Loading...
Loading...
Found 25 Skills
PR-backed and current-main optimization manual for the `MiniMaxAI/MiniMax-M2` series, including M2, M2.1, M2.5, M2.7, and M2.7-highspeed. Use when Codex needs to recover, extend, or audit MiniMax-specific optimizations, TP QK norm/all-reduce behavior, parser contracts, distributed runtime behavior, quantized loading, or backend-specific validation.
Use when creating or revising model PR optimization history documents for SGLang, vLLM, or another serving framework that cite GitHub PRs. Requires manual, per-PR source-diff review and documentation of motivation, key implementation approach, most important code excerpts, reviewed files, and validation implications instead of generated or one-line summaries.
Replay-first debug flow for SGLang serving problems. Use when a live or recent server shows health-check failures, latency or throughput regressions, queue growth, timeouts, distributed stalls, crash dumps, wrong outputs after deploys, or PD/EP/HiCache issues, and the job is to turn the problem into a replay plus the right next debug tool.
Compact SGLang torch-profiler triage skill. Use when Codex should inspect an existing `trace.json(.gz)` or profile directory, trigger `sglang.profiler` against a live server, and return one compact report with kernel, overlap-opportunity, and fuse-pattern tables. Single-trace triage is enough for quick diagnosis; mapping+formal two-trace triage gives stronger overlap conclusions.
PR-backed and current-main optimization manual for `moonshotai/Kimi-K2*` and `moonshotai/Kimi-K2.5*` in SGLang. Use when Codex needs to recover, extend, or audit Kimi optimizations, including K2 router/MoE fast paths, K2 thinking Marlin paths, K2.5 wrapper/multimodal/runtime plumbing, W4AFP8/W4A16 quant tracks, parser contracts, LoRA coverage, and backend-specific validation.
Framework-independent LLM serving benchmark skill for comparing SGLang, vLLM, TensorRT-LLM, or another serving framework. Use when a user wants to find the best deployment command for one model across multiple serving frameworks under the same workload, GPU budget, and latency SLA.
Use when an SGLang, vLLM, or TensorRT-LLM serving/model optimization task needs prior model-family PR evidence. Query and read the PR-driven history docs under model-pr-optimization-history before choosing source paths, fast paths, kernel/fusion ideas, regression risks, or validation lanes.
Unified LLM torch-profiler triage skill for `sglang`, `vllm`, and `TensorRT-LLM`. Use it to inspect an existing `trace.json(.gz)` or profile directory, or to drive live profiling against a running server and return one three-table report with kernel, overlap-opportunity, and fuse-pattern tables.
Fast structured generation and serving for LLMs with RadixAttention prefix caching. Use for JSON/regex outputs, constrained decoding, agentic workflows with tool calls, or when you need 5× faster inference than vLLM with prefix sharing. Powers 300,000+ GPUs at xAI, AMD, NVIDIA, and LinkedIn.
SSH into host `h100_sglang`, enter Docker container `sglang_bbuf`, work in `/data/bbuf/repos/sglang`, and use the ready H100 remote environment for SGLang **diffusion** development and validation. Use when a task needs diffusion model smoke tests, Triton/CUDA kernel validation, torch.compile diffusion checks, or a safe remote copy for diffusion-specific SGLang changes.
Run an autonomous Humanize-governed vLLM SOTA performance loop for one LLM model: first perform the fixed fair vLLM/SGLang/TensorRT-LLM deployment search and benchmark, then start one RLCR loop that repeatedly decides the gap, profiles the current bottleneck, runs layer/kernel pipeline analysis, patches vLLM code, optionally uses ncu-report-skill for kernel evidence, and revalidates until vLLM matches or beats the best observed framework under the same workload and SLA.
Run an autonomous Humanize-governed SGLang SOTA performance loop for one LLM model: first perform the fixed fair SGLang/vLLM/TensorRT-LLM deployment search and benchmark, then start one RLCR loop that repeatedly decides the gap, profiles the current bottleneck, runs layer/kernel pipeline analysis, patches SGLang code, optionally uses ncu-report-skill for kernel evidence, and revalidates until SGLang matches or beats the best observed framework under the same workload and SLA.