125 lines
2.8 KiB
Markdown
125 lines
2.8 KiB
Markdown
|
# pgvector Setup Guide for PostgreSQL 16
|
||
|
|
|
||
|
|
## 1. Install pgvector on your server
|
||
|
|
|
||
|
|
### Debian / Ubuntu
|
||
|
|
```bash
|
||
|
|
sudo apt install postgresql-16-pgvector
|
||
|
|
```
|
||
|
|
|
||
|
|
### From source (if package not available)
|
||
|
|
```bash
|
||
|
|
sudo apt install postgresql-server-dev-16 build-essential git
|
||
|
|
git clone https://github.com/pgvector/pgvector.git
|
||
|
|
cd pgvector
|
||
|
|
make
|
||
|
|
sudo make install
|
||
|
|
```
|
||
|
|
|
||
|
|
### Arch Linux
|
||
|
|
```bash
|
||
|
|
yay -S pgvector
|
||
|
|
# or
|
||
|
|
paru -S pgvector
|
||
|
|
```
|
||
|
|
|
||
|
|
### Docker (if running PostgreSQL in a container)
|
||
|
|
Use the official pgvector image instead:
|
||
|
|
```dockerfile
|
||
|
|
FROM pgvector/pgvector:pg16
|
||
|
|
```
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## 2. Enable the extension in your database
|
||
|
|
|
||
|
|
Connect to your database and run:
|
||
|
|
```sql
|
||
|
|
\c manel_bot
|
||
|
|
CREATE EXTENSION IF NOT EXISTS vector;
|
||
|
|
```
|
||
|
|
|
||
|
|
The bot also does this automatically on startup via `init_db()`.
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## 3. Create the vector index (do this after ~1000 messages)
|
||
|
|
|
||
|
|
Running this on an empty table does nothing useful. Wait until you have
|
||
|
|
a good amount of history, then run:
|
||
|
|
|
||
|
|
```sql
|
||
|
|
\c manel_bot
|
||
|
|
CREATE INDEX idx_history_embedding
|
||
|
|
ON history USING ivfflat (embedding vector_cosine_ops)
|
||
|
|
WITH (lists = 100);
|
||
|
|
```
|
||
|
|
|
||
|
|
**Why wait?** IVFFlat needs enough data to build meaningful clusters.
|
||
|
|
Until then the bot works fine via sequential scan — just slightly slower
|
||
|
|
for very large histories.
|
||
|
|
|
||
|
|
For smaller deployments (<100k rows) you can also use HNSW which
|
||
|
|
doesn't need pre-training:
|
||
|
|
```sql
|
||
|
|
CREATE INDEX idx_history_embedding
|
||
|
|
ON history USING hnsw (embedding vector_cosine_ops)
|
||
|
|
WITH (m = 16, ef_construction = 64);
|
||
|
|
```
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## 4. Check the dimension matches your model
|
||
|
|
|
||
|
|
nomic-embed-text-v2-moe outputs **768-dimensional** vectors.
|
||
|
|
This is set in `.env_credentials` as `EMBED_DIM=768`.
|
||
|
|
|
||
|
|
If you ever switch embedding models, update `EMBED_DIM` to match
|
||
|
|
and recreate the history table (the vector column dimension is fixed):
|
||
|
|
```sql
|
||
|
|
-- Only if switching models — this drops all history!
|
||
|
|
ALTER TABLE history DROP COLUMN embedding;
|
||
|
|
ALTER TABLE history ADD COLUMN embedding vector(NEW_DIM);
|
||
|
|
```
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## 5. Verify everything is working
|
||
|
|
|
||
|
|
```sql
|
||
|
|
\c manel_bot
|
||
|
|
SELECT extname, extversion FROM pg_extension WHERE extname = 'vector';
|
||
|
|
-- Should return: vector | 0.8.x
|
||
|
|
|
||
|
|
SELECT COUNT(*), COUNT(embedding) FROM history;
|
||
|
|
-- After some messages: total count and how many have embeddings
|
||
|
|
```
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Ollama model setup
|
||
|
|
|
||
|
|
Make sure the model is pulled on your Ollama machine:
|
||
|
|
```bash
|
||
|
|
ollama pull nomic-embed-text-v2-moe
|
||
|
|
```
|
||
|
|
|
||
|
|
Test it's reachable from the bot server:
|
||
|
|
```bash
|
||
|
|
curl http://YOUR_OLLAMA_IP:11434/api/embed \
|
||
|
|
-d '{"model":"nomic-embed-text-v2-moe","input":"test"}'
|
||
|
|
```
|
||
|
|
|
||
|
|
Should return a JSON with an `embeddings` array.
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
sudo -u postgres psql
|
||
|
|
sqlCREATE USER manel WITH PASSWORD 'choose_a_password';
|
||
|
|
CREATE DATABASE manel_bot OWNER manel;
|
||
|
|
GRANT ALL PRIVILEGES ON DATABASE manel_bot TO manel;
|
||
|
|
get in the database and
|
||
|
|
CREATE EXTENSION IF NOT EXISTS vector;
|
||
|
|
|
||
|
|
python bot.py 2>&1 | tee manel.log
|