Consolidate Gemma 4 E2B tokenizers and pin MLX multimodal to HF URLs

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: NorbertKlockiewicz

packages/react-native-executorch/src/constants/modelUrls.ts

@@ -16,18 +16,14 @@ const LLAMA3_2_1B_SPINQUANT_MODEL = `${URL_PREFIX}-llama-3.2/${PREVIOUS_VERSION_
 const LLAMA3_2_TOKENIZER = `${URL_PREFIX}-llama-3.2/${PREVIOUS_VERSION_TAG}/tokenizer.json`;
 const LLAMA3_2_TOKENIZER_CONFIG = `${URL_PREFIX}-llama-3.2/${PREVIOUS_VERSION_TAG}/tokenizer_config.json`;
 
-// GEMMA 4 — separate HF repo; tokenizer files are duplicated per-backend
-// folder, so each variant references its own.
-const GEMMA4_E2B_PREFIX = `${URL_PREFIX}-gemma-4/resolve/main/e2b`;
+// GEMMA 4 — separate HF repo; tokenizer files live at the e2b root and are
+// shared by all backend variants.
+const GEMMA4_E2B_PREFIX = `${URL_PREFIX}-gemma-4/${PREVIOUS_VERSION_TAG}/e2b`;
 const GEMMA4_E2B_MLX_MODEL = `${GEMMA4_E2B_PREFIX}/mlx/gemma4_e2b_mlx_int4.pte`;
-const GEMMA4_E2B_MLX_TOKENIZER = `${GEMMA4_E2B_PREFIX}/mlx/tokenizer.json`;
-const GEMMA4_E2B_MLX_TOKENIZER_CONFIG = `${GEMMA4_E2B_PREFIX}/mlx/tokenizer_config.json`;
 const GEMMA4_E2B_XNNPACK_MODEL = `${GEMMA4_E2B_PREFIX}/xnnpack/gemma_4_e2b_xnnpack_8da4w.pte`;
-const GEMMA4_E2B_XNNPACK_TOKENIZER = `${GEMMA4_E2B_PREFIX}/xnnpack/tokenizer.json`;
-const GEMMA4_E2B_XNNPACK_TOKENIZER_CONFIG = `${GEMMA4_E2B_PREFIX}/xnnpack/tokenizer_config.json`;
 const GEMMA4_E2B_VULKAN_MODEL = `${GEMMA4_E2B_PREFIX}/vulkan/gemma_4_e2b_vulkan_8da4w.pte`;
-const GEMMA4_E2B_VULKAN_TOKENIZER = `${GEMMA4_E2B_PREFIX}/vulkan/tokenizer.json`;
-const GEMMA4_E2B_VULKAN_TOKENIZER_CONFIG = `${GEMMA4_E2B_PREFIX}/vulkan/tokenizer_config.json`;
+const GEMMA4_E2B_TOKENIZER = `${GEMMA4_E2B_PREFIX}/tokenizer.json`;
+const GEMMA4_E2B_TOKENIZER_CONFIG = `${GEMMA4_E2B_PREFIX}/tokenizer_config.json`;
 
 /**
  * @category Models - LLM
@@ -138,18 +134,19 @@ export const QWEN3_0_6B_QUANTIZED = {
 } as const;
 
 // GEMMA 4
-const GEMMA4_E2B_XNNPACK_MM = `${URL_PREFIX}-gemma-4-multimodal/${PREVIOUS_VERSION_TAG}/e2b/xnnpack/gemma_4_e2b_xnnpack_8da4w.pte`;
 const GEMMA4_E2B_VULKAN_MM = `${URL_PREFIX}-gemma-4-multimodal/${PREVIOUS_VERSION_TAG}/e2b/vulkan/gemma_4_e2b_vulkan_8da4w.pte`;
 const GEMMA4_TOKENIZER = `${URL_PREFIX}-gemma-4/${PREVIOUS_VERSION_TAG}/e2b/tokenizer.json`;
 const GEMMA4_TOKENIZER_CONFIG = `${URL_PREFIX}-gemma-4/${PREVIOUS_VERSION_TAG}/e2b/tokenizer_config.json`;
 
+const GEMMA4_E2B_MLX_MM = `${URL_PREFIX}-gemma-4-multimodal/${PREVIOUS_VERSION_TAG}/e2b/mlx/gemma4_e2b_mlx_int4.pte`;
+
 /**
  * @category Models - VLM
  */
 export const GEMMA4_E2B_MM = {
   modelName: 'gemma4-e2b-multimodal',
   modelSource:
-    Platform.OS === `android` ? GEMMA4_E2B_VULKAN_MM : GEMMA4_E2B_XNNPACK_MM,
+    Platform.OS === `android` ? GEMMA4_E2B_VULKAN_MM : GEMMA4_E2B_MLX_MM,
   tokenizerSource: GEMMA4_TOKENIZER,
   tokenizerConfigSource: GEMMA4_TOKENIZER_CONFIG,
   capabilities: ['vision', 'audio'],
@@ -605,8 +602,8 @@ export const LFM2_VL_1_6B_QUANTIZED = LFM2_5_VL_1_6B_QUANTIZED;
 export const GEMMA4_E2B_MLX = {
   modelName: 'gemma-4-e2b-mlx',
   modelSource: GEMMA4_E2B_MLX_MODEL,
-  tokenizerSource: GEMMA4_E2B_MLX_TOKENIZER,
-  tokenizerConfigSource: GEMMA4_E2B_MLX_TOKENIZER_CONFIG,
+  tokenizerSource: GEMMA4_E2B_TOKENIZER,
+  tokenizerConfigSource: GEMMA4_E2B_TOKENIZER_CONFIG,
 } as const;
 
 /**
@@ -615,8 +612,8 @@ export const GEMMA4_E2B_MLX = {
 export const GEMMA4_E2B_XNNPACK = {
   modelName: 'gemma-4-e2b-xnnpack',
   modelSource: GEMMA4_E2B_XNNPACK_MODEL,
-  tokenizerSource: GEMMA4_E2B_XNNPACK_TOKENIZER,
-  tokenizerConfigSource: GEMMA4_E2B_XNNPACK_TOKENIZER_CONFIG,
+  tokenizerSource: GEMMA4_E2B_TOKENIZER,
+  tokenizerConfigSource: GEMMA4_E2B_TOKENIZER_CONFIG,
 } as const;
 
 /**
@@ -625,8 +622,8 @@ export const GEMMA4_E2B_XNNPACK = {
 export const GEMMA4_E2B_VULKAN = {
   modelName: 'gemma-4-e2b-vulkan',
   modelSource: GEMMA4_E2B_VULKAN_MODEL,
-  tokenizerSource: GEMMA4_E2B_VULKAN_TOKENIZER,
-  tokenizerConfigSource: GEMMA4_E2B_VULKAN_TOKENIZER_CONFIG,
+  tokenizerSource: GEMMA4_E2B_TOKENIZER,
+  tokenizerConfigSource: GEMMA4_E2B_TOKENIZER_CONFIG,
 } as const;
 
 /**