|
5ac549c…
|
lmata
|
1 |
package llm |
|
5ac549c…
|
lmata
|
2 |
|
|
5ac549c…
|
lmata
|
3 |
import ( |
|
5ac549c…
|
lmata
|
4 |
"context" |
|
5ac549c…
|
lmata
|
5 |
"fmt" |
|
5ac549c…
|
lmata
|
6 |
"net/http" |
|
5ac549c…
|
lmata
|
7 |
) |
|
5ac549c…
|
lmata
|
8 |
|
|
5ac549c…
|
lmata
|
9 |
// KnownBackends maps OpenAI-compatible backend names to their default base URLs. |
|
5ac549c…
|
lmata
|
10 |
var KnownBackends = map[string]string{ |
|
5ac549c…
|
lmata
|
11 |
"openai": "https://api.openai.com/v1", |
|
5ac549c…
|
lmata
|
12 |
"openrouter": "https://openrouter.ai/api/v1", |
|
5ac549c…
|
lmata
|
13 |
"together": "https://api.together.xyz/v1", |
|
5ac549c…
|
lmata
|
14 |
"groq": "https://api.groq.com/openai/v1", |
|
5ac549c…
|
lmata
|
15 |
"fireworks": "https://api.fireworks.ai/inference/v1", |
|
5ac549c…
|
lmata
|
16 |
"mistral": "https://api.mistral.ai/v1", |
|
5ac549c…
|
lmata
|
17 |
"ai21": "https://api.ai21.com/studio/v1", |
|
5ac549c…
|
lmata
|
18 |
"huggingface": "https://api-inference.huggingface.co/v1", |
|
5ac549c…
|
lmata
|
19 |
"deepseek": "https://api.deepseek.com/v1", |
|
5ac549c…
|
lmata
|
20 |
"cerebras": "https://api.cerebras.ai/v1", |
|
5ac549c…
|
lmata
|
21 |
"xai": "https://api.x.ai/v1", |
|
5ac549c…
|
lmata
|
22 |
// Local / self-hosted (defaults — override with base_url) |
|
1066004…
|
lmata
|
23 |
"litellm": "http://localhost:4000/v1", |
|
1066004…
|
lmata
|
24 |
"lmstudio": "http://localhost:1234/v1", |
|
1066004…
|
lmata
|
25 |
"jan": "http://localhost:1337/v1", |
|
1066004…
|
lmata
|
26 |
"localai": "http://localhost:8080/v1", |
|
1066004…
|
lmata
|
27 |
"vllm": "http://localhost:8000/v1", |
|
5ac549c…
|
lmata
|
28 |
"anythingllm": "http://localhost:3001/v1", |
|
5ac549c…
|
lmata
|
29 |
} |
|
5ac549c…
|
lmata
|
30 |
|
|
5ac549c…
|
lmata
|
31 |
// New creates a Provider from the given config. The returned value may also |
|
5ac549c…
|
lmata
|
32 |
// implement ModelDiscoverer — check with a type assertion before calling |
|
5ac549c…
|
lmata
|
33 |
// DiscoverModels. Allow/block filters in cfg are applied transparently by |
|
5ac549c…
|
lmata
|
34 |
// wrapping the discoverer. |
|
5ac549c…
|
lmata
|
35 |
func New(cfg BackendConfig) (Provider, error) { |
|
5ac549c…
|
lmata
|
36 |
hc := &http.Client{} |
|
5ac549c…
|
lmata
|
37 |
switch cfg.Backend { |
|
5ac549c…
|
lmata
|
38 |
case "anthropic": |
|
5ac549c…
|
lmata
|
39 |
return newAnthropicProvider(cfg, hc), nil |
|
5ac549c…
|
lmata
|
40 |
|
|
5ac549c…
|
lmata
|
41 |
case "gemini": |
|
5ac549c…
|
lmata
|
42 |
return newGeminiProvider(cfg, hc), nil |
|
5ac549c…
|
lmata
|
43 |
|
|
5ac549c…
|
lmata
|
44 |
case "bedrock": |
|
5ac549c…
|
lmata
|
45 |
return newBedrockProvider(cfg, hc) |
|
5ac549c…
|
lmata
|
46 |
|
|
5ac549c…
|
lmata
|
47 |
case "ollama": |
|
5ac549c…
|
lmata
|
48 |
baseURL := cfg.BaseURL |
|
5ac549c…
|
lmata
|
49 |
if baseURL == "" { |
|
5ac549c…
|
lmata
|
50 |
baseURL = "http://localhost:11434" |
|
5ac549c…
|
lmata
|
51 |
} |
|
5ac549c…
|
lmata
|
52 |
return newOllamaProvider(cfg, baseURL, hc), nil |
|
5ac549c…
|
lmata
|
53 |
|
|
5ac549c…
|
lmata
|
54 |
default: |
|
5ac549c…
|
lmata
|
55 |
// OpenAI-compatible backend. |
|
5ac549c…
|
lmata
|
56 |
baseURL := cfg.BaseURL |
|
5ac549c…
|
lmata
|
57 |
if baseURL == "" { |
|
5ac549c…
|
lmata
|
58 |
u, ok := KnownBackends[cfg.Backend] |
|
5ac549c…
|
lmata
|
59 |
if !ok { |
|
5ac549c…
|
lmata
|
60 |
return nil, fmt.Errorf("llm: unknown backend %q — set base_url for custom endpoints", cfg.Backend) |
|
5ac549c…
|
lmata
|
61 |
} |
|
5ac549c…
|
lmata
|
62 |
baseURL = u |
|
5ac549c…
|
lmata
|
63 |
} |
|
5ac549c…
|
lmata
|
64 |
model := cfg.Model |
|
5ac549c…
|
lmata
|
65 |
if model == "" { |
|
5ac549c…
|
lmata
|
66 |
model = "gpt-4o-mini" |
|
5ac549c…
|
lmata
|
67 |
} |
|
5ac549c…
|
lmata
|
68 |
return newOpenAIProvider(cfg.APIKey, baseURL, model, hc), nil |
|
5ac549c…
|
lmata
|
69 |
} |
|
5ac549c…
|
lmata
|
70 |
} |
|
5ac549c…
|
lmata
|
71 |
|
|
5ac549c…
|
lmata
|
72 |
// Discover runs model discovery for the given config, applying any allow/block |
|
5ac549c…
|
lmata
|
73 |
// filters from the config. Returns an error if the provider doesn't support |
|
5ac549c…
|
lmata
|
74 |
// discovery. |
|
5ac549c…
|
lmata
|
75 |
func Discover(ctx context.Context, cfg BackendConfig) ([]ModelInfo, error) { |
|
5ac549c…
|
lmata
|
76 |
p, err := New(cfg) |
|
5ac549c…
|
lmata
|
77 |
if err != nil { |
|
5ac549c…
|
lmata
|
78 |
return nil, err |
|
5ac549c…
|
lmata
|
79 |
} |
|
5ac549c…
|
lmata
|
80 |
d, ok := p.(ModelDiscoverer) |
|
5ac549c…
|
lmata
|
81 |
if !ok { |
|
5ac549c…
|
lmata
|
82 |
return nil, fmt.Errorf("llm: backend %q does not support model discovery", cfg.Backend) |
|
5ac549c…
|
lmata
|
83 |
} |
|
5ac549c…
|
lmata
|
84 |
models, err := d.DiscoverModels(ctx) |
|
5ac549c…
|
lmata
|
85 |
if err != nil { |
|
5ac549c…
|
lmata
|
86 |
return nil, err |
|
5ac549c…
|
lmata
|
87 |
} |
|
5ac549c…
|
lmata
|
88 |
if len(cfg.Allow) > 0 || len(cfg.Block) > 0 { |
|
5ac549c…
|
lmata
|
89 |
f, ferr := NewModelFilter(cfg.Allow, cfg.Block) |
|
5ac549c…
|
lmata
|
90 |
if ferr != nil { |
|
5ac549c…
|
lmata
|
91 |
return nil, ferr |
|
5ac549c…
|
lmata
|
92 |
} |
|
5ac549c…
|
lmata
|
93 |
models = f.Apply(models) |
|
5ac549c…
|
lmata
|
94 |
} |
|
5ac549c…
|
lmata
|
95 |
return models, nil |
|
5ac549c…
|
lmata
|
96 |
} |
|
5ac549c…
|
lmata
|
97 |
|
|
5ac549c…
|
lmata
|
98 |
// BackendNames returns a sorted list of all known backend names. |
|
5ac549c…
|
lmata
|
99 |
func BackendNames() []string { |
|
5ac549c…
|
lmata
|
100 |
names := make([]string, 0, len(KnownBackends)+4) |
|
5ac549c…
|
lmata
|
101 |
for k := range KnownBackends { |
|
5ac549c…
|
lmata
|
102 |
names = append(names, k) |
|
5ac549c…
|
lmata
|
103 |
} |
|
5ac549c…
|
lmata
|
104 |
names = append(names, "anthropic", "gemini", "bedrock", "ollama") |
|
5ac549c…
|
lmata
|
105 |
return names |
|
5ac549c…
|
lmata
|
106 |
} |