inclusionAI/AReno

areno-build-singleturn-game-demo

Autonomously research, implement, train, evaluate, visualize, and submit a genuinely new AReno RLVR mini-game whose mechanically generated samples are independent state-to-single-output tasks. Use when the user requests a complete new game demo with real…

原文语言：英语

areno-model-adaptation

Add or debug an AReno model family, including config conversion, module construction, checkpoint load/save, text or multimodal inference, training backward, tensor parallelism, CUDA graph decode, and checkpoint round trips. Use only when model-specific…

原文语言：英语

areno-add-algorithm

Add or modify an AReno algorithm, trainer, loss, advantage calculation, role model, or algorithm-specific configuration. Use for framework-level SFT, DPO, GSPO, GRPO, PPO, or new optimization method development. Do not use merely to run an existing algorithm.

原文语言：英语

areno-build-agentic-workflow

Create or debug an AReno multi-turn agentic dataset, run_agent implementation, tool schemas, tool execution, reward, loss masks, interactive TUI game with OpenAI-compatible LLM inference, or agentic training example. Use for text or multimodal agentic…

原文语言：英语

areno-debug-runtime

Diagnose failed, hung, slow, OOM, NaN, illegal-memory-access, NCCL, compilation, rollout, or training runs in AReno. Use when runtime evidence must identify the first causal stage. Do not use for routine capacity planning without a failure.

原文语言：英语

areno-develop-kernel

Develop, optimize, debug, and validate an AReno CUDA, Triton, fused, attention, convolution, routing, or MoE operator. Use when changes touch areno/accel or a runtime kernel boundary and require forward, backward, dtype, layout, CUDA graph, or benchmark…

原文语言：英语

areno-profile-performance

Measure and diagnose AReno rollout, prefill, decode, training, checkpoint, role-switch, communication, or Python scheduling performance. Use when throughput or step time is slow and evidence from metrics, py-spy, or Nsight is required. Do not optimize before…

原文语言：英语

areno-run-serving

Start, validate, debug, and stop an AReno OpenAI-compatible serving endpoint. Use for areno serve commands, API probes, streaming, cancellation, cache or CUDA graph issues, and supported image or tool-call requests. Do not use for training jobs.

原文语言：英语

areno-run-training

Run, configure, retry, and validate AReno SFT, DPO, GSPO, GRPO, PPO, and agentic training. Use for training commands, dataset or reward setup, smoke validation, real-step execution, checkpoint saving, or failed training retries. Do not use for serving-only…

原文语言：英语

areno-tune-capacity

Fit an AReno training or rollout workload to available GPUs by tuning TP, mini-batch, batch, sample count, and rollout concurrency. Use for OOM prevention, memory headroom, smoke-infer, smoke-train, or tune-params requests. Do not change semantic token limits…

原文语言：英语

areno-validate-correctness

Compare an AReno branch, model, checkpoint, algorithm, scheduler, or kernel against a baseline. Use for regression validation, logprob or metric comparison, checkpoint round trips, and correctness gates. Keep performance reporting separate.

原文语言：英语