michael/Athena-Deck
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add isolated Qwen3 TTS runtime, GUI setup and speech endpoint

Browse Source
This commit is contained in:
Mikei386
2026-09-28 23:51:56 +02:00
parent 52b1116bdb
commit 88e0edcc59
22 changed files with 413 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files
endpoint.py
+22 -5
View File
@@ -17,7 +17,7 @@ class APIError(ValueError):
class Endpoint:
def __init__(self,root,profiles,worker,scheduler,images,credentials,management_port):
self.root=Path(root);self.profiles=profiles;self.worker=worker;self.scheduler=scheduler;self.images=images;self.credentials=credentials
self.root=Path(root);self.profiles=profiles;self.worker=worker;self.scheduler=scheduler;self.images=images;self.tts=None;self.credentials=credentials
self.management_port=management_port;self.lock=threading.RLock();self.http=None;self.thread=None;self.state='stopped';self.error=None;self.inflight=0
self.allowed_ports=[int(p) for p in os.environ.get('DECK_API_PORTS','').split(',') if p]
self.config=dict(port=self.allowed_ports[0] if self.allowed_ports else 8120,enabled_profiles=[],autostart=False)
@@ -36,9 +36,9 @@ class Endpoint:
return [dict(p,enabled=p['id'] in enabled) for p in rows]
def status(self):
rows=self.rows();worker=self.worker.status();job=self.images.status()['job'];counts={}
for key,kind in [('llm','chat'),('image','image'),('tts','tts'),('stt','stt')]:
for key,kind in [('llm','chat'),('image','image'),('tts','audio'),('stt','stt')]:
subset=[p for p in rows if p['kind']==kind]
counts[key]=dict(enabled=sum(p['enabled'] for p in subset),available=sum(p['enabled'] and p['runnable'] for p in subset),loaded=bool(worker['state']=='ready' and kind=='chat') if kind=='chat' else bool(kind=='image' and job and job['state']=='running'),supported=kind in ('chat','image'))
counts[key]=dict(enabled=sum(p['enabled'] for p in subset),available=sum(p['enabled'] and p['runnable'] for p in subset),loaded=bool(worker['state']=='ready' and kind=='chat') if kind=='chat' else bool(self.tts and self.tts.status()['job'] and self.tts.status()['job']['state']=='running') if kind=='audio' else bool(kind=='image' and job and job['state']=='running'),supported=kind in ('chat','image') or (kind=='audio' and self.tts is not None))
scheduler_status=self.scheduler.status()
with self.lock:return dict(state=self.state,reachable=bool(self.thread and self.thread.is_alive() and self.state=='running'),port=self.config['port'],bind=os.environ.get('DECK_API_BIND','127.0.0.1'),base_url=f"http://127.0.0.1:{self.config['port']}/v1",allowed_ports=self.allowed_ports,error=self.error,counts=counts,worker=worker,scheduler=scheduler_status,profiles=[dict(id=p['id'],name=p['name'],kind=p['kind'],enabled=p['enabled'],runnable=p['runnable'],blockers=p['blockers']) for p in rows],active_requests=self.inflight)
def configure(self,data):
@@ -99,6 +99,7 @@ class Endpoint:
# Process shutdown does not change the user's autostart preference.
with self.lock:self.state='stopping';http=self.http
if http:http.shutdown();http.server_close()
if self.tts:self.tts.stop()
self.images.stop();self.worker.stop()
def allowed(self):
with self.lock:return self.state=='running'
@@ -151,8 +152,8 @@ class APIHandler(BaseHTTPRequestHandler):
if self.command=='GET' and self.path=='/v1/models':return self.send(ep.model_list())
if self.command=='GET' and self.path=='/health':return self.send({'status':'ok','service':'athena-deck-api'})
if self.command!='POST':raise APIError('Route nicht gefunden.',404)
if self.path in ('/v1/audio/speech','/v1/audio/transcriptions'):raise APIError('Für TTS/STT ist noch keine Deck-Laufzeit eingerichtet.',501,'not_implemented')
if self.path not in ('/v1/chat/completions','/v1/images/generations'):raise APIError('Route nicht implementiert.',404)
if self.path=='/v1/audio/transcriptions':raise APIError('Für TTS/STT ist noch keine Deck-Laufzeit eingerichtet.',501,'not_implemented')
if self.path not in ('/v1/chat/completions','/v1/images/generations','/v1/audio/speech'):raise APIError('Route nicht implementiert.',404)
if self.headers.get('Transfer-Encoding'):raise APIError('Chunked Upload wird nicht unterstützt.')
try:length=int(self.headers.get('Content-Length','0'))
except ValueError:raise APIError('Ungültige Content-Length.') from None
@@ -161,6 +162,7 @@ class APIHandler(BaseHTTPRequestHandler):
except (ValueError,UnicodeError):raise APIError('Ungültiges JSON.') from None
if not isinstance(data,dict):raise APIError('JSON-Objekt erforderlich.')
if self.path=='/v1/chat/completions':return self.chat(ep,data)
if self.path=='/v1/audio/speech':return self.speech(ep,data)
return self.image(ep,data)
except (APIError,InferenceError) as exc:
if isinstance(exc,InferenceError):
@@ -172,6 +174,21 @@ class APIHandler(BaseHTTPRequestHandler):
self.close_connection=True
if admitted:
with ep.lock:ep.inflight-=1
def speech(self,ep,data):
if not ep.tts:raise APIError('TTS-Laufzeit nicht eingerichtet.',501)
if set(data)-{'model','input','voice','response_format','speed','language'}:raise APIError('Nicht unterstützte TTS-Felder.')
if data.get('response_format','wav')!='wav':raise APIError('Derzeit wird nur WAV unterstützt; response_format=wav setzen.')
profile=ep.find_profile(data.get('model'),'audio')
try:job=ep.tts.start(profile['id'],data.get('input'),speaker=data.get('voice','Ryan'),language=data.get('language','Auto'),speed=data.get('speed'),wait=True)
except ValueError as exc:raise APIError(str(exc)) from None
end=time.monotonic()+620
while time.monotonic()<end:
current=ep.tts.status()['job']
if current and current['id']==job['id'] and current['state'] in ('failed','cancelled'):raise APIError(current['phase'],503)
try:body=ep.tts.audio(job['id'])
except (ValueError,OSError):time.sleep(.25);continue
self.sent=True;self.send_response(200);self.send_header('Content-Type','audio/wav');self.send_header('Content-Length',str(len(body)));self.send_header('Cache-Control','no-store');self.send_header('Connection','close');self.end_headers();self.wfile.write(body);return
raise APIError('TTS-Zeitlimit überschritten.',504)
def chat(self,ep,data):
supported={'model','messages','stream','stream_options','temperature','top_p','top_k','max_tokens','max_completion_tokens','stop','seed','tools','tool_choice','parallel_tool_calls','response_format','presence_penalty','frequency_penalty','logprobs','top_logprobs','user','n'}
if set(data)-supported:raise APIError('Nicht unterstützte Chat-Felder: '+', '.join(sorted(set(data)-supported)))