fix(fireworks): correct reasoning field for models and simplify to static catalog
- Set reasoning=true for gpt-oss models ("powerful reasoning, agentic tasks")
- Set reasoning=true for MiniMax-M2/M2.1 ("long-thinking")
- Set reasoning=true for GLM-4.6 ("advanced thinking controls")
- Set reasoning=false for DeepSeek V3.2 (standard chat model)
- Remove dynamic API fetching, use static catalog only
- Remove unused API constants and types
This commit is contained in:
Dylan Huang
parent
4bf61bc930
commit
a9c2ccdb3b
@ -1,8 +1,6 @@
|
||||
import type { ModelDefinitionConfig } from "../config/types.js";
|
||||
|
||||
export const FIREWORKS_BASE_URL = "https://api.fireworks.ai/inference/v1";
|
||||
export const FIREWORKS_API_BASE_URL = "https://api.fireworks.ai";
|
||||
export const FIREWORKS_ACCOUNT_ID = "fireworks";
|
||||
export const FIREWORKS_DEFAULT_MODEL_ID = "accounts/fireworks/models/deepseek-v3p2";
|
||||
export const FIREWORKS_DEFAULT_MODEL_REF = `fireworks/${FIREWORKS_DEFAULT_MODEL_ID}`;
|
||||
|
||||
@ -18,7 +16,6 @@ export const FIREWORKS_DEFAULT_COST = {
|
||||
/**
|
||||
* Static catalog of Fireworks AI serverless models.
|
||||
*
|
||||
* This catalog serves as a fallback when the Fireworks API is unreachable.
|
||||
* Only includes LLM models (no image generation), non-deprecated models,
|
||||
* and models that support serverless inference.
|
||||
*
|
||||
@ -61,17 +58,17 @@ export const FIREWORKS_MODEL_CATALOG = [
|
||||
{
|
||||
id: "accounts/fireworks/models/deepseek-v3p2",
|
||||
name: "DeepSeek V3.2",
|
||||
reasoning: true,
|
||||
reasoning: false,
|
||||
input: ["text"] as const,
|
||||
contextWindow: 163840,
|
||||
maxTokens: 8192,
|
||||
},
|
||||
|
||||
// GLM models
|
||||
// GLM models - both support "advanced thinking controls"
|
||||
{
|
||||
id: "accounts/fireworks/models/glm-4p6",
|
||||
name: "GLM-4.6",
|
||||
reasoning: false,
|
||||
reasoning: true,
|
||||
input: ["text"] as const,
|
||||
contextWindow: 202752,
|
||||
maxTokens: 8192,
|
||||
@ -85,11 +82,11 @@ export const FIREWORKS_MODEL_CATALOG = [
|
||||
maxTokens: 8192,
|
||||
},
|
||||
|
||||
// OpenAI gpt-oss models
|
||||
// OpenAI gpt-oss models - designed for "powerful reasoning, agentic tasks"
|
||||
{
|
||||
id: "accounts/fireworks/models/gpt-oss-120b",
|
||||
name: "OpenAI gpt-oss-120b",
|
||||
reasoning: false,
|
||||
reasoning: true,
|
||||
input: ["text"] as const,
|
||||
contextWindow: 131072,
|
||||
maxTokens: 8192,
|
||||
@ -97,7 +94,7 @@ export const FIREWORKS_MODEL_CATALOG = [
|
||||
{
|
||||
id: "accounts/fireworks/models/gpt-oss-20b",
|
||||
name: "OpenAI gpt-oss-20b",
|
||||
reasoning: false,
|
||||
reasoning: true,
|
||||
input: ["text"] as const,
|
||||
contextWindow: 131072,
|
||||
maxTokens: 8192,
|
||||
@ -132,11 +129,11 @@ export const FIREWORKS_MODEL_CATALOG = [
|
||||
maxTokens: 8192,
|
||||
},
|
||||
|
||||
// MiniMax models
|
||||
// MiniMax models - "long-thinking" and "complex reasoning tasks"
|
||||
{
|
||||
id: "accounts/fireworks/models/minimax-m2",
|
||||
name: "MiniMax-M2",
|
||||
reasoning: false,
|
||||
reasoning: true,
|
||||
input: ["text"] as const,
|
||||
contextWindow: 196608,
|
||||
maxTokens: 8192,
|
||||
@ -144,7 +141,7 @@ export const FIREWORKS_MODEL_CATALOG = [
|
||||
{
|
||||
id: "accounts/fireworks/models/minimax-m2p1",
|
||||
name: "MiniMax-M2.1",
|
||||
reasoning: false,
|
||||
reasoning: true,
|
||||
input: ["text"] as const,
|
||||
contextWindow: 204800,
|
||||
maxTokens: 8192,
|
||||
@ -244,130 +241,9 @@ export function buildFireworksModelDefinition(entry: FireworksCatalogEntry): Mod
|
||||
};
|
||||
}
|
||||
|
||||
// Fireworks API response types
|
||||
interface FireworksModel {
|
||||
name: string; // Full resource name: accounts/fireworks/models/<id>
|
||||
displayName?: string;
|
||||
description?: string;
|
||||
contextLength?: number;
|
||||
kind?: string; // HF_BASE_MODEL, FLUMINA_BASE_MODEL, etc.
|
||||
supportsImageInput?: boolean;
|
||||
supportsTools?: boolean;
|
||||
supportsServerless?: boolean;
|
||||
deprecationDate?: { year: number; month: number; day: number } | null;
|
||||
}
|
||||
|
||||
interface FireworksModelsResponse {
|
||||
models: FireworksModel[];
|
||||
nextPageToken?: string;
|
||||
}
|
||||
|
||||
function isDeprecated(
|
||||
deprecationDate?: { year: number; month: number; day: number } | null,
|
||||
): boolean {
|
||||
if (!deprecationDate) return false;
|
||||
const { year, month, day } = deprecationDate;
|
||||
const depDate = new Date(year, month - 1, day);
|
||||
return depDate < new Date();
|
||||
}
|
||||
|
||||
function isLlmModel(model: FireworksModel): boolean {
|
||||
// Skip image generation models (FLUX, etc.) - they use FLUMINA_BASE_MODEL kind
|
||||
// and have contextLength of 0
|
||||
if (model.kind === "FLUMINA_BASE_MODEL") return false;
|
||||
if (!model.contextLength || model.contextLength === 0) return false;
|
||||
return true;
|
||||
}
|
||||
|
||||
/**
|
||||
* Discover serverless LLM models from Fireworks API with fallback to static catalog.
|
||||
* Returns Fireworks models from the static catalog.
|
||||
*/
|
||||
export async function discoverFireworksModels(params?: {
|
||||
apiKey?: string;
|
||||
}): Promise<ModelDefinitionConfig[]> {
|
||||
// Skip API discovery in test environment
|
||||
if (process.env.NODE_ENV === "test" || process.env.VITEST) {
|
||||
return FIREWORKS_MODEL_CATALOG.map(buildFireworksModelDefinition);
|
||||
}
|
||||
|
||||
const apiKey = params?.apiKey ?? process.env.FIREWORKS_API_KEY;
|
||||
if (!apiKey) {
|
||||
console.warn("[fireworks-models] No API key available, using static catalog");
|
||||
return FIREWORKS_MODEL_CATALOG.map(buildFireworksModelDefinition);
|
||||
}
|
||||
|
||||
try {
|
||||
const allModels: FireworksModel[] = [];
|
||||
let pageToken: string | undefined;
|
||||
|
||||
// Paginate through results
|
||||
do {
|
||||
const url = new URL(`${FIREWORKS_API_BASE_URL}/v1/accounts/${FIREWORKS_ACCOUNT_ID}/models`);
|
||||
url.searchParams.set("filter", "supports_serverless=true");
|
||||
url.searchParams.set("pageSize", "200");
|
||||
if (pageToken) {
|
||||
url.searchParams.set("pageToken", pageToken);
|
||||
}
|
||||
|
||||
const response = await fetch(url.toString(), {
|
||||
headers: { Authorization: `Bearer ${apiKey}` },
|
||||
signal: AbortSignal.timeout(10000),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
console.warn(
|
||||
`[fireworks-models] Failed to list models: HTTP ${response.status}, using static catalog`,
|
||||
);
|
||||
return FIREWORKS_MODEL_CATALOG.map(buildFireworksModelDefinition);
|
||||
}
|
||||
|
||||
const data = (await response.json()) as FireworksModelsResponse;
|
||||
allModels.push(...(data.models ?? []));
|
||||
pageToken = data.nextPageToken;
|
||||
} while (pageToken);
|
||||
|
||||
// Filter and build model definitions
|
||||
const catalogById = new Map<string, FireworksCatalogEntry>(
|
||||
FIREWORKS_MODEL_CATALOG.map((m) => [m.id, m]),
|
||||
);
|
||||
const models: ModelDefinitionConfig[] = [];
|
||||
|
||||
for (const apiModel of allModels) {
|
||||
// Skip deprecated models
|
||||
if (isDeprecated(apiModel.deprecationDate)) continue;
|
||||
// Skip non-LLM models (image gen, etc.)
|
||||
if (!isLlmModel(apiModel)) continue;
|
||||
|
||||
// Use the full name as the ID (accounts/fireworks/models/<model>)
|
||||
const id = apiModel.name;
|
||||
const catalogEntry = catalogById.get(id);
|
||||
|
||||
if (catalogEntry) {
|
||||
// Use catalog metadata for known models
|
||||
models.push(buildFireworksModelDefinition(catalogEntry));
|
||||
} else {
|
||||
// Create definition for newly discovered models not in catalog
|
||||
const isReasoning =
|
||||
id.toLowerCase().includes("r1") ||
|
||||
id.toLowerCase().includes("thinking") ||
|
||||
id.toLowerCase().includes("reasoning") ||
|
||||
(apiModel.description?.toLowerCase().includes("reasoning") ?? false);
|
||||
|
||||
models.push({
|
||||
id,
|
||||
name: apiModel.displayName ?? id.replace(/^accounts\/fireworks\/models\//, ""),
|
||||
reasoning: isReasoning,
|
||||
input: apiModel.supportsImageInput ? ["text", "image"] : ["text"],
|
||||
cost: FIREWORKS_DEFAULT_COST,
|
||||
contextWindow: apiModel.contextLength ?? 128000,
|
||||
maxTokens: 8192,
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
return models.length > 0 ? models : FIREWORKS_MODEL_CATALOG.map(buildFireworksModelDefinition);
|
||||
} catch (error) {
|
||||
console.warn(`[fireworks-models] Discovery failed: ${String(error)}, using static catalog`);
|
||||
return FIREWORKS_MODEL_CATALOG.map(buildFireworksModelDefinition);
|
||||
}
|
||||
export function discoverFireworksModels(): ModelDefinitionConfig[] {
|
||||
return FIREWORKS_MODEL_CATALOG.map(buildFireworksModelDefinition);
|
||||
}
|
||||
|
||||
@ -351,8 +351,8 @@ async function buildVeniceProvider(): Promise<ProviderConfig> {
|
||||
};
|
||||
}
|
||||
|
||||
async function buildFireworksProvider(apiKey?: string): Promise<ProviderConfig> {
|
||||
const models = await discoverFireworksModels({ apiKey });
|
||||
function buildFireworksProvider(): ProviderConfig {
|
||||
const models = discoverFireworksModels();
|
||||
return {
|
||||
baseUrl: FIREWORKS_BASE_URL,
|
||||
api: "openai-completions",
|
||||
@ -416,7 +416,7 @@ export async function resolveImplicitProviders(params: {
|
||||
resolveEnvApiKeyVarName("fireworks") ??
|
||||
resolveApiKeyFromProfiles({ provider: "fireworks", store: authStore });
|
||||
if (fireworksKey) {
|
||||
providers.fireworks = { ...(await buildFireworksProvider(fireworksKey)), apiKey: fireworksKey };
|
||||
providers.fireworks = { ...buildFireworksProvider(), apiKey: fireworksKey };
|
||||
}
|
||||
|
||||
const qwenProfiles = listProfilesForProvider(authStore, "qwen-portal");
|
||||
|
||||
Loading…
Reference in New Issue
Block a user