admin/ai-service
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Switch transcription comparison to Voxtral
All checks were successful
CI / test (push) Successful in 13s
Details
Build and Deploy / build-and-deploy (push) Successful in 18s
Details

Browse Source
This commit is contained in:
Grendgi
2026-06-09 15:11:52 +03:00
parent add15f1385
commit 94e0d03580
2 changed files with 7 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
deploy/ai-server/docker-compose.audio.yml
View File

@@ -103,15 +103,15 @@ services:
- "--port" - "--port"
- "8000" - "8000"
- "--max-model-len" - "--max-model-len"
- "32768" - "16384"
- "--gpu-memory-utilization" - "--gpu-memory-utilization"
- "0.62" - "0.55"
- "--api-key" - "--api-key"
- "${VLLM_API_KEY}" - "${VLLM_API_KEY}"
- "--max-num-seqs" - "--max-num-seqs"
- "2" - "1"
- "--max-num-batched-tokens" - "--max-num-batched-tokens"
- "8192" - "4096"
healthcheck: healthcheck:
test: ["CMD", "python3", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:8000/health')"] test: ["CMD", "python3", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:8000/health')"]
interval: 30s interval: 30s

6
k8s/configmap.yaml
View File

@@ -11,16 +11,16 @@ data:
LLM_BASE_URL: "http://10.2.3.5:8002" LLM_BASE_URL: "http://10.2.3.5:8002"
LLM_MODEL: "qwen2.5-14b" LLM_MODEL: "qwen2.5-14b"
LLM_TIMEOUT: "5m" LLM_TIMEOUT: "5m"
TRANSCRIPTION_PROVIDERS: "whisperx,qwen2-audio,voxtral-small" TRANSCRIPTION_PROVIDERS: "voxtral-small"
WHISPERX_URL: "http://10.2.3.5:8001" WHISPERX_URL: "http://10.2.3.5:8001"
WHISPERX_TIMEOUT: "10m" WHISPERX_TIMEOUT: "10m"
WHISPERX_LEAD_SILENCE: "800ms" WHISPERX_LEAD_SILENCE: "800ms"
# Fill these after Qwen2-Audio and Voxtral are exposed as OpenAI-compatible # Fill these after Qwen2-Audio and Voxtral are exposed as OpenAI-compatible
# chat-completions endpoints on the AI server. # chat-completions endpoints on the AI server.
QWEN_AUDIO_BASE_URL: "http://10.2.3.5:8003" QWEN_AUDIO_BASE_URL: ""
QWEN_AUDIO_MODEL: "Qwen/Qwen2-Audio-7B-Instruct" QWEN_AUDIO_MODEL: "Qwen/Qwen2-Audio-7B-Instruct"
QWEN_AUDIO_TIMEOUT: "10m" QWEN_AUDIO_TIMEOUT: "10m"
VOXTRAL_BASE_URL: "" VOXTRAL_BASE_URL: "http://10.2.3.5:8004"
VOXTRAL_MODEL: "mistralai/Voxtral-Small-24B-2507" VOXTRAL_MODEL: "mistralai/Voxtral-Small-24B-2507"
VOXTRAL_TIMEOUT: "10m" VOXTRAL_TIMEOUT: "10m"
AUDIO_LLM_MAX_TOKENS: "4096" AUDIO_LLM_MAX_TOKENS: "4096"