{"slug":"openlair-moe-training","source_name":"openlair/moe-training","name":"Openlair/Moe Training","description":"Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace. Use when training large-scale models with limited compute (5× cost reduction vs dense models), implementing sparse architectures like Mixtral 8x7B or DeepSeek-V3, or scaling model capacity without proportional compute increase. Covers MoE architectures, routing mechanisms, load balancing, expert parallelism, and inference optimization.","version":1,"lift":{"pass_rate_delta_pts":18.18,"pass_rate_pct":90.9,"total_cases":22,"passed_cases":20,"tokens_delta_pct":182,"turns_delta_pct":0,"verdict":"mixed","benchmark_model":"gemini-3.6-flash","grading_method":"judged","completed_at":"2026-08-07T19:25:27.233827+00:00"},"skill_score":0.9091,"benchmark_models":[{"model":"gemini-3.6-flash","headline":true,"delta_pts":18.18,"with_pass_pct":90.9,"without_pass_pct":72.7,"tokens_delta_pct":182,"turns_delta_pct":0,"total_cases":22,"cases_aggregated":21,"verdict":"mixed","never_hurt":true,"completed_at":"2026-08-07T19:25:27.233827+00:00","run_id":"99f195a4-1064-457a-bd10-406473c6ccc4","version_number":1,"is_latest_version":true,"gate":null}],"trust":{"skill_safety":"passed","safety_status":"clean","intent_verdict":"safe","content_status":"clean","indexable":true},"license":"MIT","install_count":0,"manifest_hash":"c5cc2c13c991887c405d000f45fad02d9cdae702707c992b1a472ee845017a7e","raw_url":"https://app.decimal.ai/s/openlair-moe-training/SKILL.md","scorecard_url":"https://app.decimal.ai/skills/openlair-moe-training"}