e5e6908416
Fixes across 25 files addressing 46+ review comments: - pkg/ai/metrics.go: handle error from Close() on writable file handle - pkg/ansible: restore loop vars after loop, restore become settings, fix Upload with become=true and no password (use sudo -n), honour SSH timeout config, use E() helper for contextual errors, quote git refs in checkout commands - pkg/rag: validate chunk config, guard negative-to-uint64 conversion, use E() helper for errors, add context timeout to Ollama HTTP calls - pkg/deploy/python: fix exec.ExitError type assertion (was os.PathError), handle os.UserHomeDir() error - pkg/build/buildcmd: use cmd.Context() instead of context.Background() for proper Ctrl+C cancellation - install.bat: add curl timeouts, CRLF line endings, use --connect-timeout for archive downloads - install.sh: use absolute path for version check in CI mode - tools/rag: fix broken ingest.py function def, escape HTML in query.py, pin qdrant-client version, add markdown code block languages - internal/cmd/rag: add chunk size validation, env override handling Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
120 lines
No EOL
2.9 KiB
Go
120 lines
No EOL
2.9 KiB
Go
package rag
|
|
|
|
import (
|
|
"context"
|
|
"fmt"
|
|
"net/http"
|
|
"net/url"
|
|
"time"
|
|
|
|
"github.com/host-uk/core/pkg/log"
|
|
"github.com/ollama/ollama/api"
|
|
)
|
|
|
|
// OllamaConfig holds Ollama connection configuration.
|
|
type OllamaConfig struct {
|
|
Host string
|
|
Port int
|
|
Model string
|
|
}
|
|
|
|
// DefaultOllamaConfig returns default Ollama configuration.
|
|
// Host defaults to localhost for local development.
|
|
func DefaultOllamaConfig() OllamaConfig {
|
|
return OllamaConfig{
|
|
Host: "localhost",
|
|
Port: 11434,
|
|
Model: "nomic-embed-text",
|
|
}
|
|
}
|
|
|
|
// OllamaClient wraps the Ollama API client for embeddings.
|
|
type OllamaClient struct {
|
|
client *api.Client
|
|
config OllamaConfig
|
|
}
|
|
|
|
// NewOllamaClient creates a new Ollama client.
|
|
func NewOllamaClient(cfg OllamaConfig) (*OllamaClient, error) {
|
|
baseURL := &url.URL{
|
|
Scheme: "http",
|
|
Host: fmt.Sprintf("%s:%d", cfg.Host, cfg.Port),
|
|
}
|
|
|
|
client := api.NewClient(baseURL, &http.Client{
|
|
Timeout: 30 * time.Second,
|
|
})
|
|
|
|
return &OllamaClient{
|
|
client: client,
|
|
config: cfg,
|
|
}, nil
|
|
}
|
|
|
|
// EmbedDimension returns the embedding dimension for the configured model.
|
|
// nomic-embed-text uses 768 dimensions.
|
|
func (o *OllamaClient) EmbedDimension() uint64 {
|
|
switch o.config.Model {
|
|
case "nomic-embed-text":
|
|
return 768
|
|
case "mxbai-embed-large":
|
|
return 1024
|
|
case "all-minilm":
|
|
return 384
|
|
default:
|
|
return 768 // Default to nomic-embed-text dimension
|
|
}
|
|
}
|
|
|
|
// Embed generates embeddings for the given text.
|
|
func (o *OllamaClient) Embed(ctx context.Context, text string) ([]float32, error) {
|
|
req := &api.EmbedRequest{
|
|
Model: o.config.Model,
|
|
Input: text,
|
|
}
|
|
|
|
resp, err := o.client.Embed(ctx, req)
|
|
if err != nil {
|
|
return nil, log.E("rag.Ollama.Embed", "failed to generate embedding", err)
|
|
}
|
|
|
|
if len(resp.Embeddings) == 0 || len(resp.Embeddings[0]) == 0 {
|
|
return nil, log.E("rag.Ollama.Embed", "empty embedding response", nil)
|
|
}
|
|
|
|
// Convert float64 to float32 for Qdrant
|
|
embedding := resp.Embeddings[0]
|
|
result := make([]float32, len(embedding))
|
|
for i, v := range embedding {
|
|
result[i] = float32(v)
|
|
}
|
|
|
|
return result, nil
|
|
}
|
|
|
|
// EmbedBatch generates embeddings for multiple texts.
|
|
func (o *OllamaClient) EmbedBatch(ctx context.Context, texts []string) ([][]float32, error) {
|
|
results := make([][]float32, len(texts))
|
|
for i, text := range texts {
|
|
embedding, err := o.Embed(ctx, text)
|
|
if err != nil {
|
|
return nil, log.E("rag.Ollama.EmbedBatch", fmt.Sprintf("failed to embed text %d", i), err)
|
|
}
|
|
results[i] = embedding
|
|
}
|
|
return results, nil
|
|
}
|
|
|
|
// VerifyModel checks if the embedding model is available.
|
|
func (o *OllamaClient) VerifyModel(ctx context.Context) error {
|
|
_, err := o.Embed(ctx, "test")
|
|
if err != nil {
|
|
return log.E("rag.Ollama.VerifyModel", fmt.Sprintf("model %s not available (run: ollama pull %s)", o.config.Model, o.config.Model), err)
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// Model returns the configured embedding model name.
|
|
func (o *OllamaClient) Model() string {
|
|
return o.config.Model
|
|
} |