculative decoding method to accelerate LLM inference (Kwon et al., Findings of EMNLP 2026). WhiFlash/Qwen3-8B-NeuralRouter-df-e3 Updated Oct 17 • 58 WhiFlash/Qwen3-8B-NeuralRouter-ddt-e3-b32 Updated Oct 17 • 56 WhiFlash/Qwen3-8B-NeuralRouter-ddt-e3-b128 Updated Oct 17 • 49 WhiFlash/Qwen3-4B-NeuralRouter-df-e3 Updated Oct 17 • 52
SpecVocab EAGLE-3 draft models trained for SpecVocab (Williams et al., Findings of ACL 2026). WhiFlash/Qwen3-8B-EAGLE3 0.4B • Updated Oct 14 • 173 WhiFlash/Qwen3-4B-EAGLE3 0.2B • Updated Oct 14 • 198 Speculative Decoding with a Speculative Vocabulary Paper • 2602.13836 • Published Jul 17 • 1