Add a transform to relace torch.ops.auto_deploy.torch_quant_nvfp4_moe
with the optimized torch.ops.auto_deploy.trtllm_quant_nvfp4_moe_fused.
Currently generates the wrong results when the number of rows in MoE FC1 weights is not divisible by 128,
so torch.ops.auto_deploy.trtllm_quant_nvfp4_moe_fused is not set as the default FP4 MoE implementation (i.e. the transform is disabled).
Signed-off-by: Neta Zmora <96238833+nzmora-nvidia@users.noreply.github.com>
Support two model flow with no overlap scheduler or chain drafter. Drafting model is in PyTorch backend.
Signed-off-by: Govind Ramnarayan <105831528+govind-ramnarayan@users.noreply.github.com>
The trtllm (cutlass) fp8 moe operator performs W3+W1 fusion (concat) during inference and we want to move this fusion to the model optimization time.
The Cutlass MoE kernel is used thru a trtllm torch operator.
Its implementation uses two FC operations (fc1 and fc2) while the canonical MoE API defines three GEMM operations and their associated weights (W1, W2, W3) so when we switch from the torch.moe op to the trtllm.moe op we also change terminology from w1, w2, w3 to fc1, fc2.
Signed-off-by: Neta Zmora <96238833+nzmora-nvidia@users.noreply.github.com>
[#9230][refactor] Replace nemotron patches with custom model implementation
* Why?
Patching for nemotron H models was growing out of hand, and made certain
optimizations more complex than they needed to be.
* What?
This commit finally gets rid of them, and replaces them with the custom
model implementation in `modeling_nemotron_h.py`.
Closes#9230
Closes NvBug 5747867
Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
* Why?
We would like to show an alternative to monkey-patching in AutoDeploy.
* What?
This commit builds on the existing custom model implementation for
NemotronH and adds the bits relevant for MoE layers.
Part of #9150.
Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
* Why?
There were a couple of issues with the recently merged custom model
injection for AutoDeploy + the reference implementation of nemotron
H:
- `d_mlp` was left in despite being mathematically always null (could
lead to runtime issues during sharding).
- the custom model mapping was inherited by children factories.
* What?
This commit fixes these issues, and refactors the key of the custom
implementation to be based on the name of the configuration class as
well.
Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
* Why?
The reference nemotron H code on HuggingFace is out of date,
and therefore bugged, and has several untested code paths.
This makes an already hairy patching system even hairier.
The proposal is to do away with those patches, and replace the
original implementation with one that is heavily slimmed down.
* What?
This PR sets the basis for an alternative path with such a
slimmed down implementation that:
- fixes bugs in the current HF implementation
- adds no new dependencies to TensorRT-LLM
- does away with unnecessary features for TensorRT-LLM/
AutoDeploy:
- no training related code (dropout, gradient checkpointing, etc.)
- no caching logic (we want to replace it with our own anyway)
- no attention masking where possible
- reuses existing AD custom ops for mamba SSM update /
causal conv1d / attention
In order for the above to be usable in the AD apparatus,
`AutoModelForCausalLMFactory` is extended to allow registrations
of custom model implementations.
Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
