Playground chat: Gemma 3 1B Instruct replaces Llama 3.2 1B (#24014)

Summary:

On GPUs with `shader-f16` (Apple GPUs, most others), Llama 3.2 1B in the playground's On-device Chat fails on its first reply with "linear_q4gsw_bk64_qkv(resize): separate projection dispatch was not restored before the QKV resize hook ran". That is a WebGPU runtime bug in the fused Q/K/V path, which only Llama's shapes take; D126153972 fixes it upstream. Until the fixed runtime ships, the chat offers Qwen 3 0.6B and Gemma 3 1B Instruct.

- Gemma 3 1B Instruct is loaded from `executorch-community/Gemma-3-1B-IT-ExecuTorch-WebGPU` on Hugging Face, pinned to commit `91864e51`, with a sha256 check per file. Export: 4-bit groupwise weights, exact sliding-window and global attention, 8,960-token context. Its repo carries the Gemma Terms of Use, the Prohibited Use Policy and the required NOTICE. Legal approved publishing it.
- The chat shows Gemma's required notice ("Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms") when Gemma is selected. Llama leaves the chat and the Models page; its Hugging Face repo stays.
- `llm.js`: when a model's chat template can't render the conversation, the chat reports it instead of silently falling back to ChatML, which only Qwen understands.
- Model cards lay out in one row.

`fbcode/executorch/website` and `xplat/executorch/website` stay byte-identical.

Differential Revision: D126233262

Author: psiddh

website/playground/webgpu/src/config/catalog.js

@@ -28,7 +28,7 @@ export const DEMOS = [
     title: 'On-device Chat',
     blurb: 'Chat with a real language model running entirely on your GPU. Talk to it, too.',
     meta: { label: 'Models', value: 'Qwen 3 0.6B, Llama 3.2 1B' },
-    models: ['qwen3-0.6b', 'llama-3.2-1b-instruct'],
+    models: ['qwen3-0.6b', 'gemma-3-1b-it'],
     compute: COMPUTE.EXECUTORCH,
     view: 'chat',
     art: 'chat',

website/playground/webgpu/src/config/models.js

@@ -116,6 +116,56 @@ export const MODELS = [
       provenance: 'Upstream: Qwen/Qwen3-0.6B.',
     },
   },
+  {
+    id: 'gemma-3-1b-it',
+    name: 'Gemma 3 1B Instruct',
+    shortName: 'Gemma 3 1B',
+    task: 'text-generation',
+    tagline: 'Larger · by Google DeepMind',
+    approxBytes: 675e6,
+    approxBytesIsEstimate: false,
+    arch: { layers: 26, dModel: 1152, ffn: 6912, heads: 4, kvHeads: 1, vocab: 262144 },
+    assets: {
+      "baseUrl": "https://huggingface.co/executorch-community/Gemma-3-1B-IT-ExecuTorch-WebGPU/resolve/91864e518e371b57a6433c180490ce5ab31639ec/", // @lint-ignore (base URL, files resolve under it)
+      "files": [
+        "gemma3_1b_it_q4gsw.pte",
+        "tokenizer/config.json",
+        "tokenizer/generation_config.json",
+        "tokenizer/special_tokens_map.json",
+        "tokenizer/tokenizer.json",
+        "tokenizer/tokenizer_config.json"
+      ],
+      "sha256": {
+        "gemma3_1b_it_q4gsw.pte": "6056492708430182103f2b28cac2cc5d4d43a66a95236f070de3312542f11c25",
+        "tokenizer/config.json": "19cb5d28c97778271ba2b3c3df47bf76bdd6706724777a2318b3522230afe91e",
+        "tokenizer/generation_config.json": "fd9324becc53c4be610db39e13a613006f09fd6ef71a95fb6320dc33157490a3",
+        "tokenizer/special_tokens_map.json": "2f7b0adf4fb469770bb1490e3e35df87b1dc578246c5e7e6fc76ecf33213a397",
+        "tokenizer/tokenizer.json": "4667f2089529e8e7657cfb6d1c19910ae71ff5f28aa7ab2ff2763330affad795",
+        "tokenizer/tokenizer_config.json": "bfe25c2735e395407beb78456ea9a6984a1f00d8c16fa04a8b75f2a614cf53e1"
+      },
+      "spec": {
+        "pte": "gemma3_1b_it_q4gsw.pte",
+        "pteBytes": 640305924,
+        "tokenizer": "tokenizer",
+        "chat": "template",
+        "prefillMax": 512,
+        "maxCtx": 8960,
+        "eos": [
+          1,
+          106
+        ]
+      }
+    },
+    credit: { by: 'Google DeepMind', url: 'https://huggingface.co/google/gemma-3-1b-it', paper: 'https://arxiv.org/abs/2503.19786' },
+    attribution: 'Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms',
+    license: {
+      status: LICENSE_STATUS.CLEARED,
+      reviewNote: "Approved for public redistribution (legal/OSS approval confirmed by the project owner, 2026-10-16).",
+      upstream: 'Gemma Terms of Use + Gemma Prohibited Use Policy',
+      provenance:
+        'Upstream: google/gemma-3-1b-it (gated on Hugging Face). Use is subject to the Gemma Terms of Use, including the use restrictions in its Section 3.2. ',
+    },
+  },
   {
     id: 'sam2-hiera-tiny',
     name: 'SAM 2.1 Hiera Tiny',

website/playground/webgpu/src/runtime/llm.js

@@ -67,6 +67,7 @@ export class ExecuTorchLLM {
       } catch {
         ids = null;
       }
+      if (!ids?.length) throw new Error('The chat template could not render this conversation.');
     }
     if (!ids?.length) {
       // ChatML, the format of the Qwen family. "nothink" pre-closes Qwen 3's

website/playground/webgpu/styles/app.css

@@ -608,7 +608,7 @@ p { margin: 0; }
 .preload__glyph--ok { background: var(--live-soft); color: var(--live); box-shadow: none; }
 .preload__title { font-size: 24px; font-weight: 700; letter-spacing: -0.02em; }
 .preload__sub { color: var(--text-3); font-size: 14px; }
-.model-choice { display: grid; grid-template-columns: repeat(2, minmax(0, 1fr)); gap: 10px; width: 100%; max-width: 440px; }
+.model-choice { display: grid; grid-template-columns: repeat(auto-fit, minmax(160px, 1fr)); gap: 10px; width: 100%; max-width: 600px; }
 .model-choice__opt { display: flex; flex-direction: column; align-items: flex-start; gap: 2px; padding: 10px 14px; border: 1px solid var(--border); border-radius: 12px; background: var(--surface); color: var(--text); text-align: left; cursor: pointer; }
 .model-choice__opt:hover { border-color: var(--border-strong); }
 .model-choice__opt[aria-checked="true"] { border-color: var(--accent); box-shadow: inset 0 0 0 1px var(--accent); background: var(--accent-soft); }