Support symmetric memory pre-allocation to avoid fragmentation (#17089)
This commit is contained in:
@@ -41,6 +41,7 @@ SGLang supports various environment variables that can be used to configure its
|
||||
| `SGLANG_MM_BUFFER_SIZE_MB` | Size of preallocated GPU buffer (in MB) for multi-modal feature hashing optimization. When set to a positive value, temporarily moves features to GPU for faster hash computation, then moves them back to CPU to save GPU memory. Larger features benefit more from GPU hashing. Set to `0` to disable. | `0` |
|
||||
| `SGLANG_MM_PRECOMPUTE_HASH` | Enable precomputing of hash values for MultimodalDataItem | `false` |
|
||||
| `SGLANG_NCCL_ALL_GATHER_IN_OVERLAP_SCHEDULER_SYNC_BATCH` | Enable NCCL for gathering when preparing mlp sync batch under overlap scheduler (without this flag gloo is used for gathering) | `false` |
|
||||
| `SGLANG_SYMM_MEM_PREALLOC_GB_SIZE` | Size of preallocated GPU buffer (in GB) for NCCL symmetric memory pool to limit memory fragmentation. Only have an effect when server arg `--enable-symm-mem` is set. | `4` |
|
||||
|
||||
|
||||
## DeepGEMM Configuration (Advanced Optimization)
|
||||
|
||||
Reference in New Issue
Block a user