docs: LoRA branch engines for sa3-sm-music and sa3-sm-sfx (193 targets each)

docs: LoRA branch engines for sa3-sm-music and sa3-sm-sfx (193 targets each)

drop the medium DiT .xnnwc weight caches (regenerated locally)

tensorRT: add sa3-sm-sfx LoRA branch engine (sm_90, 193 targets)

tensorRT: add sa3-sm-music LoRA branch engine (sm_90, 193 targets)

small DiTs (sm-music, sm-sfx) tflite: ship the w8a8 rung ladders

tensorRT/sm_90: add the sa3-m runtime LoRA engine (229 targets, rank<=512)

tensorRT: document the runtime LoRA / DoRA engine

tensorRT: document the runtime LoRA / DoRA engine

tflite README: medium DiT default -> w8a8 + SAME-S

medium DiT tflite: ship the N=12 rung ladder (fp32 + cache-safe w8a8)

Update README.md

tflite: bake output limiter into SAME-S/SAME-L decoders (ceiling 0.977, matches TRT graft)

tflite: bake output limiter into SAME-S/SAME-L decoders (ceiling 0.977, matches TRT graft)

model card: note optimized TFLite/LiteRT rung codecs

tflite: SAME codecs -> static rung models (fp32 + w8a8); retire w16a32/w8a32/w8a8-dyn to legacy/

tflite: preserve pre-rung SAME codecs under legacy/

Move the retired fp8_fast ONNX under onnx/same-s/legacy/

Rename the autoencoder ONNX to match the engine naming scheme

Move the retired w8_bf16 ONNX under onnx/*/legacy/

Delete the miscalibrated sm_120 SAME-S fp8 encoder

Move deprecated sm_120 TRT engines under legacy/

Move deprecated sm_90 TRT engines under legacy/

build-ready ONNX: limiter grafted + ceiling baked (decoders), amax-recalibrated activation scales (fp8 encoders)

build-ready ONNX: limiter grafted + ceiling baked (decoders), amax-recalibrated activation scales (fp8 encoders)

build-ready ONNX: limiter grafted + ceiling baked (decoders), amax-recalibrated activation scales (fp8 encoders)

build-ready ONNX: limiter grafted + ceiling baked (decoders), amax-recalibrated activation scales (fp8 encoders)

build-ready ONNX: limiter grafted + ceiling baked (decoders), amax-recalibrated activation scales (fp8 encoders)

build-ready ONNX: limiter grafted + ceiling baked (decoders), amax-recalibrated activation scales (fp8 encoders)

retire fp8_fast; remove the superseded sm_90 fp8 engines (miscalibrated encoder, 8192-ceiling decoders)

retire fp8_fast; remove the superseded sm_90 fp8 engines (miscalibrated encoder, 8192-ceiling decoders)

retire fp8_fast; remove the superseded sm_90 fp8 engines (miscalibrated encoder, 8192-ceiling decoders)

retire fp8_fast; remove the superseded sm_90 fp8 engines (miscalibrated encoder, 8192-ceiling decoders)

retire fp8_fast; remove the superseded sm_90 fp8 engines (miscalibrated encoder, 8192-ceiling decoders)

retire fp8_fast; remove the superseded sm_90 fp8 engines (miscalibrated encoder, 8192-ceiling decoders)

retire fp8_fast; remove the superseded sm_90 fp8 engines (miscalibrated encoder, 8192-ceiling decoders)

SAME-L/SAME-S: bake the limiter ceiling so the chunkable decoders keep the ('latent','pcm') signature; add the SAME-S fp8 chunkable pair

SAME-L/SAME-S: bake the limiter ceiling so the chunkable decoders keep the ('latent','pcm') signature; add the SAME-S fp8 chunkable pair

SAME-L/SAME-S: bake the limiter ceiling so the chunkable decoders keep the ('latent','pcm') signature; add the SAME-S fp8 chunkable pair

SAME-L/SAME-S: bake the limiter ceiling so the chunkable decoders keep the ('latent','pcm') signature; add the SAME-S fp8 chunkable pair

SAME-L/SAME-S: bake the limiter ceiling so the chunkable decoders keep the ('latent','pcm') signature; add the SAME-S fp8 chunkable pair

SAME-S: add the chunkable bf16 pair (two profiles, limiter)

SAME-S: add the chunkable bf16 pair (two profiles, limiter)

SAME-L encoders: low band 64 -> 256 (1.3-1.5x faster, cos 0.99831 -> 0.99969)

SAME-L encoders: low band 64 -> 256 (1.3-1.5x faster, cos 0.99831 -> 0.99969)

SAME-L: remove miscalibrated enc_fp8.trt (activation scales 1.3-28x too small; superseded by enc_fp8_chunkable.trt on sm_90)

SAME-L: remove miscalibrated enc_fp8.trt (activation scales 1.3-28x too small; superseded by enc_fp8_chunkable.trt on sm_90)

SAME-L: add enc_fp8_chunkable.trt (fp16, two-profile chunkable)

SAME-L: add dec_fp8_chunkable_limiter.trt (fp16, two-profile chunkable, runtime limiter ceiling)

SAME-L: add enc_fp16_chunkable.trt (fp16, two-profile chunkable)