feat: ✨ add model alias support and update defaults

microsoft · Dec 20, 2024 · c62171d · c62171d
1 parent f16d915
commit c62171d
Show file tree

Hide file tree

Showing 7 changed files with 241 additions and 113 deletions.
diff --git a/docs/public/schemas/llms.json b/docs/public/schemas/llms.json
@@ -58,13 +58,64 @@
                     "bearerToken": {
                         "type": "boolean",
                         "description": "Indicates if bearer token is supported"
+                    },
+                    "aliases": {
+                        "type": "object",
+                        "description": "List of model aliases for the provider",
+                        "properties": {
+                            "large": {
+                                "type": "string",
+                                "description": "Alias for large model"
+                            },
+                            "small": {
+                                "type": "string",
+                                "description": "Alias for small model"
+                            },
+                            "vision": {
+                                "type": "string",
+                                "description": "Alias for vision model"
+                            },
+                            "reasoning": {
+                                "type": "string",
+                                "description": "Alias for reasoning model"
+                            },
+                            "reasoning_small": {
+                                "type": "string",
+                                "description": "Alias for reasoning small model"
+                            },
+                            "long": {
+                                "type": "string",
+                                "description": "Alias for long model"
+                            },
+                            "agent": {
+                                "type": "string",
+                                "description": "Alias for agent model"
+                            },
+                            "memory": {
+                                "type": "string",
+                                "description": "Alias for memory model"
+                            },
+                            "embeddings": {
+                                "type": "string",
+                                "description": "Alias for embeddings model"
+                            }
+                        }
                     }
                 },
                 "additionalProperties": false,
                 "required": ["id", "detail"]
             }
         }
     },
+    "aliases": {
+        "type": "object",
+        "additionalProperties": true,
+        "patternProperties": {
+            "^[a-zA-Z0-9:_-]+$": {
+                "type": "string"
+            }
+        }
+    },
     "pricings": {
         "type": "object",
         "additionalProperties": false,

diff --git a/docs/src/content/docs/reference/cli/commands.md b/docs/src/content/docs/reference/cli/commands.md
@@ -1,3 +1,78 @@
+{
+  "large": {
+    "model": "openai:gpt-4o",
+    "source": "default",
+    "candidates": [
+      "openai:gpt-4o",
+      "azure_serverless:gpt-4o",
+      "anthropic:claude-2.1",
+      "google:gemini-1.5-pro-latest",
+      "github:gpt-4o",
+      "client:gpt-4o"
+    ]
+  },
+  "small": {
+    "model": "openai:gpt-4o-mini",
+    "source": "default",
+    "candidates": [
+      "openai:gpt-4o-mini",
+      "azure_serverless:gpt-4o-mini",
+      "anthropic:claude-instant-1.2",
+      "google:gemini-1.5-flash-latest",
+      "github:gpt-4o-mini",
+      "client:gpt-4o-mini"
+    ]
+  },
+  "vision": {
+    "model": "openai:gpt-4o",
+    "source": "default",
+    "candidates": [
+      "openai:gpt-4o",
+      "azure_serverless:gpt-4o",
+      "anthropic:claude-2.1",
+      "google:gemini-1.5-flash-latest",
+      "github:gpt-4o"
+    ]
+  },
+  "embeddings": {
+    "model": "openai:text-embedding-3-small",
+    "source": "default",
+    "candidates": [
+      "openai:text-embedding-3-small",
+      "github:text-embedding-3-small"
+    ]
+  },
+  "reasoning": {
+    "model": "openai:o1",
+    "source": "default",
+    "candidates": [
+      "openai:o1",
+      "azure_serverless:o1-preview",
+      "github:o1-preview"
+    ]
+  },
+  "reasoning_small": {
+    "model": "openai:o1-mini",
+    "source": "default",
+    "candidates": [
+      "openai:o1-mini",
+      "azure_serverless:o1-mini",
+      "github:o1-mini"
+    ]
+  },
+  "agent": {
+    "model": "large",
+    "source": "default"
+  },
+  "long": {
+    "model": "large",
+    "source": "default"
+  },
+  "memory": {
+    "model": "large",
+    "source": "default"
+  }
+}
 ---
 title: Commands
 description: List of all CLI commands

diff --git a/packages/cli/src/nodehost.ts b/packages/cli/src/nodehost.ts
@@ -17,27 +17,16 @@ import {
     parseTokenFromEnv,
 } from "../../core/src/connection"
 import {
-    DEFAULT_LARGE_MODEL,
     MODEL_PROVIDER_AZURE_OPENAI,
     SHELL_EXEC_TIMEOUT,
-    DEFAULT_EMBEDDINGS_MODEL,
-    DEFAULT_SMALL_MODEL,
     AZURE_COGNITIVE_SERVICES_TOKEN_SCOPES,
     MODEL_PROVIDER_AZURE_SERVERLESS_MODELS,
     AZURE_AI_INFERENCE_TOKEN_SCOPES,
     MODEL_PROVIDER_AZURE_SERVERLESS_OPENAI,
     DOT_ENV_FILENAME,
-    DEFAULT_VISION_MODEL,
     LARGE_MODEL_ID,
     SMALL_MODEL_ID,
-    DEFAULT_SMALL_MODEL_CANDIDATES,
-    DEFAULT_LARGE_MODEL_CANDIDATES,
-    DEFAULT_EMBEDDINGS_MODEL_CANDIDATES,
-    DEFAULT_VISION_MODEL_CANDIDATES,
-    DEFAULT_REASONING_MODEL,
-    DEFAULT_REASONING_SMALL_MODEL,
-    DEFAULT_REASONING_SMALL_MODEL_CANDIDATES,
-    DEFAULT_REASONING_MODEL_CANDIDATES,
+    VISION_MODEL_ID,
 } from "../../core/src/constants"
 import { tryReadText } from "../../core/src/fs"
 import {
@@ -54,7 +43,7 @@ import {
     ModelConfiguration,
 } from "../../core/src/host"
 import { TraceOptions } from "../../core/src/trace"
-import { logError, logVerbose } from "../../core/src/util"
+import { deleteEmptyValues, logError, logVerbose } from "../../core/src/util"
 import { parseModelIdentifier } from "../../core/src/models"
 import { LanguageModel } from "../../core/src/chat"
 import { errorMessage, NotSupportedError } from "../../core/src/error"
@@ -73,6 +62,7 @@ import { resolveGlobalConfiguration } from "../../core/src/config"
 import { HostConfiguration } from "../../core/src/hostconfiguration"
 import { resolveLanguageModel } from "../../core/src/lm"
 import { CancellationOptions } from "../../core/src/cancellation"
+import LLMS from "../../../packages/core/src/llms.json"
 
 class NodeServerManager implements ServerManager {
     async start(): Promise<void> {
@@ -83,6 +73,43 @@ class NodeServerManager implements ServerManager {
     }
 }
 
+function readModelAliases(): ModelConfigurations {
+    const aliases = [
+        LARGE_MODEL_ID,
+        SMALL_MODEL_ID,
+        VISION_MODEL_ID,
+        "embeddings",
+        "reasoning",
+        "reasoning_small",
+    ]
+    const res = {
+        ...(Object.fromEntries(
+            aliases.map((alias) => [alias, readModelAlias(alias)])
+        ) as ModelConfigurations),
+        ...Object.fromEntries(
+            Object.entries(LLMS.aliases).map((kv) => [
+                kv[0],
+                {
+                    model: kv[1],
+                    source: "default",
+                } satisfies ModelConfiguration,
+            ])
+        ),
+    }
+    return res
+
+    function readModelAlias(alias: string) {
+        const candidates = Object.values(LLMS.providers)
+            .map(({ aliases }) => (aliases as Record<string, string>)?.[alias])
+            .filter((c) => !!c)
+        return deleteEmptyValues({
+            model: candidates[0],
+            source: "default",
+            candidates,
+        })
+    }
+}
+
 export class NodeHost implements RuntimeHost {
     private pulledModels: string[] = []
     readonly dotEnvPath: string
@@ -97,41 +124,7 @@ export class NodeHost implements RuntimeHost {
         "default" | "cli" | "env" | "config",
         Omit<ModelConfigurations, "large" | "small" | "vision" | "embeddings">
     > = {
-        default: {
-            large: {
-                model: DEFAULT_LARGE_MODEL,
-                source: "default",
-                candidates: DEFAULT_LARGE_MODEL_CANDIDATES,
-            },
-            small: {
-                model: DEFAULT_SMALL_MODEL,
-                source: "default",
-                candidates: DEFAULT_SMALL_MODEL_CANDIDATES,
-            },
-            vision: {
-                model: DEFAULT_VISION_MODEL,
-                source: "default",
-                candidates: DEFAULT_VISION_MODEL_CANDIDATES,
-            },
-            embeddings: {
-                model: DEFAULT_EMBEDDINGS_MODEL,
-                source: "default",
-                candidates: DEFAULT_EMBEDDINGS_MODEL_CANDIDATES,
-            },
-            reasoning: {
-                model: DEFAULT_REASONING_MODEL,
-                source: "default",
-                candidates: DEFAULT_REASONING_MODEL_CANDIDATES,
-            },
-            ["reasoning_small"]: {
-                model: DEFAULT_REASONING_SMALL_MODEL,
-                source: "default",
-                candidates: DEFAULT_REASONING_SMALL_MODEL_CANDIDATES,
-            },
-            long: { model: LARGE_MODEL_ID, source: "default" },
-            agent: { model: LARGE_MODEL_ID, source: "default" },
-            memory: { model: SMALL_MODEL_ID, source: "default" },
-        },
+        default: readModelAliases(),
         cli: {},
         env: {},
         config: {},

diff --git a/packages/cli/src/parse.ts b/packages/cli/src/parse.ts
@@ -16,7 +16,6 @@ import { YAMLParse, YAMLStringify } from "../../core/src/yaml"
 import { resolveTokenEncoder } from "../../core/src/encoders"
 import {
     CSV_REGEX,
-    DEFAULT_LARGE_MODEL,
     INI_REGEX,
     JSON5_REGEX,
     MD_REGEX,
@@ -203,7 +202,7 @@ export async function parseTokens(
     filesGlobs: string[],
     options: { excludedFiles: string[]; model: string }
 ) {
-    const { model = DEFAULT_LARGE_MODEL } = options || {}
+    const { model } = options || {}
     const { encode: encoder } = await resolveTokenEncoder(model)
 
     const files = await expandFiles(filesGlobs, options?.excludedFiles)

diff --git a/packages/core/src/constants.ts b/packages/core/src/constants.ts
@@ -57,56 +57,6 @@ export const SMALL_MODEL_ID = "small"
 export const LARGE_MODEL_ID = "large"
 export const VISION_MODEL_ID = "vision"
 export const DEFAULT_FENCE_FORMAT: FenceFormat = "xml"
-export const DEFAULT_LARGE_MODEL = "openai:gpt-4o"
-export const DEFAULT_LARGE_MODEL_CANDIDATES = [
-    "azure_serverless:gpt-4o",
-    DEFAULT_LARGE_MODEL,
-    "google:gemini-1.5-pro-latest",
-    "anthropic:claude-2.1",
-    "mistral:mistral-large-latest",
-    "github:gpt-4o",
-    "client:gpt-4",
-]
-export const DEFAULT_VISION_MODEL = "openai:gpt-4o"
-export const DEFAULT_VISION_MODEL_CANDIDATES = [
-    "azure_serverless:gpt-4o",
-    DEFAULT_VISION_MODEL,
-    "google:gemini-1.5-flash-latest",
-    "anthropic:claude-2.1",
-    "github:gpt-4o",
-]
-export const DEFAULT_SMALL_MODEL = "openai:gpt-4o-mini"
-export const DEFAULT_SMALL_MODEL_CANDIDATES = [
-    "azure_serverless:gpt-4o-mini",
-    DEFAULT_SMALL_MODEL,
-    "google:gemini-1.5-flash-latest",
-    "anthropic:claude-instant-1.2",
-    "mistral:mistral-small-latest",
-    "github:gpt-4o-mini",
-    "client:gpt-4-mini",
-]
-export const DEFAULT_EMBEDDINGS_MODEL_CANDIDATES = [
-    "azure:text-embedding-3-small",
-    "azure:text-embedding-2-small",
-    "openai:text-embedding-3-small",
-    "github:text-embedding-3-small",
-    "client:text-embedding-3-small",
-]
-export const DEFAULT_REASONING_SMALL_MODEL = "openai:o1-mini"
-export const DEFAULT_REASONING_SMALL_MODEL_CANDIDATES = [
-    "azure_serverless:o1-mini",
-    DEFAULT_REASONING_SMALL_MODEL,
-    "github:o1-mini",
-    "client:o1-mini",
-]
-export const DEFAULT_REASONING_MODEL = "openai:o1"
-export const DEFAULT_REASONING_MODEL_CANDIDATES = [
-    "azure_serverless:o1-preview",
-    DEFAULT_REASONING_MODEL,
-    "github:o1-preview",
-    "client:o1-preview",
-]
-export const DEFAULT_EMBEDDINGS_MODEL = "openai:text-embedding-ada-002"
 export const DEFAULT_TEMPERATURE = 0.8
 export const BUILTIN_PREFIX = "_builtin/"
 export const CACHE_LLMREQUEST_PREFIX = "genaiscript/cache/llm/"