trellis-2-mrp-mlx/mlx_backend
m3ultra fa972aa345
Some checks are pending
CodeQL Advanced / Analyze (${{ matrix.language }}) (none, c-cpp) (push) Waiting to run
CodeQL Advanced / Analyze (${{ matrix.language }}) (none, python) (push) Waiting to run
mlx_backend: pure-MLX FlowEuler+CFG(+interval) sampler loop
Replaces the torch CPU sampler + per-forward adapter bounces: one
conversion per stage boundary, batched dense CFG, once-per-stage
concat_cond, one eval/step. Env-gated (TRELLIS2_MLX_SAMPLER=0 reverts;
TRELLIS2_MLX_COMPILE=1 compiles the step forward). Parity: raw mesh
within 0.03% of torch-sampler baseline at seed 42. Sampling 56.8->54.4s
(m3ultra) — honest ~4%; the win is architectural (stable shapes for
compile, no host round-trips) not headline speed.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-20 03:01:57 +10:00
..
__init__.py add mlx backend for apple silicon with metal gpu acceleration 2026-07-17 07:51:05 +03:00
adapters.py add mlx backend for apple silicon with metal gpu acceleration 2026-07-17 07:51:05 +03:00
attention.py add mlx backend for apple silicon with metal gpu acceleration 2026-07-17 07:51:05 +03:00
convert.py add mlx backend for apple silicon with metal gpu acceleration 2026-07-17 07:51:05 +03:00
dinov3.py feat: add native Apple Silicon asset pipeline 2026-07-17 09:21:34 +03:00
flow_models.py inline env defaults in api_server, remove start script, drop intermediate mx.eval syncs 2026-07-17 07:51:05 +03:00
mlx_samplers.py mlx_backend: pure-MLX FlowEuler+CFG(+interval) sampler loop 2026-07-20 03:01:57 +10:00
norm.py add mlx backend for apple silicon with metal gpu acceleration 2026-07-17 07:51:05 +03:00
pipeline.py mlx_backend: pure-MLX FlowEuler+CFG(+interval) sampler loop 2026-07-20 03:01:57 +10:00
rope.py add mlx backend for apple silicon with metal gpu acceleration 2026-07-17 07:51:05 +03:00
sparse_conv.py add mlx backend for apple silicon with metal gpu acceleration 2026-07-17 07:51:05 +03:00
sparse_ops.py add mlx backend for apple silicon with metal gpu acceleration 2026-07-17 07:51:05 +03:00
sparse_tensor.py add mlx backend for apple silicon with metal gpu acceleration 2026-07-17 07:51:05 +03:00
structure_decoder.py add mlx backend for apple silicon with metal gpu acceleration 2026-07-17 07:51:05 +03:00
transformer_block.py add mlx backend for apple silicon with metal gpu acceleration 2026-07-17 07:51:05 +03:00
vae_blocks.py add mlx backend for apple silicon with metal gpu acceleration 2026-07-17 07:51:05 +03:00
vae_decoders.py add mlx backend for apple silicon with metal gpu acceleration 2026-07-17 07:51:05 +03:00