diff --git a/src/pages/ml.astro b/src/pages/ml.astro
index 6fc7a34..5be1ca1 100644
--- a/src/pages/ml.astro
+++ b/src/pages/ml.astro
@@ -1,123 +1,52 @@
---
import Base from "../layouts/Base.astro"
+import models from "../data/models-catalogue.json" with { type: "json" }
+const all = Object.values(models) as Array<{
+ id: string
+ task: string
+ precision: string
+ sizeHuman: string
+ splitParts: number | null
+ metrics: { name: string; value: number }[]
+ releaseUrl: string
+}>
-const serverModels = [
- {
- id: "ara-diac-2.0",
- task: "diacritization",
- pair: "Arabic → haraqat",
- metric: "DER 2.29 · OOD WER 17.38",
- artifact: "fp32 · 2.7 GiB · parts",
- status: "released",
- },
- {
- id: "khm-latn-1.0",
- task: "translit",
- pair: "Khmer → Latin",
- metric: "CER 27.42 · EM 59.66",
- artifact: "fp32 · 1.3 GiB",
- status: "released",
- },
- {
- id: "urd-g2p-1.0",
- task: "g2p",
- pair: "Urdu → IPA",
- metric: "CER 14.77 · EM 33.6",
- artifact: "fp32 · 1.3 GiB",
- status: "released",
- },
- {
- id: "urd-diac-1.0",
- task: "diacritization",
- pair: "Urdu → haraqat",
- metric: "CER 3.74",
- artifact: "fp32 · 1.3 GiB",
- status: "released",
- },
- {
- id: "heb-diac-1.0",
- task: "diacritization",
- pair: "Hebrew → niqqud",
- metric: "DER 29.0 greedy · 17.46 beam-4",
- artifact: "fp32 · 2.6 GiB · parts",
- status: "released",
- },
- {
- id: "tha-g2p-base-1.0",
- task: "g2p",
- pair: "Thai → IPA",
- metric: "PER 9.19 (teacher 4.43)",
- artifact: "fp32 · 2.6 GiB · parts",
- status: "released",
- },
-]
+const metricLabel = (m: { name: string; value: number }) => {
+ const names: Record
= {
+ der_total_greedy: "DER",
+ der_morph_greedy: "morph DER",
+ der_teacher_greedy: "DER (teacher)",
+ der_teacher_fullset: "DER (teacher)",
+ wer_wikinews2024_multiref: "OOD WER",
+ wer_wikinews_multiref: "OOD WER",
+ cer: "CER",
+ per: "PER",
+ em: "EM",
+ }
+ return `${names[m.name] ?? m.name} ${m.value}`
+}
-const clientModels = [
- {
- id: "ara-diac-plane-large-1.0",
- task: "diacritization · plane",
- pair: "Arabic → haraqat",
- metric: "DER 2.74 full-set — dedicated #2, ahead of Gemini-Flash",
- artifact: "int8 · 866 MiB · one graph",
- status: "released",
- },
- {
- id: "ara-diac-plane-1.0",
- task: "diacritization · plane",
- pair: "Arabic → haraqat",
- metric: "DER 3.59 · −45% size · −67% CPU latency vs seq2seq student",
- artifact: "int8 · 219 MiB · K-pass decode",
- status: "released",
- },
- {
- id: "ara-diac-small-2.1",
- task: "diacritization",
- pair: "Arabic → haraqat",
- metric: "DER 4.57 [1.91, 2.35] vs teacher (1.0 was 8.26)",
- artifact: "int8 · 264 MiB · fp16 · 674 MiB",
- status: "released",
- },
- {
- id: "ara-diac-layerdrop-1.0",
- task: "diacritization",
- pair: "Arabic → haraqat",
- metric: "DER 5.78 · 63% of the parameters",
- artifact: "int4 · 95 MiB · browser-native",
- status: "released",
- },
- {
- id: "tha-g2p-small-1.0",
- task: "g2p",
- pair: "Thai → IPA",
- metric: "PER 2.85 greedy (12.06 beam-4)",
- artifact: "int8 · 246 MiB",
- status: "released",
- },
- {
- id: "tha-g2p-small-1.0-int4",
- task: "g2p",
- pair: "Thai → IPA",
- metric: "same student · 4-bit",
- artifact: "int4 · 193 MiB",
- status: "released",
- },
- {
- id: "heb-diac-small-1.0",
- task: "diacritization",
- pair: "Hebrew → niqqud",
- metric: "DER 30.37 (teacher 24.79)",
- artifact: "fp32 · 1.3 GiB",
- status: "released",
- },
- {
- id: "fas-g2p-1.0",
- task: "g2p",
- pair: "Persian → IPA",
- metric: "CER ≈1.6 · homograph 77.34%",
- artifact: "fp32 · 2.6 GiB · parts",
- status: "released",
- },
-]
+const row = (m: (typeof all)[number]) => ({
+ id: m.id,
+ task: m.task,
+ metric: m.metrics.map(metricLabel).slice(0, 2).join(" · ") || "see RESULTS.md",
+ artifact:
+ `${m.precision} · ${m.sizeHuman}` +
+ (m.splitParts ? ` · ${m.splitParts} parts` : ""),
+ releaseUrl: m.releaseUrl,
+})
+
+const serverModels = all
+ .filter((m) => m.precision === "fp32")
+ .sort((a, b) => a.id.localeCompare(b.id))
+ .map(row)
+
+const clientModels = all
+ .filter((m) => m.precision !== "fp32")
+ .sort((a, b) => a.id.localeCompare(b.id))
+ .map(row)
+
+const modelCount = all.length
---
The catalogue
Every model, with its measured score.
- The index (v8) resolves 26 models across five languages. No model is published without a
+ The models.yaml registry resolves {modelCount} models across six languages. No model is published without a
measured metric next to its teacher's, on the same harness, in the open. Students are
distilled from frozen teachers and gated at a pre-agreed error budget. Each entry
resolves from the models.yaml index; artifacts above GitHub's 2 GiB cap ship as
@@ -173,7 +102,7 @@ const clientModels = [
{m.task}
- {m.id}
+ {m.id}
{m.pair}
{m.metric}
@@ -194,7 +123,7 @@ const clientModels = [
{m.task}
- {m.id}
+ {m.id}
{m.pair}
{m.metric}
diff --git a/test/maps-pages.test.ts b/test/maps-pages.test.ts
index c5f3fde..63b9620 100644
--- a/test/maps-pages.test.ts
+++ b/test/maps-pages.test.ts
@@ -24,7 +24,7 @@ describe("maps catalogue page", () => {
})
it("mentions the catalogue total", () => {
- expect(html).toMatch(/289 maps/)
+ expect(html).toMatch(/318 maps/)
})
it("mentions authority count", () => {