Hugging Face's logo

stabilityai
/
stable-audio-3-optimized

Text-to-Audio
Stable Audio 3
LiteRT
ONNX
English
audio-generation
music
sound-effects
diffusion
Model card Files
xet
 Community
8
stable-audio-3-optimized

Commit History

  1. cpu-amx: add dit_medium_bf16_core.bin (bf16 DiT engine)

    cortexelus
    Verified signature
    0905190

  2. cpu-amx: add dit_medium_bf16_flash.so (bf16 DiT engine)

    cortexelus
    Verified signature
    60952d8

  3. cpu-amx: add dit_medium_bf16_pin_fp32.npz (bf16 DiT engine)

    cortexelus
    Verified signature
    476830e

  4. cpu-amx: add dit_medium_bf16_core_manifest.txt (bf16 DiT engine)

    cortexelus
    Verified signature
    70aafed

  5. cpu-amx: add dit_medium_bf16.so (bf16 DiT engine)

    cortexelus
    Verified signature
    fc19859

  6. Delete old fp16mixed names + retired medium dit_bf16.trt (fp16 rename shipped)

    cortexelus
    Verified signature
    48e34e2

  7. Rename fp16mixed->fp16 DiT/T5Gemma engines+onnx (copies; deletes follow after code merge)

    cortexelus
    Verified signature
    3f29675

  8. Remove redundant w8_bf16 engines (int8 weights fold to bf16 at build → identical size/speed to the bf16 baseline)

    cortexelus
    Verified signature
    63cc70d

  9. Add sm_120 TRT engines (AOT SWA, wide profile) for the quantized tiers

    cortexelus
    Verified signature
    4e8cc21

  10. cpu-amx: add int8 encoders (same_{s,l}_encoder_int8, SmoothQuant+GPTQ) + README row

    cortexelus
    Verified signature
    be8ab06

  11. cpu-amx README: consistent <model>_<precision> filenames

    cortexelus
    Verified signature
    0d42b85

  12. cpu-amx: consistent <model>_<precision> filenames (t5gemma_bf16, dit_medium_int8, same_*_encoder_bf16)

    cortexelus
    Verified signature
    5d1992c

  13. Add sm_90 TRT engines (wide profile) for the quantized tiers

    cortexelus
    Verified signature
    d593f74

  14. Add quantized encoder+decoder tier ONNX (fp8 / w8_bf16, padded encoders)

    cortexelus
    Verified signature
    4168236

  15. Add cpu-amx: torch-free C++/AMX engines (T5Gemma bf16, medium DiT int8, SAME-S/L bf16+int8 decoders, SAME-S/L encoders) + weight blobs + README

    cortexelus
    Verified signature
    dec2054

  1. Add fp8 tier for sm-music / sm-sfx DiTs (clean fp8-linear graft on fp16mixed)

    cortexelus
    Verified signature
    308abaf

  2. Add fp8 tier for sm-music / sm-sfx DiTs (clean fp8-linear graft on fp16mixed)

    cortexelus
    Verified signature
    ac3a57e

  3. Add fp8 tier for sm-music / sm-sfx DiTs (clean fp8-linear graft on fp16mixed)

    cortexelus
    Verified signature
    c3aa6e0

  4. Add fp8 tier for sm-music / sm-sfx DiTs (clean fp8-linear graft on fp16mixed)

    cortexelus
    Verified signature
    5bcf0c7

  5. Add fp8 tier for sm-music / sm-sfx DiTs (clean fp8-linear graft on fp16mixed)

    cortexelus
    Verified signature
    6a96f1b

  6. Add fp8 tier for sm-music / sm-sfx DiTs (clean fp8-linear graft on fp16mixed)

    cortexelus
    Verified signature
    1489a17

  7. sm_120 sa3-m: add the fp8 medium-DiT tier (176 fp8 GEMM + 96 bf16 fused MHA, 1.43-1.95x vs fp16mixed) (#7)⁠

    cortexelus
    f8e0a0f

  8. sa3-m fp8: calibrated dit_fp8.onnx (real-conditioning activation + per-channel weight scales, @ryanontheinside #47; refs unchanged dit_fp8lin.onnx.data)

    cortexelus
    Verified signature
    8383b7d

  9. sa3-m fp8: ship calibrated-bakedmin engine (176 fp8 GEMM + 96 bf16 fused MHA + baked fp32 RoPE; calibration by @ryanontheinside #47)

    cortexelus
    Verified signature
    4697212

  10. sm_120 SAME-L: AOT (graph-capturable) engines + tensorRT/README.md (#6)⁠

    cortexelus
    a05f2d4

  1. Upload onnx/sa3-m/dit_fp8lin.onnx.data with huggingface_hub

    cortexelus
    Verified signature
    9efa545

  2. Upload onnx/sa3-m/dit_fp8.onnx with huggingface_hub

    cortexelus
    Verified signature
    21b6506

  3. Upload tensorRT/sm_90/sa3-m/dit_fp8.trt with huggingface_hub

    cortexelus
    Verified signature
    feddb8b

  4. Fix the bf16 medium DiT: bake RoPE's tables into the graph (clipping 3.112% -> 0.014%) (#5)⁠

    cortexelus
    6152b1d

  5. Rebuild medium DiT fp16-mixed engines (sm_90 + new sm_120): attention now fuses, 4.3x faster (#4)⁠

    cortexelus
    06debc3

  6. Fix medium DiT FP16-mixed ONNX: bound the RoPE island so attention fuses (4.3x faster, cos 1.0000) (#3)⁠

    cortexelus
    fb17708

  1. Add MLX base-model DiT weights for LoRA training (#2)⁠

    mattricesound +1: cortexelus
    08c64b9

  2. Add bf16 medium DiT TRT engine (sm_90) — FMHA-fused, medium default (#1)⁠

    mattricesound +1: cortexelus
    9a38efb

  1. tflite: replace naive same-l/enc_w8a8-dyn.tflite with GPTQ (torch-free, real-audio calibrated)

    cortexelus
    Verified signature
    c2949a4

  2. tflite: replace naive same-l/enc_w8a32.tflite with GPTQ (torch-free, real-audio calibrated)

    cortexelus
    Verified signature
    aea1c33

  3. tflite: replace naive same-s/enc_w8a8-dyn.tflite with GPTQ (torch-free, real-audio calibrated)

    cortexelus
    Verified signature
    6401d6c

  4. tflite: replace naive same-s/enc_w8a32.tflite with GPTQ (torch-free, real-audio calibrated)

    cortexelus
    Verified signature
    07885df

  1. tflite: add same-l/enc_w8a8-dyn.tflite (quantized encoder, varlen)

    cortexelus
    Verified signature
    a7970b2

  2. tflite: add same-l/enc_w8a32.tflite (quantized encoder, varlen)

    cortexelus
    Verified signature
    7de7f11

  3. tflite: add same-l/enc_w16a32.tflite (quantized encoder, varlen)

    cortexelus
    Verified signature
    4adfa45

  4. tflite: add same-s/enc_w8a8-dyn.tflite (quantized encoder, varlen)

    cortexelus
    Verified signature
    151e748

  5. tflite: add same-s/enc_w8a32.tflite (quantized encoder, varlen)

    cortexelus
    Verified signature
    2682028

  6. tflite: add same-s/enc_w16a32.tflite (quantized encoder, varlen)

    cortexelus
    Verified signature
    d35a0c9

  7. tflite: add sa3-m/dit_w16a32.tflite (quantized, varlen+varbatch)

    cortexelus
    Verified signature
    4ed3663

  8. tflite: add sa3-m/dit_w8a32.tflite (quantized, varlen+varbatch)

    cortexelus
    Verified signature
    3e6eee6

  9. tflite: add sa3-m/dit_w8a8-dyn.tflite (quantized, varlen+varbatch)

    cortexelus
    Verified signature
    78049a1

  10. tflite: add same-l/dec_w16a32.tflite (quantized, varlen+varbatch)

    cortexelus
    Verified signature
    f4cd71b

  11. tflite: add sa3-sm-sfx/dit_w16a32.tflite (quantized, varlen+varbatch)

    cortexelus
    Verified signature
    860ab42

  12. tflite: add sa3-sm-music/dit_w16a32.tflite (quantized, varlen+varbatch)

    cortexelus
    Verified signature
    e6e2e82

  13. tflite: add same-l/dec_w8a32.tflite (quantized, varlen+varbatch)

    cortexelus
    Verified signature
    cd3b3c2