deepseek

paper ,  deepseek ,  engram ,  n-gram ,  context-aware-gating ,  conditional-memory ,  new-axis-of-sparsity ,  tokenizer-compression

google-colab ,  llm-finetuning ,  ocr ,  deepseek ,  deepseek-ocr ,  unsloth

ocr ,  deepseek ,  document-image-to-text ,  deepseek-ocr

deepseek ,  deepseek-prover ,  deepseek-prover-v2

paper ,  top-ml-papers-of-the-week ,  survey-paper ,  llm-survey ,  gnn ,  scaling-law ,  deepseek ,  reasoning-strategies ,  efficient-reasoning ,  stop-overthinking ,  a-mem ,  agentic-memory ,  deepmesh ,  deep-learning-is-not-so-mysterious-or-different ,  flora-bench

llm-inference ,  deepseek ,  inference-cost ,  deepseek-r1 ,  deepseek-v3

deepseek ,  3fs ,  fire-flyer-fs ,  file-system

flashattention ,  deepseek ,  flashmla ,  cutlass

mixture-of-experts ,  deepseek ,  deepep ,  expert-parallelism

fp8 ,  deepseek ,  cutlass ,  deepgemm ,  nvidia-hopper ,  gemm-kernel ,  grouped-gemm ,  tensor-memory-accelerator

mixture-of-experts ,  deepseek ,  expert-parallelism ,  expert-parallelism-load-balancer

deepseek ,  dualpipe ,  computation-communication-overlap ,  1-forward-1-backward ,  zero-bubble-1-pipeline

deepseek ,  profiling-data ,  visualization ,  dualpipe ,  computation-communication-overlap

deepseek ,  llm-distillation ,  deepseek-r1 ,  deepseek-r1-zero ,  distilled-models

llm-reasoning ,  deepseek ,  o1-preview ,  deepseek-r1 ,  longer-reasoning

mixture-of-experts ,  deepseek-v2 ,  deepseek ,  open-weights