For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).
Python function
check_supported
check_supported()
max.nn.ep.check_supported(config, mesh)
Raises if this exchange cannot run config on mesh.
The exchange dispatches BF16 tokens, or quantizes them in transit to
FP8 (block-scaled, compressed-tensors or FBGEMM), MXFP4, MXFP6, MXFP8 or
NVFP4. NVFP4 with per-token global scales dispatches in one launch only.
With config.eplb_count_expert_load, each dispatch also counts the
tokens routed to each local expert.
-
Parameters:
-
- config (EPConfig) – The expert-parallel configuration.
- mesh (DeviceMesh) – The mesh of the devices that the exchange runs on.
-
Raises:
-
ValueError – If
configdispatches in another format, or in a format that this accelerator’s kernels do not serve, or ifmeshdoes not hold one device per GPU of the node. -
Return type:
-
None