vllm.model_executor.layers.fused_moe.oracle.nvfp4
¶
Classes:
-
NvFp4MoEKernelOracle–NVFP4 MoE backend selection for one deployment configuration.
Functions:
-
map_nvfp4_backend–Map user's MoEBackend to NvFp4MoeBackend.
-
nvfp4_backend_supports_csf_scales–Whether the backend keeps
csfcheckpoint scales compressed. -
select_nvfp4_moe_backend–Select the primary NvFP4 MoE backend
NvFp4MoEKernelOracle
¶
Bases: MoEKernelOracle[NvFp4MoeBackend]
NVFP4 MoE backend selection for one deployment configuration.
Source code in vllm/model_executor/layers/fused_moe/oracle/nvfp4.py
map_nvfp4_backend(runner_backend)
¶
Map user's MoEBackend to NvFp4MoeBackend.
Source code in vllm/model_executor/layers/fused_moe/oracle/nvfp4.py
nvfp4_backend_supports_csf_scales(backend, config, use_a16)
¶
Whether the backend keeps csf checkpoint scales compressed.
Source code in vllm/model_executor/layers/fused_moe/oracle/nvfp4.py
select_nvfp4_moe_backend(config, weight_key, activation_key)
¶
Select the primary NvFP4 MoE backend
Note: Shape-specific fallbacks may still occur at runtime.
Returns NONE and a placeholder experts class for requantization sources.
Source code in vllm/model_executor/layers/fused_moe/oracle/nvfp4.py
234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 | |