dschlueter/my_voice_assistant_v3
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

fix(llm): LLM antwortet jetzt in der konfigurierten Systemsprache

Browse source 
Bisher wurde `language` zwar an STT/TTS übergeben, aber nie an den LLM.
Lokales Modell antwortete deshalb immer auf Englisch/Deutsch, egal ob
Französisch, Spanisch o.a. eingestellt war.

Lösung: `language`-Parameter durch die gesamte LLM-Schicht gezogen:
- base.py: `lang_instruction()` helper + Signatur erweitert
- local_openai_compatible.py: Sprachanweisung ("Respond in Français.")
  wird als letzter System-Part in den Message-Stack eingefügt
- openrouter.py: Explizite Sprachanweisung ergänzt den bestehenden
  "Answer in the same language as the user"-Prompt
- fallback.py: FallbackLLMProvider leitet `language` durch
- orchestrator.py: alle 3 LLM-Aufrufstellen übergeben `language`
- Tests: alle Stub-LLMs um `language=None` ergänzt

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Dieter Schlüter 2026-06-19 14:12:16 +02:00
commit 81d4cf6fd8
14 changed files with 89 additions and 44 deletions
Download patch file Download diff file Expand all files Collapse all files

2
tests/test_emergency_llm.py
View file

@ -12,7 +12,7 @@ class StubLLM:
self.raw = raw
self.calls = 0
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
self.calls += 1
return self.raw

2
tests/test_endpoints_e2e.py
View file

@ -59,7 +59,7 @@ def test_session_route_applies(monkeypatch):
def test_chat_per_request_override_and_loopback(monkeypatch):
class StubLLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
return "Mir geht es gut, danke."
class StubTTS:

2
tests/test_memory.py
View file

@ -8,7 +8,7 @@ client = TestClient(app)
def _install_stubs(monkeypatch, captured):
class MemLLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
captured["history"] = list(history or [])
return f"Antwort auf: {text}"

2
tests/test_memory_extraction.py
View file

@ -12,7 +12,7 @@ class StubLLM:
self.raw = raw
self.calls = 0
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
self.calls += 1
return self.raw

2
tests/test_quota_safety.py
View file

@ -10,7 +10,7 @@ client = TestClient(app)
def _stub_chat(monkeypatch, answer="ok"):
class StubLLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
return answer
class StubTTS:

6
tests/test_realtime.py
View file

@ -42,10 +42,10 @@ def test_energy_vad_ignores_silence_without_speech():
def _install_slow_stream(monkeypatch):
class SlowLLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
return "fertig"
async def stream(self, text, history=None, session_id=None):
async def stream(self, text, history=None, session_id=None, language=None):
for i in range(200):
await asyncio.sleep(0.005)
yield f"t{i} "
@ -89,7 +89,7 @@ def _install_voice_stubs(monkeypatch):
return "ok"
class LLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
return "antwort"
class TTS:

20
tests/test_resilience.py
View file

@ -20,11 +20,11 @@ def _run(coro):
def test_llm_fallback_uses_second_on_error():
class BadLLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
raise RuntimeError("down")
class GoodLLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
return "ok"
chain = FallbackLLMProvider("llm", [("bad", BadLLM()), ("good", GoodLLM())])
@ -37,7 +37,7 @@ def test_llm_fallback_uses_second_on_error():
def test_llm_fallback_all_fail_raises():
class BadLLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
raise RuntimeError("x")
chain = FallbackLLMProvider("llm", [("a", BadLLM()), ("b", BadLLM())])
@ -47,18 +47,18 @@ def test_llm_fallback_all_fail_raises():
def test_llm_stream_fallback_before_first_token():
class BadStream:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
return "x"
async def stream(self, text, history=None, session_id=None):
async def stream(self, text, history=None, session_id=None, language=None):
raise RuntimeError("boom")
yield # macht die Funktion zum Generator
class GoodStream:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
return "ok"
async def stream(self, text, history=None, session_id=None):
async def stream(self, text, history=None, session_id=None, language=None):
yield "he"
yield "llo"
@ -74,11 +74,11 @@ def test_llm_stream_fallback_before_first_token():
def test_config_llm_fallback_applied(monkeypatch):
class BadLLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
raise RuntimeError("primary down")
class GoodLLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
return "rescued"
class StubTTS:
@ -102,7 +102,7 @@ def test_config_llm_fallback_applied(monkeypatch):
def test_metrics_endpoint_records_requests_and_stages(monkeypatch):
class StubLLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
return "hi"
class StubTTS:

12
tests/test_streaming.py
View file

@ -40,7 +40,7 @@ def test_sse_delta_parsing():
def test_base_stream_default_yields_full_completion():
class P(LLMProvider):
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
return "ganze Antwort"
async def run():
@ -51,10 +51,10 @@ def test_base_stream_default_yields_full_completion():
def _install_streaming(monkeypatch, tokens):
class StreamLLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
return "".join(tokens)
async def stream(self, text, history=None, session_id=None):
async def stream(self, text, history=None, session_id=None, language=None):
for tok in tokens:
yield tok
@ -95,7 +95,7 @@ def test_ws_without_stream_flag_has_no_tokens(monkeypatch):
def test_ws_stream_fallback_for_nonstreaming_llm(monkeypatch):
class OnlyComplete:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
return "komplett"
class StubTTS:
@ -118,10 +118,10 @@ def test_ws_audio_stream_sends_chunks_per_sentence(monkeypatch):
tts_calls = []
class StreamLLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
return "Satz eins. Satz zwei."
async def stream(self, text, history=None, session_id=None):
async def stream(self, text, history=None, session_id=None, language=None):
for tok in ["Satz ", "eins. ", "Satz ", "zwei."]:
yield tok

4
tests/test_ws.py
View file

@ -11,7 +11,7 @@ client = TestClient(app)
def _install_stubs(monkeypatch, captured):
class MemLLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
captured["history"] = list(history or [])
return f"Echo: {text}"
@ -74,7 +74,7 @@ def _install_voice_stubs(monkeypatch):
return f"erkannt({len(audio_bytes)})"
class StubLLM:
async def complete(self, text, history=None, session_id=None):
async def complete(self, text, history=None, session_id=None, language=None):
return f"Antwort zu {text}"
class StubTTS: