DeepSeek-V3.2
DeepSeekAs DeepSeek-V3.2-Exp, we merge reasoning, agent, and human alignment training into one RL stage.
specific method · filed under post-training
Mixed-domain reinforcement learning trains a model with GRPO across reasoning, agent, and human alignment data in a single RL stage.1 It balances performance across diverse domains while avoiding the catastrophic forgetting of multi-stage training.2
Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.
we merge reasoning, agent, and human alignment training into one RL stage— DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models
effectively balances performance across diverse domains while circumventing the catastrophic forgetting issues commonly associated with multi-stage training paradigms— DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models
Also called Mixed reinforcement learning, Mixed multi-domain reinforcement learning, mixed RL, Mixed RL Training.
5 quotes from 5 sources, one card per model, each opening with its strongest.
As DeepSeek-V3.2-Exp, we merge reasoning, agent, and human alignment training into one RL stage.
Post-training from this checkpoint (SFT, mixed RL, multi-domain on-policy distillation, and agentic RL) produces LFM3-1.7B.
Mixed reinforcement learning: many domains trained at once, from math and code to SQL and structured outputs.
many domains trained at once, from math and code to SQL and structured outputs