Files
Assistant/adapters/models/ollama.go

170 lines
6.2 KiB
Go

package models
import (
"bytes"
"context"
"encoding/json"
"errors"
"io"
"net/http"
"net/url"
"time"
"git.cin.su/m/Assistant/contracts/capability"
"git.cin.su/m/Assistant/contracts/cognitive"
)
type OllamaAdapter struct {
base *url.URL
client *http.Client
}
func NewOllamaAdapter(raw string, transport http.RoundTripper) (*OllamaAdapter, error) {
base, err := url.Parse(raw)
if err != nil || base.Scheme != "http" || base.Host == "" || base.Path != "" || base.User != nil || base.RawQuery != "" || base.Fragment != "" {
return nil, errors.New("invalid Ollama endpoint")
}
if transport == nil {
transport = http.DefaultTransport
}
return &OllamaAdapter{base: base, client: &http.Client{Transport: transport, Timeout: 10 * time.Second, CheckRedirect: func(*http.Request, []*http.Request) error { return http.ErrUseLastResponse }}}, nil
}
type ollamaTags struct {
Models []struct {
Name string `json:"name"`
Model string `json:"model"`
Size int64 `json:"size"`
Digest string `json:"digest"`
Details struct {
Format string `json:"format"`
Family string `json:"family"`
ParameterSize string `json:"parameter_size"`
Quantization string `json:"quantization_level"`
ContextLength int64 `json:"context_length"`
} `json:"details"`
Capabilities []string `json:"capabilities"`
} `json:"models"`
}
type ollamaVersion struct {
Version string `json:"version"`
}
func (a *OllamaAdapter) Probe(ctx context.Context) ProviderProbe {
var version ollamaVersion
status, err := a.request(ctx, http.MethodGet, "/api/version", nil, &version)
if err != nil {
return probeFailure(err)
}
if status != http.StatusOK || version.Version == "" {
return ProviderProbe{Health: capability.HealthUnavailable, Failure: mapCompatibleStatus(status)}
}
return ProviderProbe{Health: capability.HealthAvailable, Version: version.Version}
}
func (a *OllamaAdapter) Discover(ctx context.Context) ([]DiscoveredModel, ProviderProbe) {
probe := a.Probe(ctx)
if probe.Health != capability.HealthAvailable {
return nil, probe
}
var tags ollamaTags
status, err := a.request(ctx, http.MethodGet, "/api/tags", nil, &tags)
if err != nil {
return nil, probeFailure(err)
}
if status != http.StatusOK {
return nil, ProviderProbe{Health: capability.HealthUnavailable, Failure: mapCompatibleStatus(status)}
}
result := make([]DiscoveredModel, 0, len(tags.Models))
for _, item := range tags.Models {
if item.Name == "" || item.Digest == "" {
continue
}
capabilities := []capability.ModelCapability{capability.CapabilityUnknown}
for _, declared := range item.Capabilities {
if declared == "completion" {
capabilities = []capability.ModelCapability{capability.CapabilityAdvisory}
break
}
}
result = append(result, DiscoveredModel{ProviderModelID: capability.ProviderModelID(item.Name), Fingerprint: capability.ModelFingerprint(item.Digest), MaxContextBytes: item.Details.ContextLength * 4, Modalities: []capability.ModelModality{capability.ModalityText}, Capabilities: capabilities, SizeBytes: item.Size, Runtime: "ollama/" + probe.Version, Quantization: item.Details.Quantization})
}
return result, probe
}
type ollamaChatRequest struct {
Model string `json:"model"`
Messages []compatibleMessage `json:"messages"`
Stream bool `json:"stream"`
Think bool `json:"think"`
Options struct {
Temperature int `json:"temperature"`
NumCtx int64 `json:"num_ctx,omitempty"`
} `json:"options"`
}
type ollamaChatResponse struct {
Message compatibleMessage `json:"message"`
Done bool `json:"done"`
DoneReason string `json:"done_reason"`
PromptEvalCount int64 `json:"prompt_eval_count"`
EvalCount int64 `json:"eval_count"`
}
func (a *OllamaAdapter) Advise(ctx context.Context, request cognitive.AdvisoryRequest) (cognitive.AdvisoryResponse, error) {
if err := ValidateAdvisoryRequest(request); err != nil || request.ProviderModelID == "" || len(request.Context) != 0 {
return cognitive.AdvisoryResponse{}, ErrInvalidAdvisory
}
payload := ollamaChatRequest{Model: string(request.ProviderModelID), Messages: []compatibleMessage{{Role: "system", Content: "Provide non-canonical advisory text only. Never claim that an action occurred."}, {Role: "user", Content: request.Message}}, Stream: false, Think: false}
var result ollamaChatResponse
status, err := a.request(ctx, http.MethodPost, "/api/chat", payload, &result)
if err != nil {
return advisoryFailure(request, probeFailure(err).Failure), nil
}
if status != http.StatusOK {
return advisoryFailure(request, mapCompatibleStatus(status)), nil
}
if !result.Done || result.Message.Content == "" {
return advisoryFailure(request, cognitive.FinishInvalidProviderResponse), nil
}
response := cognitive.AdvisoryResponse{Version: request.Version, RequestID: request.RequestID, ProviderBindingID: request.ProviderBindingID, ProviderBindingRevision: request.ProviderBindingRevision, AdvisoryText: result.Message.Content, FinishReason: cognitive.FinishCompleted, Usage: &cognitive.ProviderUsage{InputUnits: result.PromptEvalCount, OutputUnits: result.EvalCount}}
if ValidateAdvisoryResponse(request, response) != nil {
return advisoryFailure(request, cognitive.FinishInvalidProviderResponse), nil
}
return response, nil
}
func (a *OllamaAdapter) request(ctx context.Context, method, path string, input, output any) (int, error) {
var body io.Reader
if input != nil {
encoded, err := json.Marshal(input)
if err != nil {
return 0, err
}
body = bytes.NewReader(encoded)
}
target := *a.base
target.Path = path
request, err := http.NewRequestWithContext(ctx, method, target.String(), body)
if err != nil {
return 0, err
}
request.Header.Set("Accept", "application/json")
if input != nil {
request.Header.Set("Content-Type", "application/json")
}
response, err := a.client.Do(request)
if err != nil {
return 0, err
}
defer response.Body.Close()
if response.StatusCode != http.StatusOK {
_, _ = io.Copy(io.Discard, io.LimitReader(response.Body, 4096))
return response.StatusCode, nil
}
decoder := json.NewDecoder(io.LimitReader(response.Body, 2<<20))
if err := decoder.Decode(output); err != nil {
return response.StatusCode, err
}
return response.StatusCode, nil
}