Experts backends

All Mixture-of-Experts (MoE) implementations perform the same high-level computation. For each token, a router selects k experts. The token hidden state is then projected through the selected experts’ parameters and aggregated with routing weights. The difference between experts backends is how those expert matrix multiplications execute.

The ExpertsInterface provides optimized experts backends. It decouples the experts implementation from the model code to simplify experimentation with different functions. Add new backends through the same interface.

experts backend	description	GPU	CPU
`"eager"`	Reference implementation that loops over selected experts and applies projections on their tokens.	Reasonable baseline performance without requiring compilation.	Slower than `grouped_mm` but faster than `batched_mm`.
`"batched_mm"`	Duplicates selected expert parameters for each token and projects all tokens in a single batched GEMM using `torch.bmm`.	Fastest for small inputs, especially with compilation. Uses more memory due to parameter duplication.	Not recommended (significantly slower than other backends).
`"grouped_mm"`	Orders tokens by selected experts and uses `torch._grouped_mm` to project all tokens in a single grouped GEMM (requires PyTorch 2.9+).	Best for larger inputs and more memory efficient as it avoids duplicating expert parameters. Fast with compilation.	Most efficient backend for all input sizes.

When using experts_implementation="grouped_mm" on GPU, the model automatically switches to "batched_mm" during the decode stage of generation (after prefill). This is because batched_mm is significantly faster on lower token count during autoregressive decoding on GPU. On CPU, grouped_mm remains active throughout generation as it is more efficient for all input sizes.

Set an experts backend

Use the experts_implementation argument in from_pretrained() to instantiate a model with a specific experts backend.

from transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen1.5-MoE-A2.7B",
    dtype="bfloat16",
    experts_implementation="batched_mm",
)

Switch between experts backends at runtime without reloading the model using set_experts_implementation().

model.set_experts_implementation("eager")

Backbone-specific experts backend

Multimodal models can have multiple sub-configs (for example, different backbones). You can set a different experts backend per sub-config by passing a dict to experts_implementation at load time.

Keys in the mapping must match sub-config names.

from transformers import AutoModelForImageTextToText

experts_implementation_per_backbone = {
    "text_config": "grouped_mm",
    "vision_config": "eager",
}

model = AutoModelForImageTextToText.from_pretrained(
    "Qwen/Qwen3-VL-Moe",
    experts_implementation=experts_implementation_per_backbone,
)

Set the experts backend globally with an empty key.

model = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen1.5-MoE-A2.7B",
    experts_implementation={"": "batched_mm"},
)

torch.compile

All three backends ("eager", "batched_mm", "grouped_mm") are compatible with torch.compile to certain extents. The following table summarizes compatibility:

Implementation	compilation modes	dtypes	`fullgraph=True`
`grouped_mm`	`None`, `max-autotune-no-cudagraphs`	`bfloat16`	Yes
`batched_mm`	all	`bfloat16`, `float16`, `float32`	Yes
`eager`	all	`bfloat16`, `float16`, `float32`	No

Notes:

The grouped_mm experts backend currently only supports bfloat16 when compiled with torch.compile. Additionally, it is not compatible with CUDA graphs, so you must use mode=None or mode="max-autotune-no-cudagraphs" when compiling.
The eager experts backend uses a data-dependent operation to find which experts are used in a forward pass. This operation is not compatible with full graph compilation (fullgraph=True).

import torch
from transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen1.5-MoE-A2.7B",
    dtype="bfloat16",
    experts_implementation="grouped_mm",
).eval().cuda()

# Works for grouped_mm (no CUDA graphs)
model.forward = torch.compile(model.forward, mode="max-autotune-no-cudagraphs")

Benchmarks

This benchmark compares different input sizes and experts implementations with and without torch.compile.

Update on GitHub

Transformers

Experts backends

Set an experts backend

Backbone-specific experts backend

torch.compile

Benchmarks