Port LEM scoring/training pipeline into CoreGo as pkg/ml with: - Inference abstraction with HTTP, llama-server, and Ollama backends - 3-tier scoring engine (heuristic, exact, LLM judge) - Capability and content probes for model evaluation - GGUF/safetensors format converters, MLX to PEFT adapter conversion - DuckDB integration for training data pipeline - InfluxDB metrics for lab dashboard - Training data export (JSONL + Parquet) - Expansion generation pipeline with distributed workers - 10 CLI commands under 'core ml' (score, probe, export, expand, status, gguf, convert, agent, worker) - 5 MCP tools (ml_generate, ml_score, ml_probe, ml_status, ml_backends) All 37 ML tests passing. Binary builds at 138MB with all commands. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
152 lines
4.2 KiB
Go
152 lines
4.2 KiB
Go
package ml
|
|
|
|
import (
|
|
"bytes"
|
|
"crypto/sha256"
|
|
"encoding/hex"
|
|
"encoding/json"
|
|
"fmt"
|
|
"io"
|
|
"net/http"
|
|
"os"
|
|
"time"
|
|
)
|
|
|
|
// OllamaBaseModelMap maps model tags to Ollama model names.
|
|
var OllamaBaseModelMap = map[string]string{
|
|
"gemma-3-1b": "gemma3:1b",
|
|
"gemma-3-4b": "gemma3:4b",
|
|
"gemma-3-12b": "gemma3:12b",
|
|
"gemma-3-27b": "gemma3:27b",
|
|
}
|
|
|
|
// HFBaseModelMap maps model tags to HuggingFace model IDs.
|
|
var HFBaseModelMap = map[string]string{
|
|
"gemma-3-1b": "google/gemma-3-1b-it",
|
|
"gemma-3-4b": "google/gemma-3-4b-it",
|
|
"gemma-3-12b": "google/gemma-3-12b-it",
|
|
"gemma-3-27b": "google/gemma-3-27b-it",
|
|
}
|
|
|
|
// ollamaUploadBlob uploads a local file to Ollama's blob store.
|
|
// Returns the sha256 digest string (e.g. "sha256:abc123...").
|
|
func ollamaUploadBlob(ollamaURL, filePath string) (string, error) {
|
|
data, err := os.ReadFile(filePath)
|
|
if err != nil {
|
|
return "", fmt.Errorf("read %s: %w", filePath, err)
|
|
}
|
|
|
|
hash := sha256.Sum256(data)
|
|
digest := "sha256:" + hex.EncodeToString(hash[:])
|
|
|
|
headReq, _ := http.NewRequest(http.MethodHead, ollamaURL+"/api/blobs/"+digest, nil)
|
|
client := &http.Client{Timeout: 5 * time.Minute}
|
|
headResp, err := client.Do(headReq)
|
|
if err == nil && headResp.StatusCode == http.StatusOK {
|
|
headResp.Body.Close()
|
|
return digest, nil
|
|
}
|
|
if headResp != nil {
|
|
headResp.Body.Close()
|
|
}
|
|
|
|
req, err := http.NewRequest(http.MethodPost, ollamaURL+"/api/blobs/"+digest, bytes.NewReader(data))
|
|
if err != nil {
|
|
return "", fmt.Errorf("blob request: %w", err)
|
|
}
|
|
req.Header.Set("Content-Type", "application/octet-stream")
|
|
|
|
resp, err := client.Do(req)
|
|
if err != nil {
|
|
return "", fmt.Errorf("blob upload: %w", err)
|
|
}
|
|
defer resp.Body.Close()
|
|
|
|
if resp.StatusCode != http.StatusCreated && resp.StatusCode != http.StatusOK {
|
|
body, _ := io.ReadAll(resp.Body)
|
|
return "", fmt.Errorf("blob upload HTTP %d: %s", resp.StatusCode, string(body))
|
|
}
|
|
return digest, nil
|
|
}
|
|
|
|
// OllamaCreateModel creates a temporary Ollama model with a LoRA adapter.
|
|
// peftDir is a local directory containing adapter_model.safetensors and adapter_config.json.
|
|
func OllamaCreateModel(ollamaURL, modelName, baseModel, peftDir string) error {
|
|
sfPath := peftDir + "/adapter_model.safetensors"
|
|
cfgPath := peftDir + "/adapter_config.json"
|
|
|
|
sfDigest, err := ollamaUploadBlob(ollamaURL, sfPath)
|
|
if err != nil {
|
|
return fmt.Errorf("upload adapter safetensors: %w", err)
|
|
}
|
|
|
|
cfgDigest, err := ollamaUploadBlob(ollamaURL, cfgPath)
|
|
if err != nil {
|
|
return fmt.Errorf("upload adapter config: %w", err)
|
|
}
|
|
|
|
reqBody, _ := json.Marshal(map[string]interface{}{
|
|
"model": modelName,
|
|
"from": baseModel,
|
|
"adapters": map[string]string{
|
|
"adapter_model.safetensors": sfDigest,
|
|
"adapter_config.json": cfgDigest,
|
|
},
|
|
})
|
|
|
|
client := &http.Client{Timeout: 10 * time.Minute}
|
|
resp, err := client.Post(ollamaURL+"/api/create", "application/json", bytes.NewReader(reqBody))
|
|
if err != nil {
|
|
return fmt.Errorf("ollama create: %w", err)
|
|
}
|
|
defer resp.Body.Close()
|
|
|
|
decoder := json.NewDecoder(resp.Body)
|
|
for decoder.More() {
|
|
var status struct {
|
|
Status string `json:"status"`
|
|
Error string `json:"error"`
|
|
}
|
|
if err := decoder.Decode(&status); err != nil {
|
|
if err == io.EOF {
|
|
break
|
|
}
|
|
return fmt.Errorf("ollama create decode: %w", err)
|
|
}
|
|
if status.Error != "" {
|
|
return fmt.Errorf("ollama create: %s", status.Error)
|
|
}
|
|
if status.Status == "success" {
|
|
return nil
|
|
}
|
|
}
|
|
|
|
if resp.StatusCode != http.StatusOK {
|
|
return fmt.Errorf("ollama create: HTTP %d", resp.StatusCode)
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// OllamaDeleteModel removes a temporary Ollama model.
|
|
func OllamaDeleteModel(ollamaURL, modelName string) error {
|
|
body, _ := json.Marshal(map[string]string{"model": modelName})
|
|
|
|
req, err := http.NewRequest(http.MethodDelete, ollamaURL+"/api/delete", bytes.NewReader(body))
|
|
if err != nil {
|
|
return fmt.Errorf("ollama delete request: %w", err)
|
|
}
|
|
req.Header.Set("Content-Type", "application/json")
|
|
|
|
client := &http.Client{Timeout: 30 * time.Second}
|
|
resp, err := client.Do(req)
|
|
if err != nil {
|
|
return fmt.Errorf("ollama delete: %w", err)
|
|
}
|
|
defer resp.Body.Close()
|
|
|
|
if resp.StatusCode != http.StatusOK {
|
|
respBody, _ := io.ReadAll(resp.Body)
|
|
return fmt.Errorf("ollama delete %d: %s", resp.StatusCode, string(respBody))
|
|
}
|
|
return nil
|
|
}
|