quant: publish the quantized encodings

.gitattributes

@@ -38,3 +38,9 @@ stable-audio-3-small-sfx-dit-0.5B-v1.0-F16.gguf filter=lfs diff=lfs merge=lfs -t
 stable-audio-3-small-sfx-dit-0.5B-v1.0-F32.gguf filter=lfs diff=lfs merge=lfs -text
 stable-audio-3-small-sfx-same-s-v1.0-F16.gguf filter=lfs diff=lfs merge=lfs -text
 stable-audio-3-small-sfx-same-s-v1.0-F32.gguf filter=lfs diff=lfs merge=lfs -text
+stable-audio-3-small-sfx-dit-0.5B-v1.0-Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+stable-audio-3-small-sfx-same-s-v1.0-Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+stable-audio-3-small-sfx-dit-0.5B-v1.0-Q5_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+stable-audio-3-small-sfx-same-s-v1.0-Q5_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+stable-audio-3-small-sfx-dit-0.5B-v1.0-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text
+stable-audio-3-small-sfx-same-s-v1.0-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text

README.md

@@ -6,6 +6,7 @@ license_name: stable-audio-community
 license_link: LICENSE.md
 pipeline_tag: text-to-audio
 base_model: stabilityai/stable-audio-3-small-sfx
+base_model_relation: quantized
 tags:
 - audio-generation
 - sound-effects
@@ -30,13 +31,34 @@ the shared **encoder + tokenizer** from
 
 | component | file | notes |
 |---|---|---|
-| DiT (diffusion transformer) | `stable-audio-3-small-sfx-dit-0.5B-v1.0-{F16,F32}.gguf` | pick one precision |
-| autoencoder (SAME-S) | `stable-audio-3-small-sfx-same-s-v1.0-{F16,F32}.gguf` | pick one precision |
+| DiT (diffusion transformer) | `stable-audio-3-small-sfx-dit-0.5B-v1.0-{F32,F16,Q8_0,Q5_K_M,Q4_K_M}.gguf` | pick one encoding |
+| autoencoder (SAME-S) | `stable-audio-3-small-sfx-same-s-v1.0-{F32,F16,Q8_0,Q5_K_M,Q4_K_M}.gguf` | **must match the DiT** |
 | conditioner | `stable-audio-3-small-sfx-conditioner-v1.0-F32.gguf` | tiny sidecar (prompt padding + seconds_total) |
 | encoder + tokenizer | → [t5gemma-b-b-ul2-GGUF](https://huggingface.co/thepatch/t5gemma-b-b-ul2-GGUF) | **shared** across all SA3 variants |
 
 > **note:** SAME-S needs an **even** `--frames` count (the packed sequence must divide the chunk size).
 
+## Encodings
+
+`sa3-generate --encoding` resolves the DiT and the SAME with the same suffix, so download the pair.
+
+| encoding | DiT | SAME-S | total |
+|---|---:|---:|---:|
+| F32 | 1751 MB | 413 MB | 2164 MB |
+| F16 | 877 MB | 207 MB | 1084 MB |
+| Q8_0 | 471 MB | 110 MB | 581 MB |
+| Q5_K_M | 333 MB | 79 MB | 412 MB |
+| Q4_K_M | 302 MB | 72 MB | 374 MB |
+
+`q4_k_m` and `q5_k_m` promote the attention V, feed-forward down and embedding tensors to Q6_K;
+`q8_0` is uniform. Every tier passes `sa3-quant-check` with `below-threshold=0` at cosine `0.990`
+against the F16 reference, for the DiT and the SAME alike.
+
+**Quantization buys footprint everywhere and speed only on some backends.** CUDA and Vulkan gain
+roughly 33% end to end. **Metal is flat** — Q8_0 is 1.7% faster and Q4_K_M 2.1% *slower* than F16,
+because the load-time saving and the added per-step dequant cancel out. On a Mac, pick a quant for
+the memory, not for the speed.
+
 ## Usage
 
 For use with [**sa3.cpp**](https://github.com/betweentwomidnights/sa3.cpp):
@@ -48,6 +70,13 @@ python tools/download_models.py --variant small-sfx --encoding f16
 sa3-generate --model small-sfx --prompt "a dog barking in a large empty hall" --out sfx.wav
 ```
 
+For a quantized set, pass the encoding to both — the downloader and the generator use the same names:
+
+```bash
+python tools/download_models.py --variant small-sfx --encoding q4_k_m
+sa3-generate --model small-sfx --encoding q4_k_m --prompt "a dog barking in a large hall" --out sfx.wav
+```
+
 ## Performance
 
 Roughly **1.7s for a 12s clip** at f16 on an 8GB laptop GPU (RTX 5070) — about 2× faster than the medium

stable-audio-3-small-sfx-dit-0.5B-v1.0-Q4_K_M.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09eaa89a98d2b483dafb1b0ec26dadccd7a0a38a2c0839582915f910c6040a13
+size 316195520

stable-audio-3-small-sfx-dit-0.5B-v1.0-Q5_K_M.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9dd02e9492db6138e794007562de172880b8b946fee9545965147f265b44b7d
+size 349274816

stable-audio-3-small-sfx-dit-0.5B-v1.0-Q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cceae541b7241d0fd67b1d0bae036e03d0cd8cbdd7a485db7d5505ddc118943a
+size 494224064

stable-audio-3-small-sfx-same-s-v1.0-Q4_K_M.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1f75f7726572ec4ab74cf9affe600b7dbea3758801d390b4a6bec985172a32b
+size 75876640

stable-audio-3-small-sfx-same-s-v1.0-Q5_K_M.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21c6c4bccf09b81da3719c727e2558f231cb9a6654a76ce6c3f0abf9a07bf8a2
+size 82315552

stable-audio-3-small-sfx-same-s-v1.0-Q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1b4e26213df03032eabd72668cd33d06deb5d93cfaa19d1d6d3fba3bcb7adc8
+size 115345696