miles/examples

Examples

These examples are runnable starting points for your own RL workflow. A few are purely demonstrative, but most are verifiable against a concrete performance score.

Recipes

End-to-end training workflows — the place to start.

• geo3k_vlm: Training VLMs with FSDP using GRPO on the GEO3K dataset.
  • multi_turn: The same dataset over multiple turns, with the model cropping images through an interactive environment.
• lora: LoRA fine-tuning with the Megatron backend.
• multi_lora: Fully-async multi-adapter LoRA training with a slot-keyed adapter page table.
• multi_policy: Two policies in one run — a solver answering gsm8k, and a verifier scored on ruling correctly about the solver's answers.
• on_policy_distillation: Teacher–student distillation on the student's own rollouts, run inside the on-policy training loop.
  • qwen3_5_35b_selfdistill: Two-phase self-distillation of Qwen3.5-35B-A3B on one 8xH200 node, with an in-process Megatron teacher.
• ppo: Actor-critic PPO with GAE advantages, where the critic shares the actor's train GPUs.
• retool_v2: Tool-enabled language model generation with sandboxed Python code execution interleaved with thinking.
• swe-agent-harbor-docker: Trains coding and terminal agents with Harbor-managed local Docker sandboxes and verifier rewards.

Infra Features

Runtime and infrastructure plumbing rather than training recipes — how miles moves data and weights around.

• fully_async: Demonstrates fully asynchronous rollout generation for higher efficiency.
• hot_restart: Replaces the orchestration script and rollout executor of a live run, keeping trainers and engines up.
• low_precision: Examples of FP8 training and inference, plus INT4 QAT, for improved throughput and stability.
• p2p_weight_transfer: Point-to-point weight transfer between training and rollout engines.
• random_async: Dataset-free stress test of the async rollout ↔ trainer loop.
• split_deployment: Installs one run as several helm releases — trainer, engines and orchestration script apart.
• train_infer_mismatch_helper: Algorithmic methods for rollout correction (e.g., TIS, MIS).
• true_on_policy: Ensures strictly equal log probabilities between inference (SGLang) and training engines.

Experimental

Not fully verified — for experimental and development use.

• agentenv: Rollouts against AgentENV, a self-hosted platform running agent sandboxes on Firecracker microVMs.
• DrGRPO: Custom reducer for Dr.GRPO algorithm.
• eval: Documentation and setup for evaluation environments using NeMo-Skills.
• eval_multi_task: Example for supporting OOD evaluation tasks, e.g., GPQA, IFBench.
• formal_math: Examples related to formal math reasoning tasks, including a single round demo.
• multi_agent: Example of running multi-agent RL with miles.
• nemo-gym: SWE-agent training with NVIDIA NeMo Gym as the environment ecosystem.
• openenv: Rollouts against OpenEnv-hosted environments.
• reproducibility: Guides on achieving bitwise experiment reproduction using deterministic modes.
• search-r1: A minimal reproduction of Search-R1, featuring multi-turn conversation and tool-calling.
• strands_sglang: Integration example with the Strands-Agents scaffolding framework.
• swe-agent-harbor-daytona: The swe-agent-harbor-docker pipeline with task sandboxes hosted on Daytona instead of local Docker.
• tau-bench: Training in an agentic multi-turn tool use environment (Tau-bench).
• verifiers: Training on a Prime Intellect Verifiers environment instead of a Miles prompt dataset.

These READMEs are the documentation site

Every README outside experimental/ is mirrored onto miles.radixark.com/docs/examples by scripts/tools/sync_example_docs.py, which pre-commit runs for you. The docs site is generated from this directory and never edited directly, so a new example needs nothing beyond its README and an entry in the list above — the sync fails if either is missing. Three things that list controls: the level-1 heading of each README becomes the page title, the one-line description becomes the page's meta description (keep it under 160 characters), and the bullet order is the sidebar order. Content between docs:exclude:start / docs:exclude:end HTML comments (like this section) stays on GitHub but is left out of the site.