@aliou/pi-neuralwatt 0.15.0 → 0.15.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
|
@@ -53,6 +53,19 @@ const GLM_5_2: NeuralwattModelFamily = {
|
|
|
53
53
|
},
|
|
54
54
|
};
|
|
55
55
|
|
|
56
|
+
// ZhipuAI. GLM-5.3 ships as a GLM-5.2 weight swap in gated preview, with
|
|
57
|
+
// GLM-5.2 pricing parity (per the API metadata; review at launch). Unlike
|
|
58
|
+
// 5.2, reasoning is mandatory and `none` is not offered: efforts are
|
|
59
|
+
// max/high/low (default max).
|
|
60
|
+
const GLM_5_3: NeuralwattModelFamily = {
|
|
61
|
+
cost: { input: 1.45, output: 4.5, cacheRead: 0.145 },
|
|
62
|
+
vision: false,
|
|
63
|
+
reasoningMetadata: {
|
|
64
|
+
supported_efforts: ["max", "high", "low"],
|
|
65
|
+
mandatory: true,
|
|
66
|
+
},
|
|
67
|
+
};
|
|
68
|
+
|
|
56
69
|
// MoonshotAI. K3 supports reasoning efforts low/high/max (default max) and
|
|
57
70
|
// can be turned off (`mandatory: false`). The `-fast` endpoint is a shorthand
|
|
58
71
|
// to set thinking to off.
|
|
@@ -65,12 +78,15 @@ const KIMI_K3: NeuralwattModelFamily = {
|
|
|
65
78
|
},
|
|
66
79
|
};
|
|
67
80
|
|
|
68
|
-
// MoonshotAI.
|
|
69
|
-
//
|
|
70
|
-
// map with `off: null`, matching the upstream binary thinking toggle.
|
|
81
|
+
// MoonshotAI. K2.7 Code has mandatory reasoning with no selectable efforts
|
|
82
|
+
// (`supported_efforts: []`), so `buildThinkingLevelMap` nulls out every level.
|
|
71
83
|
const KIMI_K2_7_CODE: NeuralwattModelFamily = {
|
|
72
84
|
cost: { input: 0.95, output: 4.0, cacheRead: 0.095 },
|
|
73
85
|
vision: true,
|
|
86
|
+
reasoningMetadata: {
|
|
87
|
+
supported_efforts: [],
|
|
88
|
+
mandatory: true,
|
|
89
|
+
},
|
|
74
90
|
};
|
|
75
91
|
|
|
76
92
|
// Qwen. Qwen3.6 35B only advertises `high` and `none`.
|
|
@@ -180,6 +196,18 @@ const FAMILIES: [NeuralwattModelFamily, NeuralwattVariantSpec[]][] = [
|
|
|
180
196
|
},
|
|
181
197
|
],
|
|
182
198
|
],
|
|
199
|
+
[
|
|
200
|
+
GLM_5_3,
|
|
201
|
+
[
|
|
202
|
+
{
|
|
203
|
+
id: "glm-5.3",
|
|
204
|
+
name: "GLM-5.3",
|
|
205
|
+
contextWindow: 1048560,
|
|
206
|
+
maxOutputTokens: null,
|
|
207
|
+
reasoning: true,
|
|
208
|
+
},
|
|
209
|
+
],
|
|
210
|
+
],
|
|
183
211
|
// The kimi-k3 endpoint rejects anything above 327,680 total tokens with
|
|
184
212
|
// `400: max_completion_tokens is too large … supports at most 327680
|
|
185
213
|
// completion tokens` (verified at runtime), even though the API advertises
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@aliou/pi-neuralwatt",
|
|
3
|
-
"version": "0.15.
|
|
3
|
+
"version": "0.15.2",
|
|
4
4
|
"license": "MIT",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"private": false,
|
|
@@ -52,6 +52,7 @@
|
|
|
52
52
|
"@types/node": "^25.0.10",
|
|
53
53
|
"husky": "^9.1.7",
|
|
54
54
|
"ts-json-schema-generator": "^2.4.0",
|
|
55
|
+
"tsx": "4.23.12",
|
|
55
56
|
"typescript": "^5.9.3",
|
|
56
57
|
"vitest": "^4.0.18"
|
|
57
58
|
},
|
|
@@ -67,6 +68,7 @@
|
|
|
67
68
|
"format": "biome check --write",
|
|
68
69
|
"test": "vitest run",
|
|
69
70
|
"test:watch": "vitest",
|
|
71
|
+
"check:models": "tsx scripts/check-models.ts",
|
|
70
72
|
"gen:schema": "pi-settings-schema -p src/config/types.ts -t NeuralwattConfig -o schema.json --version 0.12.0",
|
|
71
73
|
"check:schema": "pi-settings-schema -p src/config/types.ts -t NeuralwattConfig -o schema.json --version 0.12.0 --check",
|
|
72
74
|
"check:lockfile": "pnpm install --frozen-lockfile --ignore-scripts",
|
|
@@ -1,7 +1,6 @@
|
|
|
1
1
|
import type { Migration } from "@aliou/pi-utils-settings";
|
|
2
2
|
import type { NeuralwattConfig } from "../types";
|
|
3
3
|
|
|
4
|
-
export { disableLegacyModelIdsByDefaultMigration } from "./01-disable-legacy-model-ids-by-default";
|
|
5
4
|
export {
|
|
6
5
|
backupConfig,
|
|
7
6
|
flatToNestedConfigMigration,
|
|
@@ -9,7 +8,6 @@ export {
|
|
|
9
8
|
export { renameHiddenToEarlyAccessMigration } from "./03-rename-hidden-to-early-access";
|
|
10
9
|
export { enableAliasesForLegacyUsersMigration } from "./04-enable-aliases-for-legacy-users";
|
|
11
10
|
|
|
12
|
-
import { disableLegacyModelIdsByDefaultMigration } from "./01-disable-legacy-model-ids-by-default";
|
|
13
11
|
import { flatToNestedConfigMigration } from "./02-flat-to-nested-config";
|
|
14
12
|
import { renameHiddenToEarlyAccessMigration } from "./03-rename-hidden-to-early-access";
|
|
15
13
|
import { enableAliasesForLegacyUsersMigration } from "./04-enable-aliases-for-legacy-users";
|
|
@@ -18,7 +16,6 @@ import { enableAliasesForLegacyUsersMigration } from "./04-enable-aliases-for-le
|
|
|
18
16
|
// applies them in sequence on the raw config record, so they are cast to the
|
|
19
17
|
// current config type for the array.
|
|
20
18
|
export const migrations = [
|
|
21
|
-
disableLegacyModelIdsByDefaultMigration,
|
|
22
19
|
flatToNestedConfigMigration,
|
|
23
20
|
renameHiddenToEarlyAccessMigration,
|
|
24
21
|
enableAliasesForLegacyUsersMigration,
|
|
@@ -1,30 +0,0 @@
|
|
|
1
|
-
import type { Migration } from "@aliou/pi-utils-settings";
|
|
2
|
-
|
|
3
|
-
/** Flat config shape before the 0.8.x nested migration. */
|
|
4
|
-
interface FlatNeuralwattConfig {
|
|
5
|
-
$schema?: string;
|
|
6
|
-
quotaCommand?: boolean;
|
|
7
|
-
quotaWarnings?: boolean;
|
|
8
|
-
subBarIntegration?: boolean;
|
|
9
|
-
includeLegacyModelIds?: boolean;
|
|
10
|
-
includeHiddenModels?: boolean;
|
|
11
|
-
}
|
|
12
|
-
|
|
13
|
-
function isPreviousConfigWithoutLegacyDefault(
|
|
14
|
-
config: FlatNeuralwattConfig,
|
|
15
|
-
): boolean {
|
|
16
|
-
return config.includeLegacyModelIds === undefined;
|
|
17
|
-
}
|
|
18
|
-
|
|
19
|
-
export const disableLegacyModelIdsByDefaultMigration: Migration<FlatNeuralwattConfig> =
|
|
20
|
-
{
|
|
21
|
-
name: "disable-legacy-model-ids-by-default",
|
|
22
|
-
version: "0.8.0",
|
|
23
|
-
shouldRun: isPreviousConfigWithoutLegacyDefault,
|
|
24
|
-
message:
|
|
25
|
-
"[neuralwatt] legacy model IDs (ids including the provider and the quantization) are disabled by default. You can enable them with /neuralwatt:settings.",
|
|
26
|
-
run: (config) => ({
|
|
27
|
-
...config,
|
|
28
|
-
includeLegacyModelIds: false,
|
|
29
|
-
}),
|
|
30
|
-
};
|