SA3-m DiT FP8 ONNX + sm_120 TRT engine (PR47 recipe + dequant-frontend)

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+onnx/sa3-m/dit_fp8.onnx.data filter=lfs diff=lfs merge=lfs -text
+tensorRT/sm_120/sa3-m/dit_fp8.trt filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,43 @@
+---
+license: other
+license_name: stabilityai-community
+license_link: https://huggingface.co/stabilityai/stable-audio-3-medium/blob/main/LICENSE.md
+base_model: stabilityai/stable-audio-3-medium
+tags:
+- stable-audio
+- tensorrt
+- fp8
+- quantization
+---
+
+# SA3-medium DiT — FP8 TensorRT artifacts
+
+FP8 GEMM-trunk quantization of the Stable Audio 3 medium DiT, built from
+`stabilityai/stable-audio-3-optimized` `onnx/sa3-m/dit_fp16mixed.onnx` with the
+producer recipe in [Stability-AI/stable-audio-3 PR #47](https://github.com/Stability-AI/stable-audio-3/pull/47)
+(`build/make_calib.py` + `build/build_dit_fp8.py`), including the
+dequant-front-end recipe step. This is a derivative of Stability AI's model
+weights and is distributed under the Stability AI Community License; see the
+base model for terms.
+
+## Contents
+
+- `onnx/sa3-m/dit_fp8.onnx` + `dit_fp8.onnx.data` — the quantized ONNX
+  (arch-independent; compile with `build_from_onnx.py sa3-m-fp8`, plain
+  STRONGLY_TYPED, no ModelOpt needed)
+- `tensorRT/sm_120/sa3-m/dit_fp8.trt` — prebuilt engine for RTX 50xx
+  (sm_120), TensorRT 10.16.1.11. TRT engines are not portable across GPU
+  architectures or TRT minor versions; rebuild from the ONNX for anything else.
+
+## Validation (vs the FP16-mixed engine, 47 prompts x 8 sigmas, L=646, RTX 5090)
+
+- worst single-step latent cosine (x + dt*v): 0.9993 (n=376)
+- 8-step compounded final-latent cosine over 47 prompts: mean 0.967,
+  median 0.972, worst 0.901 (the rollout is chaotic; a 1e-3 input
+  perturbation alone compounds to ~0.967, so this is a guide, not a gate)
+- step latency B=1 L=646: ~10.6-11.0 ms vs ~18.7-19.4 ms FP16-mixed (~1.8x)
+- under the stochastic pingpong sampler the engine produces a different but
+  comparable sample
+
+Inputs/outputs are FP32, drop-in for the FP16-mixed DiT engine
+(`sa3_trt --precision fp8`, paired with the FP16-mixed decoder).

onnx/sa3-m/dit_fp8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af67c494125d1081a3d2d769ceec454d32be3237b46d9de8843a69fd45571758
+size 12382984

onnx/sa3-m/dit_fp8.onnx.data

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f425d71772126f9a5bec68f50fed4f822e117e8bf8cc48a1645e343988b156be
+size 2905153535

tensorRT/sm_120/sa3-m/dit_fp8.trt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ed8394389ccb9389805d6833a537cd0d249e796ad9e9d777ccfcb68d1addfdc
+size 1498263724