Normalize base-model inheritance

2026-06-04 12:52:03 -05:00
163 changed files with 53 additions and 264 deletions
@@ -12,7 +12,7 @@ import {
 } from "./schema.js";

 const BaseModel = AuthoredModelShape
-  .partial()
+  .deepPartial()
  .extend({
    id: z.string(),
    base_model: z.string().min(1, "Base model cannot be empty"),
@@ -211,7 +211,8 @@ function applyOmit(target: Record<string, unknown>, paths: string[]) {

    for (let index = parents.length - 1; index >= 0; index--) {
      const parent = parents[index];
-      const value = parent?.value[parent.key];
+      if (parent === undefined) continue;
+      const value = parent.value[parent.key];
      if (
        value === null ||
        value === undefined ||
@@ -10,14 +10,21 @@ import { openrouter } from "./providers/openrouter.js";
 import { ovhcloud } from "./providers/ovhcloud.js";
 import { xai } from "./providers/xai.js";

-const ExistingModel = AuthoredModelShape.partial()
+const ExistingModelType = AuthoredModelShape.partial()
  .extend({
    base_model: z.string().optional(),
    base_model_omit: z.array(z.string()).optional(),
  })
  .strict();

-const SyncedBaseModel = AuthoredModelShape.partial()
+const ExistingModel = AuthoredModelShape.deepPartial()
+  .extend({
+    base_model: z.string().optional(),
+    base_model_omit: z.array(z.string()).optional(),
+  })
+  .strict();
+
+const SyncedBaseModel = AuthoredModelShape.deepPartial()
  .extend({
    id: z.string(),
    base_model: z.string(),
@@ -27,7 +34,7 @@ const SyncedBaseModel = AuthoredModelShape.partial()

 const SyncedAuthoredModel = z.union([AuthoredModel, SyncedBaseModel]);

-export type ExistingModel = z.infer<typeof ExistingModel>;
+export type ExistingModel = z.infer<typeof ExistingModelType>;
 export type SyncedFullModel = Omit<z.infer<typeof AuthoredModelShape>, "id">;
 export type SyncedBaseModel = Omit<z.infer<typeof SyncedBaseModel>, "id">;
 export type SyncedModel = SyncedFullModel | SyncedBaseModel;
@@ -240,7 +247,7 @@ async function readExisting(modelsDir: string) {
      throw parsed.error;
    }

-    const authored = parsed.data;
+    const authored = parsed.data as ExistingModel;
    if (authored.base_model !== undefined && modelMetadata === undefined) {
      modelMetadata = await readModelMetadata(modelsDir);
    }
@@ -304,7 +311,7 @@ function resolveBaseModel(
    parsed.error.cause = { modelPath, toml: merged };
    throw parsed.error;
  }
-  return parsed.data;
+  return parsed.data as ExistingModel;
 }

 function inheritableModelMetadata(model: Record<string, unknown>) {
@@ -128,6 +128,44 @@ output = 32_000
    });
  });

+  test("base_model can inherit sibling fields from partial object overrides", async () => {
+    await withFixture(async (root) => {
+      await write(root, "providers/provider/provider.toml", providerToml("Provider"));
+      await write(root, "models/lab/model.toml", modelMetadataToml());
+      await write(
+        root,
+        "providers/provider/models/model.toml",
+        `base_model = "lab/model"
+open_weights = true
+
+[cost]
+input = 1.25
+output = 2.50
+
+[limit]
+context = 200_000
+
+[modalities]
+input = ["text"]
+`,
+      );
+
+      const providers = await generate(path.join(root, "providers"));
+      const model = providers.provider?.models.model;
+
+      expect(model?.open_weights).toBe(true);
+      expect(model?.limit).toEqual({
+        context: 200_000,
+        input: 272_000,
+        output: 128_000,
+      });
+      expect(model?.modalities).toEqual({
+        input: ["text"],
+        output: ["text"],
+      });
+    });
+  });
+
  test("repository provider TOMLs do not use legacy extends tables", async () => {
    const root = path.join(import.meta.dirname, "..", "..", "..");
    const matches: string[] = [];
@@ -14,5 +14,4 @@ cache_read = 0
 cache_write = 0

 [limit]
-context = 262_144
 output = 32_768
@@ -12,5 +12,4 @@ cache_read = 0
 cache_write = 0

 [limit]
-context = 262_144
 output = 16_384
@@ -11,4 +11,3 @@ cache_write = 0

 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -11,4 +11,3 @@ cache_write = 0

 [limit]
 context = 202_752
-output = 131_072
@@ -8,4 +8,3 @@ cache_read = 0.075

 [modalities]
 input = ["text", "image", "pdf"]
-output = ["text"]
@@ -8,4 +8,3 @@ cache_read = 0.02

 [modalities]
 input = ["text", "image", "pdf"]
-output = ["text"]
@@ -8,4 +8,3 @@ cache_read = 0.1

 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -8,4 +8,3 @@ cache_read = 0.5

 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -8,4 +8,3 @@ cache_read = 0.075

 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -8,4 +8,3 @@ cache_read = 1.25

 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -4,7 +4,3 @@ base_model_omit = ["limit.input"]
 [cost]
 input = 15
 output = 120
-
-[limit]
-context = 400_000
-output = 272_000
@@ -5,7 +5,3 @@ base_model_omit = ["limit.input"]
 input = 1.25
 output = 10
 cache_read = 0.125
-
-[limit]
-context = 400_000
-output = 128_000
@@ -8,4 +8,3 @@ cache_read = 0.075

 [modalities]
 input = ["text", "image", "pdf"]
-output = ["text"]
@@ -8,4 +8,3 @@ cache_read = 0.02

 [modalities]
 input = ["text", "image", "pdf"]
-output = ["text"]
@@ -8,4 +8,3 @@ cache_read = 0.5

 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -11,10 +11,5 @@ input = 0.39
 output = 2.34
 cache_read = 0.195

-[limit]
-context = 262_144
-output = 65_536
-
 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -8,10 +8,5 @@ input = 0.195
 output = 1.56
 cache_read = 0.0975

-[limit]
-context = 262_144
-output = 65_536
-
 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -9,7 +9,3 @@ structured_output = true
 input = 0.09
 output = 0.29
 cache_read = 0.045
-
-[limit]
-context = 262_144
-output = 65_536
@@ -3,7 +3,6 @@
 base_model = "deepseek/deepseek-r1"
 name = "DeepSeek R1 0528 TEE"
 structured_output = true
-open_weights = true

 [interleaved]
 field = "reasoning_content"
@@ -15,5 +15,4 @@ output = 2
 cache_read = 0.22

 [limit]
-context = 262_144
 output = 65_535
@@ -13,5 +13,4 @@ output = 4
 cache_read = 0.475

 [limit]
-context = 262_144
 output = 65_535
@@ -2,7 +2,6 @@
 # Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 base_model = "zai/glm-5.1"
 name = "GLM 5.1 TEE"
-open_weights = true

 [interleaved]
 field = "reasoning_content"
@@ -9,4 +9,3 @@ output = 4

 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -11,4 +11,3 @@ cache_read = 0.0028

 [limit]
 context = 1_048_576
-output = 384_000
@@ -11,4 +11,3 @@ cache_read = 0.003625

 [limit]
 context = 1_048_576
-output = 384_000
@@ -11,9 +11,7 @@ input = 0.444
 output = 3.106

 [limit]
-context = 262_144
 output = 262_144

 [modalities]
 input = ["text"]
-output = ["text"]
@@ -16,4 +16,3 @@ output = 250_000

 [modalities]
 input = ["text"]
-output = ["text"]
@@ -9,7 +9,6 @@ output = 0.21
 cache_read = 0.003

 [limit]
-context = 1_000_000
 output = 131_072

 [provider]
@@ -9,7 +9,6 @@ output = 1.6
 cache_read = 0.02

 [limit]
-context = 1_000_000
 output = 131_072

 [provider]
@@ -9,7 +9,6 @@ output = 0.8
 cache_read = 0.003

 [limit]
-context = 1_000_000
 output = 131_072

 [provider]
@@ -6,7 +6,6 @@ output = 0.3
 cache_read = 0.02

 [limit]
-context = 262_144
 output = 262_144

 [provider]
@@ -8,7 +8,6 @@ cache_write = 0

 [limit]
 context = 202_752
-output = 131_072

 [provider]
 npm = "@ai-sdk/openai-compatible"
@@ -6,7 +6,6 @@ output = 1.75
 cache_read = 0.07

 [limit]
-context = 262_144
 output = 262_144

 [provider]
@@ -6,7 +6,6 @@ output = 1.5
 cache_read = 0.04

 [limit]
-context = 262_144
 output = 262_144

 [provider]
@@ -15,5 +15,4 @@ output = 6
 cache_read = 0.4

 [limit]
-context = 1_048_576
 output = 16_384
@@ -8,5 +8,4 @@ input = 1.25
 output = 3.89

 [limit]
-context = 200_000
 output = 128_000
@@ -8,4 +8,3 @@ cache_read = 0.5

 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -7,6 +7,4 @@ cache_read = 0.1
 cache_write = 1.25

 [limit]
-context = 200_000
 input = 136_000
-output = 64_000
@@ -7,6 +7,5 @@ cache_read = 0.5
 cache_write = 6.25

 [limit]
-context = 200_000
 input = 168_000
 output = 32_000
@@ -7,6 +7,5 @@ cache_read = 0.3
 cache_write = 3.75

 [limit]
-context = 200_000
 input = 168_000
 output = 32_000
@@ -14,7 +14,6 @@ cache_read = 0.5
 [limit]
 context = 400_000
 input = 272_000
-output = 128_000

 [experimental.modes.fast]
 cost = { input = 5, output = 30, cache_read = 0.5 }
@@ -14,7 +14,6 @@ cache_read = 1
 [limit]
 context = 400_000
 input = 272_000
-output = 128_000

 [experimental.modes.fast]
 cost = { input = 12.5, output = 75, cache_read = 1.25 }
@@ -7,8 +7,6 @@ cache_read = 0.5

 [limit]
 context = 409_600
-output = 128_000

 [modalities]
 input = ["text"]
-output = ["text"]
@@ -7,11 +7,9 @@ cache_read = 0.45

 [limit]
 context = 409_600
-output = 128_000

 [modalities]
 input = ["text"]
-output = ["text"]

 [experimental.modes.fast]
 cost = { input = 30, output = 150, cache_read = 3, cache_write = 37.5 }
@@ -7,8 +7,6 @@ cache_read = 0.3

 [limit]
 context = 409_600
-output = 64_000

 [modalities]
 input = ["text"]
-output = ["text"]
@@ -10,4 +10,3 @@ cache_read = 0.022

 [limit]
 context = 1_048_575
-output = 384_000
@@ -10,4 +10,3 @@ cache_read = 0.116

 [limit]
 context = 1_048_576
-output = 384_000
@@ -14,4 +14,3 @@ output = 65_536

 [modalities]
 input = ["text"]
-output = ["text"]
@@ -10,4 +10,3 @@ cache_read = 0.12

 [limit]
 context = 202_752
-output = 131_072
@@ -10,4 +10,3 @@ cache_read = 0.182

 [limit]
 context = 202_752
-output = 131_072
@@ -14,5 +14,4 @@ cache_read = 0.6
 cache_write = 7.5

 [limit]
-context = 1_000_000
 output = 128_000
@@ -14,7 +14,6 @@ cache_read = 0.6
 cache_write = 7.5

 [limit]
-context = 1_000_000
 output = 128_000

 [provider]
@@ -11,4 +11,3 @@ cache_write = 0

 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -7,5 +7,4 @@ output = 2
 cache_read = 0.08

 [limit]
-context = 262_144
 output = 65_536
@@ -14,4 +14,3 @@ output = 262_100

 [modalities]
 input = ["text", "image", "video"]
-output = ["text"]
@@ -12,5 +12,4 @@ output = 1.2
 cache_read = 0.055

 [limit]
-context = 204_800
 output = 204_800
@@ -10,4 +10,3 @@ cache_read = 0.2

 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -1,7 +1,6 @@
 base_model = "zai/glm-5.1"
 name = "GLM 5.1"
 knowledge = "2025-04"
-open_weights = true

 [interleaved]
 field = "reasoning_content"
@@ -13,4 +12,3 @@ cache_read = 0.27

 [limit]
 context = 202_800
-output = 131_072
@@ -10,4 +10,3 @@ output = 32_768

 [modalities]
 input = ["text"]
-output = ["text"]
@@ -7,9 +7,7 @@ output = 1.1
 cache_read = 0.056

 [limit]
-context = 262_144
 output = 32_768

 [modalities]
 input = ["text"]
-output = ["text"]
@@ -8,4 +8,3 @@ cache_write = 6.25

 [limit]
 context = 200_000
-output = 128_000
@@ -7,4 +7,3 @@ cache_read = 0.026

 [modalities]
 input = ["text"]
-output = ["text"]
@@ -16,4 +16,3 @@ cache_write = 7.5

 [limit]
 context = 1_000_000
-output = 64_000
@@ -15,5 +15,4 @@ cache_read = 0.6
 cache_write = 7.5

 [limit]
-context = 1_000_000
 output = 128_000
@@ -10,4 +10,3 @@ cache_read = 0.003625

 [limit]
 context = 1_048_576
-output = 384_000
@@ -8,5 +8,4 @@ cache_read = 0.01
 cache_write = 0.083333

 [limit]
-context = 1_048_576
 output = 65_535
@@ -8,5 +8,4 @@ cache_read = 0.03
 cache_write = 0.083333

 [limit]
-context = 1_048_576
 output = 65_535
@@ -5,9 +5,7 @@ input = 0.06
 output = 0.33

 [limit]
-context = 262_144
 output = 262_144

 [modalities]
 input = ["image", "text", "video"]
-output = ["text"]
@@ -8,4 +8,3 @@ output = 0

 [modalities]
 input = ["image", "text", "video"]
-output = ["text"]
@@ -5,9 +5,7 @@ input = 0.12
 output = 0.37

 [limit]
-context = 262_144
 output = 16_384

 [modalities]
 input = ["image", "text", "video"]
-output = ["text"]
@@ -8,4 +8,3 @@ output = 0

 [modalities]
 input = ["image", "text", "video"]
-output = ["text"]
@@ -7,4 +7,3 @@ output = 1.2

 [limit]
 context = 196_608
-output = 131_072
@@ -10,5 +10,4 @@ output = 1
 cache_read = 0.03

 [limit]
-context = 196_608
 output = 196_608
@@ -10,4 +10,3 @@ cache_read = 0.04

 [modalities]
 input = ["text", "pdf"]
-output = ["text"]
@@ -8,4 +8,3 @@ cache_read = 0.05

 [modalities]
 input = ["text", "image", "pdf"]
-output = ["text"]
@@ -12,4 +12,3 @@ cache_read = 0.09

 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -10,4 +10,3 @@ cache_read = 0.144

 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -12,4 +12,3 @@ output = 0

 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -7,4 +7,3 @@ cache_read = 0.025

 [modalities]
 input = ["image", "text", "pdf"]
-output = ["text"]
@@ -6,4 +6,3 @@ output = 15

 [modalities]
 input = ["text", "image", "pdf"]
-output = ["text"]
@@ -7,4 +7,3 @@ cache_read = 1.25

 [modalities]
 input = ["text", "image", "pdf"]
-output = ["text"]
@@ -7,4 +7,3 @@ cache_read = 1.25

 [modalities]
 input = ["text", "image", "pdf"]
-output = ["text"]
@@ -6,7 +6,3 @@ attachment = true
 input = 1.25
 output = 10
 cache_read = 0.125
-
-[limit]
-context = 400_000
-output = 128_000
@@ -6,10 +6,5 @@ input = 0.25
 output = 2
 cache_read = 0.025

-[limit]
-context = 400_000
-output = 128_000
-
 [modalities]
 input = ["text", "image", "pdf"]
-output = ["text"]
@@ -6,10 +6,5 @@ input = 0.05
 output = 0.4
 cache_read = 0.01

-[limit]
-context = 400_000
-output = 128_000
-
 [modalities]
 input = ["text", "image", "pdf"]
-output = ["text"]
@@ -6,9 +6,7 @@ input = 15
 output = 120

 [limit]
-context = 400_000
 output = 128_000

 [modalities]
 input = ["image", "text", "pdf"]
-output = ["text"]
@@ -5,7 +5,3 @@ base_model_omit = ["limit.input"]
 input = 1.25
 output = 10
 cache_read = 0.125
-
-[limit]
-context = 400_000
-output = 128_000
@@ -7,5 +7,4 @@ output = 2
 cache_read = 0.025

 [limit]
-context = 400_000
 output = 100_000
@@ -5,7 +5,3 @@ base_model_omit = ["limit.input"]
 input = 1.25
 output = 10
 cache_read = 0.13
-
-[limit]
-context = 400_000
-output = 128_000
@@ -6,10 +6,5 @@ input = 1.25
 output = 10
 cache_read = 0.13

-[limit]
-context = 400_000
-output = 128_000
-
 [modalities]
 input = ["image", "text", "pdf"]
-output = ["text"]
@@ -6,10 +6,5 @@ input = 1.75
 output = 14
 cache_read = 0.175

-[limit]
-context = 400_000
-output = 128_000
-
 [modalities]
 input = ["text", "image"]
-output = ["text"]
@@ -6,10 +6,5 @@ structured_output = true
 input = 21
 output = 168

-[limit]
-context = 400_000
-output = 128_000
-
 [modalities]
 input = ["image", "text", "pdf"]
-output = ["text"]
@@ -6,10 +6,5 @@ input = 1.75
 output = 14
 cache_read = 0.175

-[limit]
-context = 400_000
-output = 128_000
-
 [modalities]
 input = ["pdf", "image", "text"]
-output = ["text"]
@@ -5,7 +5,3 @@ base_model_omit = ["limit.input"]
 input = 1.75
 output = 14
 cache_read = 0.175
-
-[limit]
-context = 400_000
-output = 128_000
@@ -6,10 +6,5 @@ input = 0.75
 output = 4.5
 cache_read = 0.075

-[limit]
-context = 400_000
-output = 128_000
-
 [modalities]
 input = ["pdf", "image", "text"]
-output = ["text"]
@@ -6,10 +6,5 @@ input = 0.2
 output = 1.25
 cache_read = 0.02

-[limit]
-context = 400_000
-output = 128_000
-
 [modalities]
 input = ["pdf", "image", "text"]
-output = ["text"]
@@ -7,4 +7,3 @@ output = 180

 [modalities]
 input = ["text", "image", "pdf"]
-output = ["text"]
--- a/Show More
+++ b/Show More