pi-lilac-provider 1.6.0 → 1.6.2

package/index.ts

@@ -398,8 +398,10 @@ function transformApiModel(apiModel: any): JsonModel | null {
   const hasImage = modalities.includes("image");
   const pricing = apiModel.pricing || {};
 
-  // Lilac API returns per-token pricing (e.g. "0.0000007" = $0.70/M tokens)
-  const toPerM = (v: any) => Math.round((typeof v === "string" ? parseFloat(v) : (v || 0)) * 1_000_000 * 100) / 100;
+  // Lilac API returns per-token pricing (e.g. "0.0000007" = $0.70/M tokens).
+  // Convert to $/M then round to 6 decimals: normalizes float noise from the ×1e6
+  // multiply and preserves sub-cent cache prices like 0.003 ($/M).
+  const toPerM = (v: any) => Math.round((typeof v === "string" ? parseFloat(v) : (v || 0)) * 1_000_000 * 1e6) / 1e6;
 
   const inputTypes: string[] = ["text"];
   if (hasImage) inputTypes.push("image");
@@ -491,9 +493,20 @@ function mergeWithEmbedded(liveModels: JsonModel[], embeddedModels: JsonModel[])
     const embedded = embeddedMap.get(liveModel.id);
     seen.add(liveModel.id);
     if (embedded) {
+      // Self-heal: live API pricing is authoritative field-by-field. Prefer the
+      // live cost when the API reports it (non-zero); fall back to embedded when
+      // the API is silent (0) so curated cacheRead/cacheWrite isn't clobbered and
+      // providers whose /models endpoint exposes no pricing keep their curated
+      // cost. Curation (reasoning/input/compat/name) still wins via ...embedded.
       result.push({
         ...liveModel,
         ...embedded,
+        cost: {
+          input: liveModel.cost.input || embedded.cost.input,
+          output: liveModel.cost.output || embedded.cost.output,
+          cacheRead: liveModel.cost.cacheRead || embedded.cost.cacheRead,
+          cacheWrite: liveModel.cost.cacheWrite || embedded.cost.cacheWrite,
+        },
         contextWindow: liveModel.contextWindow || embedded.contextWindow,
       });
     } else {
@@ -561,7 +574,9 @@ function applyDiscounts(models: JsonModel[], discounts: Map<string, JsonDiscount
     // E.g. "0.75" means pay 75% of list price. For MiniMax with "1.00" there's no discount.
     // discountPercent is informational (it equals (1 - creditMultiplier) * 100).
     const factor = discount.creditMultiplier;
-    const applyFactor = (n: number) => n > 0 ? Math.round(n * factor * 10000) / 10000 : n;
+    // Round to 6 decimals of $/M so a discounted sub-cent cache price
+    // (e.g. 0.003 × 0.75 = 0.00225) survives; 4-decimal rounding trimmed it to 0.0023.
+    const applyFactor = (n: number) => n > 0 ? Math.round(n * factor * 1e6) / 1e6 : n;
     return {
       ...model,
       cost: {
@@ -604,7 +619,8 @@ function applyDiscountInPlace(
   // discount was removed since it was last priced.
   const rawFactor = discounts?.get(model.id)?.creditMultiplier;
   const factor = Number.isFinite(rawFactor) && rawFactor !== undefined ? rawFactor : 1;
-  const applyFactor = (n: number) => n > 0 ? Math.round(n * factor * 10000) / 10000 : n;
+  // Round to 6 decimals of $/M so a discounted sub-cent cache price survives.
+  const applyFactor = (n: number) => n > 0 ? Math.round(n * factor * 1e6) / 1e6 : n;
   model.cost.input = applyFactor(list.cost.input);
   model.cost.output = applyFactor(list.cost.output);
   model.cost.cacheRead = applyFactor(list.cost.cacheRead);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-lilac-provider",
-  "version": "1.6.0",
+  "version": "1.6.2",
   "description": "Lilac provider extension for pi - Access Kimi K2.6, GLM 5.1, and Gemma 4 models through Lilac's OpenAI-compatible API on idle GPUs",
   "type": "module",
   "main": "index.ts",

package/scripts/update-models.js

@@ -55,7 +55,9 @@ function saveJson(filePath, data) {
 // Convert per-token pricing from API to per-million-tokens
 function toPerMillion(val) {
   if (val === '' || val === null || val === undefined) return null;
-  return Math.round(parseFloat(val) * 1_000_000 * 100) / 100;
+  // Round to 6 decimals of $/M: normalizes float noise from the ×1e6 multiply
+  // and preserves sub-cent cache prices like 0.003 ($/M).
+  return Math.round(parseFloat(val) * 1_000_000 * 1e6) / 1e6;
 }
 
 // ─── API fetch ───────────────────────────────────────────────────────────────