Spaces:
Running
on
T4
Running
on
T4
mintox
#1
by
mortimerp9
- opened
- Dockerfile +1 -1
- README.md +1 -1
- app.py +5 -7
- requirements.txt +1 -1
- whl/seamless_communication-1.0.0-py3-none-any.whl +2 -2
Dockerfile
CHANGED
@@ -45,7 +45,7 @@ RUN pyenv install ${PYTHON_VERSION} && \
|
|
45 |
|
46 |
COPY --chown=1000 . ${HOME}/app
|
47 |
RUN pip install -r ${HOME}/app/requirements.txt && \
|
48 |
-
pip install fairseq2 --pre --extra-index-url https://fair.pkg.atmeta.com/fairseq2/pt2.1.0/cu121 && \
|
49 |
pip install ${HOME}/app/whl/seamless_communication-1.0.0-py3-none-any.whl
|
50 |
|
51 |
ENV PYTHONPATH=${HOME}/app \
|
|
|
45 |
|
46 |
COPY --chown=1000 . ${HOME}/app
|
47 |
RUN pip install -r ${HOME}/app/requirements.txt && \
|
48 |
+
pip install fairseq2 --pre --extra-index-url https://fair.pkg.atmeta.com/fairseq2/whl/nightly/pt2.1.0/cu121 && \
|
49 |
pip install ${HOME}/app/whl/seamless_communication-1.0.0-py3-none-any.whl
|
50 |
|
51 |
ENV PYTHONPATH=${HOME}/app \
|
README.md
CHANGED
@@ -7,7 +7,7 @@ sdk: docker
|
|
7 |
pinned: false
|
8 |
suggested_hardware: t4-medium
|
9 |
models:
|
10 |
-
- facebook/seamless-m4t-
|
11 |
- facebook/SONAR
|
12 |
---
|
13 |
|
|
|
7 |
pinned: false
|
8 |
suggested_hardware: t4-medium
|
9 |
models:
|
10 |
+
- facebook/seamless-m4t-large
|
11 |
- facebook/SONAR
|
12 |
---
|
13 |
|
app.py
CHANGED
@@ -23,7 +23,7 @@ from lang_list import (
|
|
23 |
|
24 |
CHECKPOINTS_PATH = pathlib.Path(os.getenv("CHECKPOINTS_PATH", "/home/user/app/models"))
|
25 |
if not CHECKPOINTS_PATH.exists():
|
26 |
-
snapshot_download(repo_id="
|
27 |
asset_store.env_resolvers.clear()
|
28 |
asset_store.env_resolvers.append(lambda: "demo")
|
29 |
demo_metadata = [
|
@@ -45,8 +45,7 @@ DESCRIPTION = """\
|
|
45 |
[SeamlessM4T](https://github.com/facebookresearch/seamless_communication) is designed to provide high-quality
|
46 |
translation, allowing people from different linguistic communities to communicate effortlessly through speech and text.
|
47 |
This unified model enables multiple tasks like Speech-to-Speech (S2ST), Speech-to-Text (S2TT), Text-to-Speech (T2ST)
|
48 |
-
translation and more, without relying on multiple separate models.
|
49 |
-
[SeamlessM4T demo website](https://seamless.metademolab.com/m4t?utm_source=huggingface&utm_medium=web&utm_campaign=seamless&utm_content=m4tspace).
|
50 |
"""
|
51 |
|
52 |
CACHE_EXAMPLES = os.getenv("CACHE_EXAMPLES") == "1" and torch.cuda.is_available()
|
@@ -105,8 +104,8 @@ def run_s2tt(input_audio: str, source_language: str, target_language: str) -> st
|
|
105 |
out_texts, _ = translator.predict(
|
106 |
input=input_audio,
|
107 |
task_str="S2TT",
|
108 |
-
src_lang=source_language_code,
|
109 |
tgt_lang=target_language_code,
|
|
|
110 |
)
|
111 |
return str(out_texts[0])
|
112 |
|
@@ -117,8 +116,8 @@ def run_t2st(input_text: str, source_language: str, target_language: str) -> tup
|
|
117 |
out_texts, out_audios = translator.predict(
|
118 |
input=input_text,
|
119 |
task_str="T2ST",
|
120 |
-
src_lang=source_language_code,
|
121 |
tgt_lang=target_language_code,
|
|
|
122 |
)
|
123 |
out_text = str(out_texts[0])
|
124 |
out_wav = out_audios.audio_wavs[0].cpu().detach().numpy()
|
@@ -131,8 +130,8 @@ def run_t2tt(input_text: str, source_language: str, target_language: str) -> str
|
|
131 |
out_texts, _ = translator.predict(
|
132 |
input=input_text,
|
133 |
task_str="T2TT",
|
134 |
-
src_lang=source_language_code,
|
135 |
tgt_lang=target_language_code,
|
|
|
136 |
)
|
137 |
return str(out_texts[0])
|
138 |
|
@@ -143,7 +142,6 @@ def run_asr(input_audio: str, target_language: str) -> str:
|
|
143 |
out_texts, _ = translator.predict(
|
144 |
input=input_audio,
|
145 |
task_str="ASR",
|
146 |
-
src_lang=target_language_code,
|
147 |
tgt_lang=target_language_code,
|
148 |
)
|
149 |
return str(out_texts[0])
|
|
|
23 |
|
24 |
CHECKPOINTS_PATH = pathlib.Path(os.getenv("CHECKPOINTS_PATH", "/home/user/app/models"))
|
25 |
if not CHECKPOINTS_PATH.exists():
|
26 |
+
snapshot_download(repo_id="meta-private/M4Tv2", repo_type="model", local_dir=CHECKPOINTS_PATH)
|
27 |
asset_store.env_resolvers.clear()
|
28 |
asset_store.env_resolvers.append(lambda: "demo")
|
29 |
demo_metadata = [
|
|
|
45 |
[SeamlessM4T](https://github.com/facebookresearch/seamless_communication) is designed to provide high-quality
|
46 |
translation, allowing people from different linguistic communities to communicate effortlessly through speech and text.
|
47 |
This unified model enables multiple tasks like Speech-to-Speech (S2ST), Speech-to-Text (S2TT), Text-to-Speech (T2ST)
|
48 |
+
translation and more, without relying on multiple separate models.
|
|
|
49 |
"""
|
50 |
|
51 |
CACHE_EXAMPLES = os.getenv("CACHE_EXAMPLES") == "1" and torch.cuda.is_available()
|
|
|
104 |
out_texts, _ = translator.predict(
|
105 |
input=input_audio,
|
106 |
task_str="S2TT",
|
|
|
107 |
tgt_lang=target_language_code,
|
108 |
+
src_lang=source_language_code,
|
109 |
)
|
110 |
return str(out_texts[0])
|
111 |
|
|
|
116 |
out_texts, out_audios = translator.predict(
|
117 |
input=input_text,
|
118 |
task_str="T2ST",
|
|
|
119 |
tgt_lang=target_language_code,
|
120 |
+
src_lang=source_language_code,
|
121 |
)
|
122 |
out_text = str(out_texts[0])
|
123 |
out_wav = out_audios.audio_wavs[0].cpu().detach().numpy()
|
|
|
130 |
out_texts, _ = translator.predict(
|
131 |
input=input_text,
|
132 |
task_str="T2TT",
|
|
|
133 |
tgt_lang=target_language_code,
|
134 |
+
src_lang=source_language_code,
|
135 |
)
|
136 |
return str(out_texts[0])
|
137 |
|
|
|
142 |
out_texts, _ = translator.predict(
|
143 |
input=input_audio,
|
144 |
task_str="ASR",
|
|
|
145 |
tgt_lang=target_language_code,
|
146 |
)
|
147 |
return str(out_texts[0])
|
requirements.txt
CHANGED
@@ -1,4 +1,4 @@
|
|
1 |
-
gradio==4.
|
2 |
omegaconf==2.3.0
|
3 |
torch==2.1.0
|
4 |
torchaudio==2.1.0
|
|
|
1 |
+
gradio==4.5.0
|
2 |
omegaconf==2.3.0
|
3 |
torch==2.1.0
|
4 |
torchaudio==2.1.0
|
whl/seamless_communication-1.0.0-py3-none-any.whl
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:c3380dc7d6613c4dc9ef4d78fd3dcf1a50f7c6a659b8ba0b37ad5237533d002e
|
3 |
+
size 234811
|