Skip to content

CONFIGURATION.md

Latest commit

 

History

History
236 lines (170 loc) · 6.44 KB

CONFIGURATION.md

File metadata and controls

236 lines (170 loc) · 6.44 KB

Configuration Guide

Complete reference for configuring MeetMemo.

Environment Variables

Create a .env file in the project root:

cp example.env .env

Required Variables

Variable Description Example
HF_TOKEN Hugging Face API token for PyAnnote models hf_abc...
LLM_API_URL LLM endpoint base URL (no /v1/chat/completions suffix) http://localhost:1234
LLM_MODEL_NAME Model identifier qwen2.5-14b-instruct
DATABASE_URL PostgreSQL connection string (auto-set in Docker) postgresql://...

Optional Variables

Variable Description Default
LLM_API_KEY API key for LLM service Empty (none)
POSTGRES_PASSWORD PostgreSQL password changeme
WHISPER_MODEL_NAME Whisper model for transcription turbo
COMPUTE_TYPE Inference precision (float16/int8) float16
TIMEZONE_OFFSET Timezone offset from UTC (hours) +8
NVIDIA_VISIBLE_DEVICES GPU selection (all, 0, 0,1) all
HTTP_PORT External HTTP port for nginx 80
HTTPS_PORT External HTTPS port for nginx 443

Whisper Model Selection

MeetMemo uses faster-whisper with CTranslate2 for 4x faster transcription compared to openai-whisper, while maintaining the same accuracy and 99+ language support.

Set WHISPER_MODEL_NAME environment variable:

Model VRAM Speed Accuracy Use Case
tiny ~1GB Fastest Basic Quick drafts, testing
base ~1GB Fast Good General use
small ~2GB Moderate Better Most meetings
medium ~5GB Slow High Important recordings
large ~10GB Slowest Highest Critical accuracy needs
turbo ~6GB Fast High Default - Best balance
large-v3 ~10GB Very Slow Highest+ 10-20% better than large-v2

Compute Type Configuration

Control inference precision with COMPUTE_TYPE environment variable:

Compute Type Memory Usage Speed Quality Best For
float16 Medium Fast High GPU (default, recommended)
int8 Low Very Fast Good CPU or low VRAM
int8_float16 Low-Medium Fast Good-High Hybrid scenarios

Example:

# In .env file
WHISPER_MODEL_NAME=turbo
COMPUTE_TYPE=float16

File Storage Limits

Configure in backend/config.py:

max_file_size: int = 100 * 1024 * 1024  # 100MB
allowed_audio_types: list[str] = [
    'audio/wav', 'audio/mpeg', 'audio/mp4',
    'audio/x-m4a', 'audio/webm', 'audio/flac', 'audio/ogg'
]

Cleanup Configuration

Auto-cleanup of old jobs and exports:

cleanup_interval_hours: int = 1      # Check every hour
job_retention_hours: int = 12        # Keep jobs for 12 hours
export_retention_hours: int = 24     # Keep exports for 24 hours

Docker Volumes

All runtime data is stored in named Docker volumes:

Volume Purpose Path in Container
meetmemo_audiofiles Uploaded audio files /app/audiofiles
meetmemo_transcripts Transcription JSONs /app/transcripts
meetmemo_summary Summary files /app/summary
meetmemo_exports PDF/Markdown exports /app/exports
meetmemo_logs Application logs /app/logs
meetmemo_whisper_cache Legacy (unused) /root/.cache/whisper
meetmemo_huggingface_cache Whisper + PyAnnote models /root/.cache/huggingface
meetmemo_torch_cache PyTorch cache /root/.cache/torch
meetmemo_postgres_data Database /var/lib/postgresql/data

Database Configuration

PostgreSQL connection pool settings in backend/config.py:

db_pool_min_size: int = 5    # Minimum connections
db_pool_max_size: int = 20   # Maximum connections

Logging Configuration

Configure in backend/config.py:

log_level: str = "INFO"              # DEBUG, INFO, WARNING, ERROR, CRITICAL
log_file: str = "logs/app.log"
log_max_bytes: int = 10 * 1024 * 1024   # 10MB per file
log_backup_count: int = 5                # Keep 5 rotated files
log_to_console: bool = True              # Also output to stdout

GPU Configuration

Select Specific GPUs

In .env:

NVIDIA_VISIBLE_DEVICES=0        # Use only GPU 0
NVIDIA_VISIBLE_DEVICES=0,1      # Use GPUs 0 and 1
NVIDIA_VISIBLE_DEVICES=all      # Use all GPUs (default)

Disable GPU (CPU-only mode)

Remove the runtime: nvidia line from docker-compose.yml:

meetmemo-backend:
  # runtime: nvidia   # Comment out this line

Note: CPU-only mode is significantly slower for transcription and diarization.

Timezone Configuration

Set your local timezone offset:

TIMEZONE_OFFSET=+8     # GMT+8 (Asia/Singapore, Beijing)
TIMEZONE_OFFSET=-5     # GMT-5 (US Eastern)
TIMEZONE_OFFSET=0      # GMT (UTC)

This affects timestamps in generated files and logs.

LLM Configuration

OpenAI-Compatible APIs

MeetMemo works with any OpenAI-compatible API:

# Local LM Studio
LLM_API_URL=http://localhost:1234
LLM_MODEL_NAME=qwen2.5-14b-instruct
LLM_API_KEY=

# Ollama with OpenAI compatibility
LLM_API_URL=http://localhost:11434/v1
LLM_MODEL_NAME=llama3
LLM_API_KEY=

# OpenAI
LLM_API_URL=https://api.openai.com/v1
LLM_MODEL_NAME=gpt-4
LLM_API_KEY=sk-...

Timeout Settings

Adjust LLM timeout in backend/config.py:

llm_timeout: float = 60.0  # 60 seconds

Port Configuration

Using Environment Variables (Recommended)

Configure external ports via .env file:

HTTP_PORT=8080
HTTPS_PORT=8443

Then access via https://localhost:8443

This is the recommended approach as it keeps your configuration in one place and doesn't require editing docker-compose.yml.

Manual Configuration

Alternatively, you can edit docker-compose.yml directly:

nginx:
  ports:
    - "8080:80"      # HTTP (redirects to HTTPS)
    - "8443:443"     # HTTPS

Note: If both environment variables and manual configuration are present, environment variables take precedence.

Nginx Configuration

SSL/TLS settings in nginx/nginx.conf:

ssl_protocols TLSv1.2 TLSv1.3;
ssl_ciphers HIGH:!aNULL:!MD5;
client_max_body_size 500M;  # Max upload size

Proxy timeouts for long-running transcriptions:

proxy_connect_timeout 600s;
proxy_send_timeout 600s;
proxy_read_timeout 600s;