hukyhukygo commited on
Commit
e4133ef
·
verified ·
1 Parent(s): de8829c

Update official model card

Browse files
Files changed (1) hide show
  1. README.md +9 -11
README.md CHANGED
@@ -11,7 +11,8 @@ tags:
11
 
12
  # FireRedTTS3-Base for SGLang Omni
13
 
14
- We provide our FireRedTTS3-Base checkpoint for inference with
 
15
  [SGLang Omni](https://github.com/sgl-project/sglang-omni). It supports
16
  zero-shot voice cloning from one reference recording and its transcript.
17
  Audio output is 24 kHz mono, with WAV responses and streaming PCM available
@@ -39,20 +40,15 @@ source .venv/bin/activate
39
  uv pip install --prerelease=allow -e '.[firered-tts3]'
40
  ```
41
 
42
- Use a revision containing the FireRedTTS3 integration. Before that integration
43
- is merged, check out the corresponding PR revision before installing; an older
44
- PyPI release or checkout does not provide this model. Run the following commands
45
- from the `sglang-omni` repository root.
46
 
47
  ## Download this checkpoint
48
 
49
- For PR testing, use our temporary checkpoint at
50
- [CKermit/FireRedTTS3-Base](https://huggingface.co/CKermit/FireRedTTS3-Base).
51
- After this PR, we plan to publish the official checkpoint at
52
- [FireRedTeam/FireRedTTS3-Base](https://huggingface.co/FireRedTeam/FireRedTTS3-Base).
53
 
54
  ```bash
55
- hf download CKermit/FireRedTTS3-Base \
56
  --local-dir ./models/FireRedTTS3-Base
57
  ```
58
 
@@ -182,7 +178,7 @@ if __name__ == "__main__":
182
  ## Serving notes
183
 
184
  - The current deployment supports a single GPU (`tp_size=1`).
185
- - HTTP `input` is limited to 400 characters.
186
  - Use an explicit language: `Chinese` (default), `English`, `Japanese`, `Korean`,
187
  `German`, `French`, `Russian`, `Portuguese`, `Spanish`, or `Italian`.
188
  Automatic language detection (`Auto`) is unsupported. These are the stock
@@ -193,6 +189,8 @@ if __name__ == "__main__":
193
  - Put flow-solver settings under `stage_params.talker`, for example
194
  `{"n_timesteps": 10, "inference_cfg": 2.0, "max_patches": 200}`. Unknown
195
  top-level fields are not forwarded as model settings.
 
 
196
  - `seed` controls the per-request flow sampler. Different precisions, hardware,
197
  or batching schedules can produce different results with the same seed.
198
  - For our published model-quality results, see the
 
11
 
12
  # FireRedTTS3-Base for SGLang Omni
13
 
14
+ This repository contains the officially released HF-format FireRedTTS3-Base
15
+ checkpoint for inference with
16
  [SGLang Omni](https://github.com/sgl-project/sglang-omni). It supports
17
  zero-shot voice cloning from one reference recording and its transcript.
18
  Audio output is 24 kHz mono, with WAV responses and streaming PCM available
 
40
  uv pip install --prerelease=allow -e '.[firered-tts3]'
41
  ```
42
 
43
+ Run the following commands from the `sglang-omni` repository root.
 
 
 
44
 
45
  ## Download this checkpoint
46
 
47
+ Download the official HF-format checkpoint from
48
+ [FireRedTeam/FireRedTTS3-Base-HF](https://huggingface.co/FireRedTeam/FireRedTTS3-Base-HF).
 
 
49
 
50
  ```bash
51
+ hf download FireRedTeam/FireRedTTS3-Base-HF \
52
  --local-dir ./models/FireRedTTS3-Base
53
  ```
54
 
 
178
  ## Serving notes
179
 
180
  - The current deployment supports a single GPU (`tp_size=1`).
181
+ - The `input` text is limited to 400 characters.
182
  - Use an explicit language: `Chinese` (default), `English`, `Japanese`, `Korean`,
183
  `German`, `French`, `Russian`, `Portuguese`, `Spanish`, or `Italian`.
184
  Automatic language detection (`Auto`) is unsupported. These are the stock
 
189
  - Put flow-solver settings under `stage_params.talker`, for example
190
  `{"n_timesteps": 10, "inference_cfg": 2.0, "max_patches": 200}`. Unknown
191
  top-level fields are not forwarded as model settings.
192
+ - HTTP, batch, and WebSocket requests do not forward `extra_params`. The SDK
193
+ `GenerateRequest.extra_params` field remains available for SDK requests.
194
  - `seed` controls the per-request flow sampler. Different precisions, hardware,
195
  or batching schedules can produce different results with the same seed.
196
  - For our published model-quality results, see the