labs-molt vs slime
Molt: NVIDIA's agentic-first RL framework in ~9K lines — Ray for placement, vLLM for rollout, AutoModel + FSDP2 for training — fully async, multimodal, multi-turn, scaling to 1T-class MoE. — versus — THUDM's RL post-training framework behind the GLM releases — Megatron training plus SGLang rollouts with native arg pass-through, and pluggable reward, verifier and agentic data-generation workflows.
Both are research RL frameworks behind frontier-scale releases. slime is THUDM's Megatron + SGLang stack from the GLM line; Molt is NVIDIA's PyTorch-native, Ray + vLLM stack with a Gymnasium-shaped agent contract.
| labs-molt | slime | |
|---|---|---|
| Stars | 1.0k | 8.4k |
| Forks | 100 | 1.2k |
| Language | Python | Python |
| License | Apache-2.0 | Apache-2.0 |
| Last activity | yesterday | 2 days ago |
| Topics | training | training |
| Curated connections | 5 | 5 |
labs-molt — the curator's take
The RL stack to read end-to-end: the agent is the program (one Gymnasium-aligned Env.step() or ChatAgent.run(), reward is any Python you write), the trainer is a single actor with an optional KL reference, and Ray owns the async queues between rollout, training and weight sync. Token-first contract keeps ids, logprobs, action ranges, rewards and multimodal tensors aligned so multi-turn tool use and VLM environments share one format. Same script trains 8B and DeepSeek-V3-class MoE with expert parallelism. NOT for breadth — it deliberately has fewer algorithms and integrations than verl or TRL — and NOT for SFT-only work; it's for research groups running agentic RL who want every gradient one file away.
slime — the curator's take
One of the few open RL stacks proven on frontier releases (GLM-4.5 through 5.2, with Qwen/DeepSeek/Llama support): the Megatron+SGLang-only bet keeps the dataflow explicit and upstream engine features usable instead of flattened behind a multi-backend abstraction, and rollout-only/train-only debug paths take RL's silent-bug problem seriously. NOT an afternoon tool — you need Megatron-scale GPU infrastructure and RL literacy; for single-node SFT or LoRA use a lighter trainer. And if your rollout engine must be vLLM, this is the wrong framework by design.