GeWu-Lab commited on
Commit
e6ad93c
·
verified ·
1 Parent(s): 634aadc

Simplify model size and total inference memory requirements

Browse files
Files changed (2) hide show
  1. README.md +9 -11
  2. manifest.json +2 -2
README.md CHANGED
@@ -88,19 +88,17 @@ For pitch and volume, use `Increase` or `Reduce` with an explicit magnitude. Vol
88
 
89
  Input files are downmixed to mono, resampled to 16 kHz, and truncated or right-padded to 10 seconds. Output is a 10-second mono waveform at 16 kHz. The default runtime uses BF16, PyTorch SDPA and 50 Euler steps, with guidance scale 8.5 for Addition and 3.5 for the other tasks. Use `--seed` with the CLI or `seed=...` with the Python API for repeatable sampling.
90
 
91
- ## Estimated GPU memory
92
 
93
- For one 10-second sample at batch size 1 with all components on the GPU:
 
 
 
 
 
 
94
 
95
- | Component | Parameters | BF16 weight storage (GiB) |
96
- |---|---:|---:|
97
- | DiT editor | 673,349,696 | 1.254 |
98
- | Audio VAE | 509,336,866 | 0.949 |
99
- | T5 encoder | 109,628,544 | 0.204 |
100
- | HiFi-GAN vocoder | 55,264,897 | 0.103 |
101
- | Total | 1,347,580,003 | 2.510 |
102
-
103
- Estimated total inference memory is **4–8 GiB VRAM**; **8 GiB or more is recommended**. This is an estimate, not a CUDA measurement. The total includes weight storage, activations, workspaces and runtime overhead, and can vary with hardware and library versions. Inference uses no CPU offload. See [MEMORY.md](https://github.com/GeWu-Lab/SteerSound/blob/main/MEMORY.md) for details.
104
 
105
  ## Files and data
106
 
 
88
 
89
  Input files are downmixed to mono, resampled to 16 kHz, and truncated or right-padded to 10 seconds. Output is a 10-second mono waveform at 16 kHz. The default runtime uses BF16, PyTorch SDPA and 50 Euler steps, with guidance scale 8.5 for Addition and 3.5 for the other tasks. Use `--seed` with the CLI or `seed=...` with the Python API for repeatable sampling.
90
 
91
+ ## Model size and GPU memory
92
 
93
+ | Component | Parameters |
94
+ |---|---:|
95
+ | DiT editor | 673,349,696 |
96
+ | Audio VAE | 509,336,866 |
97
+ | T5 encoder | 109,628,544 |
98
+ | HiFi-GAN vocoder | 55,264,897 |
99
+ | Total | **1,347,580,003** |
100
 
101
+ **GPU memory for inference: approximately 8 GB** (BF16, batch size 1, 10-second audio).
 
 
 
 
 
 
 
 
102
 
103
  ## Files and data
104
 
manifest.json CHANGED
@@ -6,8 +6,8 @@
6
  "sha256": "1851bd54305514445591afe73dec20f7142d93892ae24d0fc62dc2855f4d0009"
7
  },
8
  "README.md": {
9
- "bytes": 5283,
10
- "sha256": "1a0488a3dfabd191e0d46d6a591f8c6302a61b10a0e1d59f5b47c4715758052f"
11
  },
12
  "dit/model.safetensors": {
13
  "bytes": 1346719328,
 
6
  "sha256": "1851bd54305514445591afe73dec20f7142d93892ae24d0fc62dc2855f4d0009"
7
  },
8
  "README.md": {
9
+ "bytes": 4848,
10
+ "sha256": "9331b1f1e0c132b07bbee95b3865924da68dcd38fa805617a4816a9d9635302d"
11
  },
12
  "dit/model.safetensors": {
13
  "bytes": 1346719328,