JD.com Unveils JoyAI-LLM-Flash Model to Drive AI Progress

JD.com made its latest large language model, JoyAI-LLM-Flash, publicly available on the Hugging Face platform on February 14th. The model boasts a total of 4.8 billion parameters, with 3 billion activated during inference. It was pre-trained on 20 trillion text tokens, exhibiting strong proficiency in understanding contemporary knowledge, logical reasoning, and programming tasks.
JoyAI-LLM-Flash utilizes a novel FiberPO optimization framework, which incorporates fiber bundle theory into reinforcement learning. Combined with the Muon optimizer and dense multi-token prediction (MTP) technology, this approach effectively addresses the instability challenges common when scaling traditional models. Compared to versions without MTP, its throughput has increased by 1.3 to 1.7 times, significantly boosting both training efficiency and practical application potential.
The model is built on a mixture-of-experts (MoE) architecture, featuring 40 layers. It supports a context length of 128K tokens and has a vocabulary of 129K, representing a substantial step forward for JD.com in the field of artificial intelligence.
Related article
Stripe to Acquire OpenRouter as AI Model Routing Gains Momentum
Stripe has agreed to acquire OpenRouter, an AI model-routing platform that gives developers access to hundreds of models through a single interface. The deal adds model selection and routing to Stripe’s existing work around AI usage and token-based b
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber
Related Special Topic Recommendations
Comments (1)
0/500

JD.com made its latest large language model, JoyAI-LLM-Flash, publicly available on the Hugging Face platform on February 14th. The model boasts a total of 4.8 billion parameters, with 3 billion activated during inference. It was pre-trained on 20 trillion text tokens, exhibiting strong proficiency in understanding contemporary knowledge, logical reasoning, and programming tasks.
JoyAI-LLM-Flash utilizes a novel FiberPO optimization framework, which incorporates fiber bundle theory into reinforcement learning. Combined with the Muon optimizer and dense multi-token prediction (MTP) technology, this approach effectively addresses the instability challenges common when scaling traditional models. Compared to versions without MTP, its throughput has increased by 1.3 to 1.7 times, significantly boosting both training efficiency and practical application potential.
The model is built on a mixture-of-experts (MoE) architecture, featuring 40 layers. It supports a context length of 128K tokens and has a vocabulary of 129K, representing a substantial step forward for JD.com in the field of artificial intelligence.
Stripe to Acquire OpenRouter as AI Model Routing Gains Momentum
Stripe has agreed to acquire OpenRouter, an AI model-routing platform that gives developers access to hundreds of models through a single interface. The deal adds model selection and routing to Stripe’s existing work around AI usage and token-based b
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber





Home






