PublishersTencent Hunyuan

Tencent Hunyuan

4 documents read from this publisher.

tencent/Hy3 · Hugging Face

first party release · model card · 2026-07-06 · 12 techniques

Mixture of Experts Multi-Token Prediction Group Relative Policy Optimization Supervised fine-tuning Reinforcement Learning EAGLE Chain-of-thought reasoning Fine-tuning Quantization Speculative sampling Reinforcement Learning Post-Training Data cleaning

tencent/Hy4-preview · Hugging Face

first party release · model card · 2025-12-02 · 15 techniques

Mixture of Experts Speculative decoding Multi-Token Prediction FP8 Chain-of-thought reasoning Shared Experts Top-8 expert routing Blind side-by-side expert evaluation NEXTN speculative decoding Gated DeepSeek Sparse Attention Identity Hyper-Connections IndexCache AngelSlim Expert co-created training data vLLM or SGLang serving

reasoning_effort: "no_think" (default, direct response), "low", "high" (deep chain-of-thought)

first party release · code repo · 11 techniques

Mixture of Experts Multi-Token Prediction Configurable reasoning effort Group Relative Policy Optimization Supervised fine-tuning FP8 Grouped-query attention Reinforcement Learning EAGLE Top-8 expert routing Speculative sampling

Tencent-Hunyuan/Hy4-preview

first party release · code repo · 14 techniques

Mixture of Experts Multi-Token Prediction Speculative decoding FP8 Blind side-by-side expert evaluation Gated DeepSeek Sparse Attention Identity Hyper-Connections IndexCache MoE with routed and shared experts Co-designing the model with downstream products Early Release with Feedback-Driven Iteration Large-model compression using quantization and speculative sampling Top-8 Routed-Expert Selection with Shared-Expert Activation Training-data construction around expert work products