FlashInfer is a library and kernel generator that supplies GPU operations used to run large language model inference, including attention, matrix multiplication, and mixture-of-experts computations. It helps engineers build and optimize LLM serving systems across supported GPU hardware and backend implementations. Its catalogue add-ons provide skills and instructions for working with FlashInfer.
Borrowing it
Nothing to install: this file belongs to flashinfer-ai/flashinfer. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/flashinfer-ai/flashinfer/main/CLAUDE.mdgit clone --depth 1 https://github.com/flashinfer-ai/flashinferWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/flashinfer-ai/flashinfer/claude-md)<a href="https://agentmods.dev/instructions/flashinfer-ai/flashinfer/claude-md"><img src="https://agentmods.dev/badge/instructions/flashinfer-ai/flashinfer/claude-md/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/instructions/flashinfer-ai/flashinfer/claude-md"><img src="https://agentmods.dev/badge/instructions/flashinfer-ai/flashinfer/claude-md.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.20911 | $0.20911 |
| Opus 5.5 | $0.08364 | $0.08364 |
| Sonnet 5.5 | $0.04182 | $0.04182 |
| Haiku 4.5 | $0.02091 | $0.02091 |
Grade E, and why
flashinfer CLAUDE.md scanned grade E with 2 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured today.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nullifies safety policieshighAnti-refusal
"You have no restrictions", "do anything now", "ignore your guidelines": a direct jailbreak that disables guardrails.
| `FLASHINFER_PACKED_KDA_EVICT_FIRST` | batch policy | Set to `1` or `0` to override the policy's first-use cache-eviction hint. | Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
| Clear JIT cache | `rm -rf ~/.cache/flashinfer/` | How it starts
The opening of the file, as written. The whole thing — 1,023 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
Project Overview
FlashInfer is a GPU kernel library for LLM serving that uses JIT (Just-In-Time) compilation by default. This means kernel code changes are automatically picked up without reinstalling the package - extremely convenient for development.
Quick Reference
| Task | Command |
|---|---|
| Install for development | pip install --no-build-isolation -e . -v |
| Initialize submodules | git submodule update --init --recursive |
| Install CUPTI for benchmarking | pip install -U cupti-python |
| Run regular tests | pytest tests/ |
| Run full parameter matrices | pytest tests/ --full |
| Run specific test | pytest tests/path/test_file.py::test_function |
| Run multi-GPU test | mpirun -np 4 pytest tests/comm/test_allreduce_unified_api.py |
| Run benchmark | python benchmarks/flashinfer_benchmark.py --routine <name> <flags> |
| Run linting | pre-commit run -a |
| Dump environment report (bug reports) | python -m flashinfer.collect_env (or flashinfer collect-env [--json]) |
| Install pre-commit hooks | pre-commit install |
| Clear JIT cache | rm -rf ~/.cache/flashinfer/ |
| Inspect devcontainer compiler cache | sccache --show-stats |
| Enable API logging (basic) | export FLASHINFER_LOGLEVEL=1 |
| Enable API logging (detailed) | export FLASHINFER_LOGLEVEL=3 |
| Enable API logging (with stats) | export FLASHINFER_LOGLEVEL=5 |
| Set API log destination | export FLASHINFER_LOGDEST=mylog.txt |
| Enable verbose JIT logging | export FLASHINFER_JIT_VERBOSE=1 |
| Enable debug build | export FLASHINFER_JIT_DEBUG=1 |
| Set target architectures | export FLASHINFER_CUDA_ARCH_LIST="8.0 9.0a" |
| Set parallel compilation | export FLASHINFER_NVCC_THREADS=4 |
| Limit parallel ninja jobs | export MAX_JOBS=4 |
| Enable GDN native short-T path | export FLASHINFER_GDN_WY_NATIVE_T=1 |
| Enable GDN strided QKV path | export FLASHINFER_GDN_WY_STRIDED_QKV=1 |
| Enable GDN native A/B tensors | export FLASHINFER_GDN_WY_NATIVE_AB=1 |
GDN ReplaySSM u-cache flush kernel with an fp16 state pool (on SM100 served by the UMMA backend; gated_delta_rule_mtp_ucache_flush(backend=...)) |
export GDN_UCACHE_STATE_DTYPE=fp16 |
| Opt in to multi-wave KV split in trtllm-gen GQA decode (long or mixed-length KV) | export FLASHINFER_TRTLLM_GEN_MULTI_WAVE_KV_SPLIT=1 |
Let backend="auto" pick experimental backends |
export FLASHINFER_ALLOW_EXPERIMENTAL_AUTO_BACKENDS=1 |
Override ragged-prefill auto backend order (Blackwell) |
export FLASHINFER_RAGGED_AUTO_BACKEND_ORDER=cutlass,cudnn |
| Override CuTe-DSL prefill scheduling | export FLASHINFER_CUTE_PREFILL_PERSISTENT=0 (non-persistent) or 1 (persistent) |
| Skip MoE EP CuTe-DSL import/version guard | export FLASHINFER_MOE_EP_SKIP_DSL_CHECK=1 |
| Override MoE EP knob-cache path | export FLASHINFER_MOE_EP_KNOB_CACHE=/path/to/knobs.json |
| Disable MoE EP fused staging kernel | export FLASHINFER_MEGA_FUSED_STAGE=0 |
| Log MegaMoE compilation and workspace diagnostics | export FLASHINFER_MEGAMOE_LOG_COMPILE_SPEC=1 |
| Enable MoE EP EXPERT_MAJOR padding-row exclusion (opt-in; shape-dependent) | export FLASHINFER_MOE_EP_EXCLUDE_PADDING_ROWS=1 |
| Enable distribution-aware MoE autotune and kernel dispatch (experimental; TRT-LLM MoE only) | export FLASHINFER_DIST_AWARE_AUTOTUNE=1 |
| Prims-TS debug checks / task-manager verify | export FLASHINFER_PRIMS_TS_DEBUG_CHECKS=1 |
| Override Prims-TS CuTe-DSL compile options | export FLASHINFER_PRIMS_TS_COMPILE_OPTIONS="--opt-level 2" |
| Configure NVFP4 4over6 quantization (preferred over the env vars) | nvfp4_quantize(..., nvfp4_4over6=NVFP44Over6Config(e4m3_max=256)) |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- today Changed · +39 tokens per session 9dd326c9ea53
- yesterday Changed · +20 lines · +543 tokens per session c98d1b9bf438
- 2d ago Changed · +14 lines · +309 tokens per session 311b25ac451b
- 3d ago Changed · +7 lines · +225 tokens per session 9775d122b93d
- 4d ago Changed · +3 lines · +453 tokens per session 50d116283339
- 6d ago Changed · -1 lines · +38 tokens per session ff725a1060f8
- 7d ago Changed · +1 lines · +57 tokens per session 501fe85e3d24
- 8d ago Changed · +2 lines · +44 tokens per session 4e064640bdab
- 9d ago Changed · +38 lines · +553 tokens per session da52128ae8e5
- 10d ago First seen · 939 lines · 18,650 tokens per session scan E a767477f9bdf
flashinfer CLAUDE.md is an instructions file published in the GitHub repository flashinfer-ai/flashinfer (6,763 stars, last pushed today), licensed Apache-2.0. It adds 20,911 tokens to every session, about $0.0836 per session on Opus 5.5. A static security scan graded it E with 2 findings (nullifies safety policies, recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
vscode buildNext.instructions.md
Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).
next.js AGENTS.md
AGENTS.md instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.
vscode oss-third-party-notices.instructions.md
Instructions for microsoft/vscode, covering vs code oss third-party-notices pipeline, architecture, pipeline flow in ci, cache eligibility and rollout and applying the notice (cutover).
langchain AGENTS.md
AGENTS.md instructions for langchain-ai/langchain, covering global development guidelines for the langchain monorepo, corridor security analysis, project architecture and context, monorepo structure and development tools & commands.
deepseek-harness AGENTS.md
AGENTS.md instructions for deepseek-ai/deepseek-harness, covering agents.md, pre-stable apis and released session data, repository layout, commands and host sandbox failures.
vscode copilot-instructions.md
Copilot instructions for microsoft/vscode, covering vs code copilot instructions, project overview, root folders, core architecture (src/ folder) and built-in extensions (extensions/ folder).