mirror of
https://github.com/openclaw/openclaw.git
synced 2026-05-06 05:30:42 +00:00
fix(deepseek): add provider-policy-api to hydrate contextWindow and cost from catalog (#74326)
DeepSeek models had no provider-policy-api.ts, so materializeRuntimeConfig filled contextWindow with DEFAULT_CONTEXT_TOKENS (200k) and cost with zeros for all DeepSeek models. This caused premature session compaction at ~125k instead of using the full 1M window, and zero-cost display for v4 models. Add a normalizeConfig surface that hydrates missing contextWindow, maxTokens, and cost from the bundled DeepSeek model catalog for matching model ids. Explicit user overrides are preserved. Fixes #74245
This commit is contained in:
Alex Knight
235
extensions/deepseek/provider-policy-api.test.ts
Normal file
235
extensions/deepseek/provider-policy-api.test.ts
Normal file
@@ -0,0 +1,235 @@
|
||||
import type { ModelProviderConfig } from "openclaw/plugin-sdk/provider-model-types";
|
||||
import { describe, expect, it } from "vitest";
|
||||
import { normalizeConfig } from "./provider-policy-api.js";
|
||||
|
||||
describe("deepseek provider-policy-api", () => {
|
||||
it("hydrates contextWindow and cost from catalog for known models", () => {
|
||||
const providerConfig: ModelProviderConfig = {
|
||||
baseUrl: "https://api.deepseek.com",
|
||||
api: "openai-completions",
|
||||
models: [
|
||||
{
|
||||
id: "deepseek-v4-flash",
|
||||
name: "DeepSeek V4 Flash",
|
||||
reasoning: true,
|
||||
input: ["text"],
|
||||
} as never,
|
||||
],
|
||||
};
|
||||
|
||||
const result = normalizeConfig({ provider: "deepseek", providerConfig });
|
||||
|
||||
expect(result).not.toBe(providerConfig);
|
||||
const model = result.models[0];
|
||||
expect(model.contextWindow).toBe(1_000_000);
|
||||
expect(model.maxTokens).toBe(384_000);
|
||||
expect(model.cost).toEqual({
|
||||
input: 0.14,
|
||||
output: 0.28,
|
||||
cacheRead: 0.028,
|
||||
cacheWrite: 0,
|
||||
});
|
||||
});
|
||||
|
||||
it("hydrates deepseek-v4-pro with correct metadata", () => {
|
||||
const providerConfig: ModelProviderConfig = {
|
||||
baseUrl: "https://api.deepseek.com",
|
||||
api: "openai-completions",
|
||||
models: [
|
||||
{
|
||||
id: "deepseek-v4-pro",
|
||||
name: "DeepSeek V4 Pro",
|
||||
reasoning: true,
|
||||
input: ["text"],
|
||||
} as never,
|
||||
],
|
||||
};
|
||||
|
||||
const result = normalizeConfig({ provider: "deepseek", providerConfig });
|
||||
const model = result.models[0];
|
||||
expect(model.contextWindow).toBe(1_000_000);
|
||||
expect(model.maxTokens).toBe(384_000);
|
||||
expect(model.cost).toEqual({
|
||||
input: 1.74,
|
||||
output: 3.48,
|
||||
cacheRead: 0.145,
|
||||
cacheWrite: 0,
|
||||
});
|
||||
});
|
||||
|
||||
it("hydrates deepseek-chat with 131k context", () => {
|
||||
const providerConfig: ModelProviderConfig = {
|
||||
baseUrl: "https://api.deepseek.com",
|
||||
api: "openai-completions",
|
||||
models: [
|
||||
{
|
||||
id: "deepseek-chat",
|
||||
name: "DeepSeek Chat",
|
||||
reasoning: false,
|
||||
input: ["text"],
|
||||
} as never,
|
||||
],
|
||||
};
|
||||
|
||||
const result = normalizeConfig({ provider: "deepseek", providerConfig });
|
||||
const model = result.models[0];
|
||||
expect(model.contextWindow).toBe(131_072);
|
||||
});
|
||||
|
||||
it("preserves explicit user contextWindow override", () => {
|
||||
const providerConfig: ModelProviderConfig = {
|
||||
baseUrl: "https://api.deepseek.com",
|
||||
api: "openai-completions",
|
||||
models: [
|
||||
{
|
||||
id: "deepseek-v4-flash",
|
||||
name: "DeepSeek V4 Flash",
|
||||
reasoning: true,
|
||||
input: ["text"],
|
||||
contextWindow: 500_000,
|
||||
} as never,
|
||||
],
|
||||
};
|
||||
|
||||
const result = normalizeConfig({ provider: "deepseek", providerConfig });
|
||||
const model = result.models[0];
|
||||
expect(model.contextWindow).toBe(500_000);
|
||||
// cost should still be hydrated since it was missing
|
||||
expect(model.cost).toEqual({
|
||||
input: 0.14,
|
||||
output: 0.28,
|
||||
cacheRead: 0.028,
|
||||
cacheWrite: 0,
|
||||
});
|
||||
});
|
||||
|
||||
it("preserves explicit user cost override", () => {
|
||||
const userCost = { input: 99, output: 99, cacheRead: 99, cacheWrite: 99 };
|
||||
const providerConfig: ModelProviderConfig = {
|
||||
baseUrl: "https://api.deepseek.com",
|
||||
api: "openai-completions",
|
||||
models: [
|
||||
{
|
||||
id: "deepseek-v4-flash",
|
||||
name: "DeepSeek V4 Flash",
|
||||
reasoning: true,
|
||||
input: ["text"],
|
||||
cost: userCost,
|
||||
} as never,
|
||||
],
|
||||
};
|
||||
|
||||
const result = normalizeConfig({ provider: "deepseek", providerConfig });
|
||||
const model = result.models[0];
|
||||
expect(model.cost).toEqual(userCost);
|
||||
// contextWindow should still be hydrated since it was missing
|
||||
expect(model.contextWindow).toBe(1_000_000);
|
||||
});
|
||||
|
||||
it("preserves explicit user maxTokens override", () => {
|
||||
const providerConfig: ModelProviderConfig = {
|
||||
baseUrl: "https://api.deepseek.com",
|
||||
api: "openai-completions",
|
||||
models: [
|
||||
{
|
||||
id: "deepseek-v4-flash",
|
||||
name: "DeepSeek V4 Flash",
|
||||
reasoning: true,
|
||||
input: ["text"],
|
||||
maxTokens: 100_000,
|
||||
} as never,
|
||||
],
|
||||
};
|
||||
|
||||
const result = normalizeConfig({ provider: "deepseek", providerConfig });
|
||||
const model = result.models[0];
|
||||
expect(model.maxTokens).toBe(100_000);
|
||||
});
|
||||
|
||||
it("returns providerConfig unchanged when all models already have metadata", () => {
|
||||
const providerConfig: ModelProviderConfig = {
|
||||
baseUrl: "https://api.deepseek.com",
|
||||
api: "openai-completions",
|
||||
models: [
|
||||
{
|
||||
id: "deepseek-v4-flash",
|
||||
name: "DeepSeek V4 Flash",
|
||||
reasoning: true,
|
||||
input: ["text"],
|
||||
contextWindow: 1_000_000,
|
||||
maxTokens: 384_000,
|
||||
cost: { input: 0.14, output: 0.28, cacheRead: 0.028, cacheWrite: 0 },
|
||||
} as never,
|
||||
],
|
||||
};
|
||||
|
||||
const result = normalizeConfig({ provider: "deepseek", providerConfig });
|
||||
expect(result).toBe(providerConfig);
|
||||
});
|
||||
|
||||
it("passes through unknown model ids unchanged", () => {
|
||||
const providerConfig: ModelProviderConfig = {
|
||||
baseUrl: "https://api.deepseek.com",
|
||||
api: "openai-completions",
|
||||
models: [
|
||||
{
|
||||
id: "deepseek-custom-finetune",
|
||||
name: "Custom Fine-tune",
|
||||
reasoning: false,
|
||||
input: ["text"],
|
||||
} as never,
|
||||
],
|
||||
};
|
||||
|
||||
const result = normalizeConfig({ provider: "deepseek", providerConfig });
|
||||
expect(result).toBe(providerConfig);
|
||||
});
|
||||
|
||||
it("returns providerConfig unchanged when models array is empty", () => {
|
||||
const providerConfig: ModelProviderConfig = {
|
||||
baseUrl: "https://api.deepseek.com",
|
||||
api: "openai-completions",
|
||||
models: [],
|
||||
};
|
||||
|
||||
const result = normalizeConfig({ provider: "deepseek", providerConfig });
|
||||
expect(result).toBe(providerConfig);
|
||||
});
|
||||
|
||||
it("hydrates only the models that need it in a mixed list", () => {
|
||||
const providerConfig: ModelProviderConfig = {
|
||||
baseUrl: "https://api.deepseek.com",
|
||||
api: "openai-completions",
|
||||
models: [
|
||||
{
|
||||
id: "deepseek-v4-flash",
|
||||
name: "DeepSeek V4 Flash",
|
||||
reasoning: true,
|
||||
input: ["text"],
|
||||
contextWindow: 1_000_000,
|
||||
maxTokens: 384_000,
|
||||
cost: { input: 0.14, output: 0.28, cacheRead: 0.028, cacheWrite: 0 },
|
||||
} as never,
|
||||
{
|
||||
id: "deepseek-v4-pro",
|
||||
name: "DeepSeek V4 Pro",
|
||||
reasoning: true,
|
||||
input: ["text"],
|
||||
} as never,
|
||||
],
|
||||
};
|
||||
|
||||
const result = normalizeConfig({ provider: "deepseek", providerConfig });
|
||||
expect(result).not.toBe(providerConfig);
|
||||
// First model should be unchanged (same reference)
|
||||
expect(result.models[0]).toBe(providerConfig.models[0]);
|
||||
// Second model should be hydrated
|
||||
expect(result.models[1].contextWindow).toBe(1_000_000);
|
||||
expect(result.models[1].cost).toEqual({
|
||||
input: 1.74,
|
||||
output: 3.48,
|
||||
cacheRead: 0.145,
|
||||
cacheWrite: 0,
|
||||
});
|
||||
});
|
||||
});
|
||||
97
extensions/deepseek/provider-policy-api.ts
Normal file
97
extensions/deepseek/provider-policy-api.ts
Normal file
@@ -0,0 +1,97 @@
|
||||
import type { ModelDefinitionConfig } from "openclaw/plugin-sdk/provider-model-shared";
|
||||
import type { ModelProviderConfig } from "openclaw/plugin-sdk/provider-model-types";
|
||||
import { DEEPSEEK_MODEL_CATALOG } from "./models.js";
|
||||
|
||||
type ModelDefinitionDraft = Partial<ModelDefinitionConfig> &
|
||||
Pick<ModelDefinitionConfig, "id" | "name">;
|
||||
|
||||
/**
|
||||
* Build a lookup from the bundled DeepSeek model catalog so we can hydrate
|
||||
* missing metadata (contextWindow, cost, maxTokens) into user-configured
|
||||
* model rows without overwriting explicit overrides.
|
||||
*/
|
||||
function buildCatalogIndex(): Map<string, ModelDefinitionConfig> {
|
||||
const index = new Map<string, ModelDefinitionConfig>();
|
||||
for (const model of DEEPSEEK_MODEL_CATALOG) {
|
||||
index.set(model.id, model);
|
||||
}
|
||||
return index;
|
||||
}
|
||||
|
||||
function isPositiveNumber(value: unknown): value is number {
|
||||
return typeof value === "number" && Number.isFinite(value) && value > 0;
|
||||
}
|
||||
|
||||
function hasCostValues(cost: unknown): cost is ModelDefinitionConfig["cost"] {
|
||||
if (!cost || typeof cost !== "object") {
|
||||
return false;
|
||||
}
|
||||
const c = cost as Record<string, unknown>;
|
||||
return (
|
||||
typeof c.input === "number" ||
|
||||
typeof c.output === "number" ||
|
||||
typeof c.cacheRead === "number" ||
|
||||
typeof c.cacheWrite === "number"
|
||||
);
|
||||
}
|
||||
|
||||
/**
|
||||
* Provider policy surface for DeepSeek.
|
||||
*
|
||||
* Hydrates missing `contextWindow`, `cost`, and `maxTokens` from the bundled
|
||||
* catalog for matching model ids. Explicit user overrides are preserved.
|
||||
*/
|
||||
export function normalizeConfig(params: {
|
||||
provider: string;
|
||||
providerConfig: ModelProviderConfig;
|
||||
}): ModelProviderConfig {
|
||||
const { providerConfig } = params;
|
||||
if (!Array.isArray(providerConfig.models) || providerConfig.models.length === 0) {
|
||||
return providerConfig;
|
||||
}
|
||||
|
||||
const catalog = buildCatalogIndex();
|
||||
let mutated = false;
|
||||
|
||||
const nextModels = providerConfig.models.map((model) => {
|
||||
const raw = model as ModelDefinitionDraft;
|
||||
const catalogEntry = catalog.get(raw.id);
|
||||
if (!catalogEntry) {
|
||||
return model;
|
||||
}
|
||||
|
||||
let modelMutated = false;
|
||||
const patched: Record<string, unknown> = {};
|
||||
|
||||
// Hydrate contextWindow from catalog when missing or not a positive number.
|
||||
if (!isPositiveNumber(raw.contextWindow) && isPositiveNumber(catalogEntry.contextWindow)) {
|
||||
patched.contextWindow = catalogEntry.contextWindow;
|
||||
modelMutated = true;
|
||||
}
|
||||
|
||||
// Hydrate maxTokens from catalog when missing or not a positive number.
|
||||
if (!isPositiveNumber(raw.maxTokens) && isPositiveNumber(catalogEntry.maxTokens)) {
|
||||
patched.maxTokens = catalogEntry.maxTokens;
|
||||
modelMutated = true;
|
||||
}
|
||||
|
||||
// Hydrate cost from catalog when missing or when all fields are zero/absent.
|
||||
if (!hasCostValues(raw.cost) && hasCostValues(catalogEntry.cost)) {
|
||||
patched.cost = catalogEntry.cost;
|
||||
modelMutated = true;
|
||||
}
|
||||
|
||||
if (!modelMutated) {
|
||||
return model;
|
||||
}
|
||||
|
||||
mutated = true;
|
||||
return { ...raw, ...patched };
|
||||
});
|
||||
|
||||
if (!mutated) {
|
||||
return providerConfig;
|
||||
}
|
||||
|
||||
return { ...providerConfig, models: nextModels as ModelDefinitionConfig[] };
|
||||
}
|
||||
Reference in New Issue
Block a user