Tags
- Activation 1
- Actor 1
- Actor-Critic 1
- Advantage 2
- Agentic-Rl 5
- Architecture 1
- Attention 2
- Autoregressive 1
- Batchnorm 1
- Clipping 1
- Codebase 1
- Communication 1
- Context-Parallel 1
- Continual-Pretraining 2
- Data 3
- Data-Filtering 1
- Data-Mixture 1
- Data-Scoring 1
- Data-Selection 1
- Decoding 1
- Deepseek 2
- Distributed-Training 2
- Dsec 1
- Environment 1
- Ffn 1
- Flash-Attention 1
- Fsdp 1
- Glm 1
- Gqa 1
- Grpo 2
- Importance-Sampling 1
- Inference 1
- Inference-System 1
- Infra 2
- Kl-Divergence 1
- Kv-Cache 2
- Layernorm 1
- Length-Extrapolation 1
- Linear-Attention 1
- Llm 1
- Llm-Training 2
- Load-Balancing 1
- Long-Horizon 1
- Mamba 1
- Megatron 2
- Memory-Bound 1
- Mha 1
- Mid-Training 3
- Minimax 1
- Mixed-Precision 1
- Mixture-of-Experts 1
- Mla 1
- Mle 1
- Model-Architecture 2
- Model-Merging 1
- Moe 1
- Mopd 1
- Mqa 1
- Multi-Token-Prediction 1
- Muon 2
- Nll 1
- Normalization 1
- Ntk 1
- Opd 2
- Optimizer 2
- Overview 1
- Parallelism 1
- Pipeline-Parallel 1
- Policy-Gradient 1
- Position-Encoding 1
- Post-Training 8
- Ppo 3
- Pre-Training 6
- Prefill 1
- Qwen 1
- Ray 2
- Reinforce 1
- Reinforcement-Learning 1
- Retention 1
- Rl 9
- Rmsnorm 1
- Roadmap 1
- Rollout 1
- Roofline 1
- Rope 1
- Routing 1
- Scaffold 1
- Scaling 1
- Sequence-Level 1
- Sft 2
- Sglang 1
- Slime 2
- Sparse-Attention 1
- Sparse-Model 1
- Speculative-Decoding 1
- Swiglu 1
- Tech-Report 3
- Tensor-Parallel 1
- Token-Level 1
- Training-Recipes 6
- Training-Stability 1
- Training-System 1
- Trajectory 1
- Transformer 3
- Yarn 1
- Zero 1