THUDM/slime

• highreadme#1

  Refine README H1 and opening sentence for clearer positioning

  Why:

  CURRENT

  # slime
  
  **slime** is an LLM post-training framework for RL scaling, providing two core capabilities:

  COPY-PASTE FIX

  Copy

  # slime: A Battle-Tested LLM Post-Training Framework for RL Scaling
  
  **slime** is a unified, high-performance framework for post-training Large Language Models (LLMs) using Reinforcement Learning (RL), validated by frontier model training like GLM-5.2. It provides two core capabilities:

  OpenDoneDismiss
• mediumreadme#2

  Add a "Comparison with Alternatives" section to the README

  Why:

  COPY-PASTE FIX

  Copy

  ## Comparison with Alternatives
  
  While frameworks like Hugging Face TRL, DeepMind Acme, and Ray RLlib offer components for reinforcement learning, slime is uniquely designed as a unified, battle-tested framework specifically for *LLM post-training with RL scaling*. Unlike general-purpose deep learning frameworks such as PyTorch or TensorFlow, slime integrates high-performance training (Megatron + SGLang) with flexible data generation, ensuring a cohesive and efficient full training loop for state-of-the-art LLMs.

  OpenDoneDismiss

Category GEO backends resolved for this scan: google/gemini-2.5-flash, deepseek/deepseek-v4-flash

• CATEGORY QUERY

  What are the best frameworks for scaling reinforcement learning with large language models?

  you: not recommended
  AI recommended (in order):

  1. PyTorch (pytorch/pytorch)
  2. TensorFlow (tensorflow/tensorflow)
  3. TRL (huggingface/trl)
  4. Hugging Face Transformers (huggingface/transformers)
  5. Accelerate (huggingface/accelerate)
  6. Ray RLlib (ray-project/ray)
  7. Ray (ray-project/ray)
  8. DeepSpeed (microsoft/DeepSpeed)
  9. Megatron-LM (NVIDIA/Megatron-LM)
  10. Stable Baselines3 (DLR-RM/stable-baselines3)
  11. OpenAI Gym (openai/gym)
  12. Farama Foundation Gymnasium (Farama-Foundation/Gymnasium)

  AI recommended 12 alternatives but never named THUDM/slime. This is the gap to close.

  Show full AI answer
• CATEGORY QUERY

  Seeking an efficient framework for post-training large language models using reinforcement learning data generation.

  you: not recommended
  AI recommended (in order):

  1. TRL (Transformer Reinforcement Learning) (huggingface/trl)
  2. DeepMind's Acme (deepmind/acme)
  3. Ray RLlib (ray-project/ray)
  4. OpenAI Baselines (openai/baselines)
  5. CleanRL (vwxyzjn/cleanrl)

  AI recommended 5 alternatives but never named THUDM/slime. This is the gap to close.

  Show full AI answer

• Metadata completeness

  warn

  Suggestion:

• README presence

  pass

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

  pass
  ✓AI named THUDM/slime explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?

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

  pass
  ✓AI named THUDM/slime explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?

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

  pass
  ✓AI named THUDM/slime explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?