Add Genie NPU models (Qwen3 4B, Llama 3.2 1B) with new HuggingFace URLs

- Update NPUChip: new base URL (runanywhere/genie-npu-models), new naming
  convention ({slug}-genie-w4a16-{chipSuffix}.tar.gz)
- Add Llama 3.2 1B Instruct (both 8 Elite and 8 Elite Gen 5)
- Add Qwen3 4B (8 Elite Gen 5 only)
- Extensible GenieModelDef pattern for easy addition of future models

Author: Siddhesh2377

examples/android/RunAnywhereAI/app/src/main/java/com/runanywhere/runanywhereai/data/ModelList.kt

@@ -4,6 +4,7 @@ import timber.log.Timber
 import com.runanywhere.runanywhereai.data.models.AppModel
 import com.runanywhere.sdk.core.onnx.ONNX
 import com.runanywhere.sdk.core.types.InferenceFramework
+import com.runanywhere.sdk.core.types.NPUChip
 import com.runanywhere.sdk.llm.llamacpp.LlamaCPP
 import com.runanywhere.sdk.llm.genie.Genie
 import com.runanywhere.sdk.public.RunAnywhere
@@ -129,18 +130,44 @@ object ModelList {
 
     // Genie NPU Models — URLs are built dynamically based on detected chipset.
     // getChip() returns the NPUChip for this device, or null if unsupported.
+    // Each entry specifies which chips it supports; only matching models are shown.
+
+    private data class GenieModelDef(
+        val slug: String,
+        val name: String,
+        val memoryRequirement: Long,
+        val supportedChips: Set<NPUChip>,
+    )
+
+    private val genieModelDefs = listOf(
+        GenieModelDef(
+            slug = "qwen3-4b",
+            name = "Qwen3 4B",
+            memoryRequirement = 2_800_000_000,
+            supportedChips = setOf(NPUChip.SNAPDRAGON_8_ELITE_GEN5),
+        ),
+        GenieModelDef(
+            slug = "llama-v3.2-1b-instruct",
+            name = "Llama 3.2 1B Instruct",
+            memoryRequirement = 1_200_000_000,
+            supportedChips = setOf(NPUChip.SNAPDRAGON_8_ELITE, NPUChip.SNAPDRAGON_8_ELITE_GEN5),
+        ),
+    )
+
     private fun genieModels(): List<AppModel> {
         val chip = RunAnywhere.getChip() ?: return emptyList()
-        return listOf(
-            AppModel(
-                id = "qwen-npu-${chip.identifier}",
-                name = "Qwen3 4B (NPU - ${chip.displayName})",
-                url = chip.downloadUrl("qwen"),
-                framework = InferenceFramework.GENIE,
-                category = ModelCategory.LANGUAGE,
-                memoryRequirement = 2_800_000_000,
-            ),
-        )
+        return genieModelDefs
+            .filter { chip in it.supportedChips }
+            .map { def ->
+                AppModel(
+                    id = "${def.slug}-npu-${chip.identifier}",
+                    name = "${def.name} (NPU - ${chip.displayName})",
+                    url = chip.downloadUrl(def.slug),
+                    framework = InferenceFramework.GENIE,
+                    category = ModelCategory.LANGUAGE,
+                    memoryRequirement = def.memoryRequirement,
+                )
+            }
     }
 
     // VLM

sdk/runanywhere-kotlin/src/commonMain/kotlin/com/runanywhere/sdk/core/types/NPUChip.kt

@@ -3,34 +3,37 @@ package com.runanywhere.sdk.core.types
 /**
  * Supported NPU chipsets for on-device Genie model inference.
  *
- * Each chip has an [identifier] used to construct dynamic download URLs
- * for chipset-specific NPU model binaries.
+ * Each chip has an [identifier] used in model IDs and an [npuSuffix] used
+ * to construct download URLs from the HuggingFace model repository.
  *
  * Example URL construction:
  * ```
  * val chip = RunAnywhere.getChip()
- * val url = "${NPUChip.BASE_URL}qwen-${chip.identifier}.zip?download=true"
+ * val url = chip.downloadUrl("qwen3-4b")
+ * // → "https://huggingface.co/runanywhere/genie-npu-models/resolve/main/qwen3-4b-genie-w4a16-8elite-gen5.tar.gz"
  * ```
  */
 enum class NPUChip(
     val identifier: String,
     val displayName: String,
     val socModel: String,
+    val npuSuffix: String,
 ) {
-    SNAPDRAGON_8_ELITE("gen1", "Snapdragon 8 Elite", "SM8750"),
-    SNAPDRAGON_8_ELITE_GEN5("gen2", "Snapdragon 8 Elite Gen 5", "SM8850"),
+    SNAPDRAGON_8_ELITE("8elite", "Snapdragon 8 Elite", "SM8750", "8elite"),
+    SNAPDRAGON_8_ELITE_GEN5("8elite-gen5", "Snapdragon 8 Elite Gen 5", "SM8850", "8elite-gen5"),
     ;
 
     /**
      * Build a HuggingFace download URL for this chip.
-     * @param modelName Model prefix (e.g. "qwen") → produces "qwen-gen1.zip"
+     * @param modelSlug Model slug (e.g. "qwen3-4b") → produces
+     *   "qwen3-4b-genie-w4a16-8elite-gen5.tar.gz"
      */
-    fun downloadUrl(modelName: String): String =
-        "${BASE_URL}${modelName}-${identifier}.zip?download=true"
+    fun downloadUrl(modelSlug: String): String =
+        "${BASE_URL}${modelSlug}-genie-w4a16-${npuSuffix}.tar.gz"
 
     companion object {
         /** Base URL for NPU model downloads on HuggingFace. */
-        const val BASE_URL = "https://huggingface.co/Void2377/npu-models/resolve/main/"
+        const val BASE_URL = "https://huggingface.co/runanywhere/genie-npu-models/resolve/main/"
 
         /**
          * Match an NPU chip from a SoC model string (e.g. "SM8750").

sdk/runanywhere-kotlin/src/commonMain/kotlin/com/runanywhere/sdk/public/extensions/RunAnywhere+Device.kt

@@ -9,12 +9,12 @@ import com.runanywhere.sdk.public.RunAnywhere
  * Returns the [NPUChip] if the device has a supported Qualcomm SoC,
  * or null if the device does not support NPU inference.
  *
- * Use [NPUChip.identifier] to construct chipset-specific download URLs:
+ * Use [NPUChip.downloadUrl] to construct chipset-specific download URLs:
  * ```kotlin
  * val chip = RunAnywhere.getChip()
  * if (chip != null) {
- *     val url = "https://example.com/models/qwen-${chip.identifier}.zip"
- *     RunAnywhere.registerModel(id = "qwen-npu", name = "Qwen NPU", url = url, ...)
+ *     val url = chip.downloadUrl("qwen3-4b")
+ *     RunAnywhere.registerModel(id = "qwen3-4b-npu", name = "Qwen3 4B NPU", url = url, ...)
  * }
  * ```
  */