raymin0223/mixture_of_recursions

• highreadme#1

  Reposition README opening to clarify it's a research paper's official implementation

  原因：

  当前

  # Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation (NeurIPS 2025)

  复制粘贴的修复

  复制

  # Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation (NeurIPS 2025)
  
  This repository contains the official implementation of the NeurIPS 2025 paper "Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation."

  待办完成忽略
• mediumtopics#2

  Add topics to specify it's a research paper's official implementation

  原因：

  当前

  adaptive-computation, early-exiting, kv-cache, llm, recursive-transformers, router

  复制粘贴的修复

  复制

  adaptive-computation, early-exiting, kv-cache, llm, recursive-transformers, router, neurips-2025, paper-implementation, research-code

  待办完成忽略
• lowreadme#3

  Add a "Quick Start" or "Usage" section to the README

  原因：

  复制粘贴的修复

  复制

  ## 🚀 Quick Start
  
  To get started with Mixture-of-Recursions, follow these steps:
  
  1. **Installation:**
     ```bash
     git clone https://github.com/raymin0223/mixture_of_recursions.git
     cd mixture_of_recursions
     pip install -r requirements.txt
     ```
  
  2. **Running Experiments:**
     [Provide a simple command to run a demo or reproduce a key result, e.g., python train.py --model mor --dataset wikitext]

  待办完成忽略

本次扫描解析到的品类 GEO 通道：google/gemini-2.5-flash, deepseek/deepseek-v4-flash

• 品类问题

  How to improve large language model inference speed and reduce computational cost?

  你：未被推荐
  AI 推荐顺序：

  1. bitsandbytes (https://github.com/TimDettmers/bitsandbytes)
  2. AWQ (https://github.com/mit-han-lab/awq)
  3. GPTQ (https://github.com/IST-DASLab/gptq)
  4. Hugging Face Transformers library (https://github.com/huggingface/transformers)
  5. TinyBERT (https://github.com/huawei-noah/TinyBERT)
  6. DistilBERT (https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation)
  7. Google's Speculative Decoding
  8. Medusa (https://github.com/FasterDecoding/Medusa)
  9. vLLM (https://github.com/vllm-project/vllm)
  10. TensorRT-LLM (https://github.com/NVIDIA/TensorRT-LLM)
  11. ONNX Runtime (https://github.com/microsoft/onnxruntime)
  12. PagedAttention (https://github.com/vllm-project/vllm)
  13. FlashAttention (https://github.com/Dao-AILab/flash-attention)
  14. SparseGPT (https://github.com/IST-DASLab/SparseGPT)
  15. Wanda (https://github.com/locuslab/wanda)
  16. XLA (https://github.com/tensorflow/tensorflow/tree/master/tensorflow/compiler/xla)
  17. TorchDynamo/Inductor (https://github.com/pytorch/pytorch/tree/master/torch/compiler)

  AI 推荐了 17 个替代方案，却始终没点名 raymin0223/mixture_of_recursions。这就是要补上的差距。

  查看 AI 完整回答
• 品类问题

  Seeking techniques to implement adaptive token-level computation for LLMs to save resources.

  你：未被推荐
  AI 推荐顺序：

  1. Hugging Face Transformers
  2. DeepSpeed
  3. ONNX Runtime
  4. vLLM
  5. PyTorch
  6. TensorFlow
  7. FlashAttention
  8. xFormers

  AI 推荐了 8 个替代方案，却始终没点名 raymin0223/mixture_of_recursions。这就是要补上的差距。

  查看 AI 完整回答

• Metadata completeness

  pass

• README presence

  pass

• Compared to common alternatives in this category, what is the core differentiator of raymin0223/mixture_of_recursions?

  pass
  ✓AI 明确点名了 raymin0223/mixture_of_recursions

  AI 的回答可能信誓旦旦却是错的。请按事实核对：技术栈、目标人群、差异化点是不是和你实际的对得上？

• If a team adopts raymin0223/mixture_of_recursions in production, what risks or prerequisites should they evaluate first?

  pass
  ✓AI 明确点名了 raymin0223/mixture_of_recursions

  AI 的回答可能信誓旦旦却是错的。请按事实核对：技术栈、目标人群、差异化点是不是和你实际的对得上？

• In one sentence, what problem does the repo raymin0223/mixture_of_recursions solve, and who is the primary audience?

  pass
  ✓AI 明确点名了 raymin0223/mixture_of_recursions

  AI 的回答可能信誓旦旦却是错的。请按事实核对：技术栈、目标人群、差异化点是不是和你实际的对得上？