> ## Documentation Index > Fetch the complete documentation index at: https://docs.sglang.io/llms.txt > Use this file to discover all available pages before exploring further. # Environment Variables > Configure SGLang diffusion behavior with environment variables. ## Runtime
Environment Variable Default Description
SGLANG\_DIFFUSION\_TARGET\_DEVICE cuda Target device for inference (cuda, rocm, xpu, npu, musa, mps, cpu)
SGLANG\_DIFFUSION\_ATTENTION\_BACKEND not set Override attention backend via env var (e.g. fa, torch\_sdpa, sage\_attn)
SGLANG\_DIFFUSION\_ATTENTION\_CONFIG not set Path to attention backend configuration file (JSON/YAML)
SGLANG\_DIFFUSION\_STAGE\_LOGGING false Enable per-stage timing logs
SGLANG\_DIFFUSION\_SERVER\_DEV\_MODE false Enable dev-only HTTP endpoints for debugging
SGLANG\_DIFFUSION\_TORCH\_PROFILER\_DIR not set Directory for torch profiler traces (absolute path). Enables profiling when set
SGLANG\_DIFFUSION\_CACHE\_ROOT \~/.cache/sgl\_diffusion Root directory for cache files
SGLANG\_DIFFUSION\_CONFIG\_ROOT \~/.config/sgl\_diffusion Root directory for configuration files
SGLANG\_DIFFUSION\_LOGGING\_LEVEL INFO Default logging level
SGLANG\_DIFFUSION\_WORKER\_MULTIPROC\_METHOD fork Multiprocess context for workers (fork or spawn)
SGLANG\_USE\_RUNAI\_MODEL\_STREAMER true Use Run:AI model streamer for model loading
## Platform-Specific ### Apple MPS
Environment Variable Default Description
SGLANG\_USE\_MLX not set Set to 1 to enable MLX fused Metal kernels for norm ops on MPS
### ROCm (AMD GPUs)
Environment Variable Default Description
SGLANG\_USE\_ROCM\_VAE false Use AITer GroupNorm in VAE for improved performance on ROCm
SGLANG\_USE\_ROCM\_CUDNN\_BENCHMARK false Enable MIOpen auto-tuning for VAE conv layers on ROCm
### Quantization
Environment Variable Default Description
SGLANG\_DIFFUSION\_FLASHINFER\_FP4\_GEMM\_BACKEND not set FlashInfer FP4 GEMM backend for generic NVFP4 fallback
## Caching Acceleration These variables configure caching acceleration for Diffusion Transformer (DiT) models. SGLang supports multiple caching strategies - see [caching documentation](./caching-acceleration) for an overview. ### Cache-DiT Configuration See [cache-dit documentation](./cache_dit) for detailed configuration.
Environment Variable Default Description
`SGLANG_CACHE_DIT_ENABLED` false Enable Cache-DiT acceleration
`SGLANG_CACHE_DIT_FN` 1 First N blocks to always compute
`SGLANG_CACHE_DIT_BN` 0 Last N blocks to always compute
`SGLANG_CACHE_DIT_WARMUP` 4 Warmup steps before caching
`SGLANG_CACHE_DIT_RDT` 0.24 Residual difference threshold
`SGLANG_CACHE_DIT_MC` 3 Max continuous cached steps
`SGLANG_CACHE_DIT_TAYLORSEER` false Enable TaylorSeer calibrator
`SGLANG_CACHE_DIT_TS_ORDER` 1 TaylorSeer order (1 or 2)
`SGLANG_CACHE_DIT_SCM_PRESET` none SCM preset (none/slow/medium/fast/ultra)
`SGLANG_CACHE_DIT_SCM_POLICY` dynamic SCM caching policy
`SGLANG_CACHE_DIT_SCM_COMPUTE_BINS` not set Custom SCM compute bins
`SGLANG_CACHE_DIT_SCM_CACHE_BINS` not set Custom SCM cache bins
### Cache-DiT Secondary Transformer For dual-transformer models (e.g., Wan2.2 with high/low-noise experts), these variables configure caching for the secondary transformer. Each falls back to its primary counterpart if not set.
Environment Variable Default Description
SGLANG\_CACHE\_DIT\_SECONDARY\_FN (from primary) First N blocks to always compute
SGLANG\_CACHE\_DIT\_SECONDARY\_BN (from primary) Last N blocks to always compute
SGLANG\_CACHE\_DIT\_SECONDARY\_WARMUP (from primary) Warmup steps before caching
SGLANG\_CACHE\_DIT\_SECONDARY\_RDT (from primary) Residual difference threshold
SGLANG\_CACHE\_DIT\_SECONDARY\_MC (from primary) Max continuous cached steps
SGLANG\_CACHE\_DIT\_SECONDARY\_TAYLORSEER (from primary) Enable TaylorSeer calibrator
SGLANG\_CACHE\_DIT\_SECONDARY\_TS\_ORDER (from primary) TaylorSeer order (1 or 2)
## Cloud Storage These variables configure S3-compatible cloud storage for automatically uploading generated images and videos.
Environment Variable Default Description
`SGLANG_CLOUD_STORAGE_TYPE` not set Set to `s3` to enable cloud storage
`SGLANG_S3_BUCKET_NAME` not set The name of the S3 bucket
`SGLANG_S3_ENDPOINT_URL` not set Custom endpoint URL (for MinIO, OSS, etc.)
`SGLANG_S3_REGION_NAME` us-east-1 AWS region name
`SGLANG_S3_ACCESS_KEY_ID` not set AWS Access Key ID
`SGLANG_S3_SECRET_ACCESS_KEY` not set AWS Secret Access Key
## CUDA Crash Debugging These variables enable kernel API logging and optional input/output dumps around diffusion CUDA kernel call boundaries. They are useful when tracking down CUDA crashes such as illegal memory access, device-side assert, or shape mismatches in custom kernels.
Environment Variable Default Description
SGLANG\_KERNEL\_API\_LOGLEVEL 0 Controls crash-debug kernel API logging. 1 logs API names, 3 logs tensor metadata, 5 adds tensor statistics, and 10 also writes dump snapshots.
SGLANG\_KERNEL\_API\_LOGDEST stdout Destination for crash-debug kernel API logs. Use stdout, stderr, or a file path. %i is replaced with the process PID.
SGLANG\_KERNEL\_API\_DUMP\_DIR sglang\_kernel\_api\_dumps Output directory for level-10 kernel API dumps. %i is replaced with the process PID.
SGLANG\_KERNEL\_API\_DUMP\_INCLUDE not set Comma-separated wildcard patterns for kernel API names to include in level-10 dumps.
SGLANG\_KERNEL\_API\_DUMP\_EXCLUDE not set Comma-separated wildcard patterns for kernel API names to exclude from level-10 dumps.