chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

Move some quant args to its own section in environ variables doc (#15722)

Browse Source 
Signed-off-by: vincentzed <207368749+vincentzed@users.noreply.github.com>
This commit is contained in:
vincentzed
2025-12-23 23:11:08 -05:00
committed by GitHub
parent 6292c24437
commit ac320a6f04
1 changed files with 13 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
docs/references/environment_variables.md
View File

@@ -56,7 +56,6 @@ SGLang supports various environment variables that can be used to configure its
| Environment Variable | Description | Default Value |
| --- | --- | --- |
| `SGLANG_DEEPEP_BF16_DISPATCH` | Use Bfloat16 for dispatch | `"false"` |
| `SGLANG_MOE_NVFP4_DISPATCH` | Use nvfp4 for moe dispatch (on flashinfer_cutlass or flashinfer_cutedsl moe runner backend) | `"false"` |
| `SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK` | The maximum number of dispatched tokens on each GPU | `"128"` |
| `SGLANG_DEEPEP_LL_COMBINE_SEND_NUM_SMS` | Number of SMs used for DeepEP combine when single batch overlap is enabled | `"32"` |
@@ -68,22 +67,28 @@ SGLang supports various environment variables that can be used to configure its
| `SGLANG_CLIP_MAX_NEW_TOKENS_ESTIMATION` | Clip max new tokens estimation for memory planning | `4096` |
| `SGLANG_DETOKENIZER_MAX_STATES` | Maximum states for detokenizer | Default value based on system |
| `SGL_DISABLE_TP_MEMORY_INBALANCE_CHECK` | Disable checks for memory imbalance across Tensor Parallel ranks | Not set (defaults to enabled check) |
| `SGLANG_NVFP4_CKPT_FP8_GEMM_IN_ATTN` | Quantize q_b_proj from BF16 to FP8 when launching DeepSeek NVFP4 checkpoint | `false` |
| `SGLANG_PER_TOKEN_GROUP_QUANT_8BIT_V2` | Apply per token group quantization kernel with fused silu and mul and masked m | `false` |
## Model-Specific Options
| Environment Variable | Description | Default Value |
| --- | --- | --- |
| `SGLANG_USE_AITER` | Use AITER optimize implementation | `false` |
| `SGLANG_INT4_WEIGHT` | Enable INT4 weight quantization | `false` |
| `SGLANG_MOE_PADDING` | Enable MoE padding (sets padding size to 128 if value is `1`, often set to `1` in Docker builds) | `0` |
| `SGLANG_FORCE_FP8_MARLIN` | Force using FP8 MARLIN kernels even if other FP8 kernels are available | `false` |
| `SGLANG_ENABLE_FLASHINFER_FP8_GEMM` (deprecated) | Use flashinfer kernels when running blockwise fp8 GEMM on Blackwell GPUs. **DEPRECATED**: Please use `--fp8-gemm-backend=flashinfer_trtllm` instead. | `false` |
| `SGLANG_FLASHINFER_FP4_GEMM_BACKEND` | Select backend for `mm_fp4` on Blackwell GPUS | `` |
| `SGLANG_SUPPORT_CUTLASS_BLOCK_FP8` (deprecated) | Use Cutlass kernels when running blockwise fp8 GEMM on Hopper or Blackwell GPUs. **DEPRECATED**: Please use `--fp8-gemm-backend=cutlass` instead. | `false` |
| `SGLANG_CUTLASS_MOE` (deprecated) | Use Cutlass FP8 MoE kernel on Blackwell GPUs (deprecated, use --moe-runner-backend=cutlass) | `false` |
## Quantization
| Environment Variable | Description | Default Value |
| --- | --- | --- |
| `SGLANG_INT4_WEIGHT` | Enable INT4 weight quantization | `false` |
| `SGLANG_PER_TOKEN_GROUP_QUANT_8BIT_V2` | Apply per token group quantization kernel with fused silu and mul and masked m | `false` |
| `SGLANG_FORCE_FP8_MARLIN` | Force using FP8 MARLIN kernels even if other FP8 kernels are available | `false` |
| `SGLANG_FLASHINFER_FP4_GEMM_BACKEND` | Select backend for `mm_fp4` on Blackwell GPUS | `` |
| `SGLANG_NVFP4_CKPT_FP8_GEMM_IN_ATTN` | Quantize q_b_proj from BF16 to FP8 when launching DeepSeek NVFP4 checkpoint | `false` |
| `SGLANG_MOE_NVFP4_DISPATCH` | Use nvfp4 for moe dispatch (on flashinfer_cutlass or flashinfer_cutedsl moe runner backend) | `"false"` |
| `SGLANG_ENABLE_FLASHINFER_FP8_GEMM` (deprecated) | Use flashinfer kernels when running blockwise fp8 GEMM on Blackwell GPUs. **DEPRECATED**: Please use `--fp8-gemm-backend=flashinfer_trtllm` instead. | `false` |
| `SGLANG_SUPPORT_CUTLASS_BLOCK_FP8` (deprecated) | Use Cutlass kernels when running blockwise fp8 GEMM on Hopper or Blackwell GPUs. **DEPRECATED**: Please use `--fp8-gemm-backend=cutlass` instead. | `false` |
## Distributed Computing