Skip to content

Open-source AI browser agent for Chrome and Firefox (monorepo) 🧠

feat: add MiniCPM5-2B WebGPU preset with XML tool-call support

Emre Sokullub3d4efe85876
diff --git a/docs/providers-and-models.md b/docs/providers-and-models.md
index ad3ddee28..6759de8fc 100644
--- a/docs/providers-and-models.md
+++ b/docs/providers-and-models.md
@@ -47,7 +47,7 @@ class BaseLLMProvider {
| `gpt4all` | `openai` | local | (loaded model) | Yes (default on) |
| `local_openai_proxy` | `openai` | local | (required) | Off / manual toggle |
| `unsloth` | `openai` | local | (required) | Off / manual toggle |
-| `webgpu` (Chromium) | `webgpu` | local | Seven shipped LFM2.5/Nanbeige/Bonsai presets; experimental custom HF ONNX repos | LFM2.5-VL presets |
+| `webgpu` (Chromium) | `webgpu` | local | Eight shipped LFM2.5/Nanbeige/MiniCPM5/Bonsai presets; experimental custom HF ONNX repos | LFM2.5-VL presets |
| `azure_openai` | `azure_openai` | cloud | (deployment) | Manual toggle |
| `aws_bedrock` | `aws_bedrock` | cloud | (model id) | No |
| `openai` | `openai` | cloud | `gpt-5.6-terra` | Model-name regex |
@@ -150,7 +150,7 @@ duplicate request.
### Local Providers
On Chromium, **WebGPU (In-browser)** is an endpoint-free local provider. Its
-Apocalypse text picker offers seven shipped presets:
+Apocalypse text picker offers eight shipped presets:
- [`LiquidAI/LFM2.5-2.6B-ONNX`](https://huggingface.co/LiquidAI/LFM2.5-2.6B-ONNX/)
(`q4f16`, about 1.55 GB) through the packaged Transformers.js 4.2 / ONNX
@@ -185,6 +185,15 @@ Apocalypse text picker offers seven shipped presets:
implementation is the [Nanbeige Browser Lab](https://huggingface.co/spaces/borkiss/nanbeige4-2-3b-browser-lab)
Space, which drives the same export through a hand-written ONNX Runtime Web
loop rather than Transformers.js.
+- [`RASMUS/MiniCPM5-2B-ONNX`](https://huggingface.co/RASMUS/MiniCPM5-2B-ONNX)
+ (`q4f16`, about 1.83 GB), a browser-ready ONNX export of
+ [`openbmb/MiniCPM5-2B`](https://huggingface.co/openbmb/MiniCPM5-2B) built with
+ `onnxruntime-genai -e webgpu` (`MatMulNBits` + `GroupQueryAttention`, default
+ `onnx/model_q4f16.onnx` layout). It needs an adapter with `shader-f16` and
+ emits XML-style tool calls
+ (`<function name="..."><param name="...">...</param></function>`, CDATA-wrapped
+ when values contain `<`, `&`, or newlines), which the local fallback parser
+ accepts. Its practical context is 16k.
- [`prism-ml/Bonsai-27B-gguf`](https://huggingface.co/prism-ml/Bonsai-27B-gguf)
(`Q1_0`, about 3.8 GB) through a dedicated vendored [bitgpu](https://github.com/stfurkan/bitgpu)
worker. Bonsai is opt-in: WebBrain never auto-downloads the 27B weights.
@@ -201,11 +210,11 @@ Do not point Transformers.js at the Bonsai GGUF — 27B is not an ONNX pipeline.
The provider defaults to the Compact prompt tier with a conservative 16k
practical context setting (4k for Nanbeige and Bonsai). Text-only presets
reject image blocks; the two LFM2.5-VL presets route image and text blocks
-through the local image-text runtime. LFM2.5 2.6B, 1.2B Thinking, and
-Nanbeige4.2-3B use reasoning paths; WebBrain keeps completed thinking out of
+through the local image-text runtime. LFM2.5 2.6B, 1.2B Thinking,
+Nanbeige4.2-3B, and MiniCPM5-2B use reasoning paths; WebBrain keeps completed thinking out of
the visible answer and reports an error when a template that opens `<think>`
in the generation prompt exhausts its output budget before closing it. Those
-three presets decode with their publisher's recommended sampling settings; the
+four presets decode with their publisher's recommended sampling settings; the
remaining ONNX presets stay greedy. Bonsai uses bitgpu
`think: true` with a 128-token think budget and the same post-think visible-
answer UX. Each repository is cached separately in Chrome.
diff --git a/docs/remote-downloads.md b/docs/remote-downloads.md
index 0f7c8f49e..24507c133 100644
--- a/docs/remote-downloads.md
+++ b/docs/remote-downloads.md
@@ -22,7 +22,7 @@ WebBrain is designed to minimize remote network dependencies. All downloads fall
| Component | Remote Server / Origin | Origin Description | Typical Size | Protocol / Method | Checksum & Integrity | Local Storage Destination |
| :--- | :--- | :--- | :--- | :--- | :--- | :--- |
-| **WebGPU Text / Multimodal Models** | `huggingface.co` / Hugging Face CDN | Seven shipped presets: LiquidAI LFM2.5 2.6B, 1.2B Instruct, 1.2B Thinking, VL 1.6B, VL 3B, Michionlion Nanbeige4.2-3B, and prism-ml Bonsai 27B | ~0.76 – 4.0 GB per selected preset | HTTPS GET (Transformers.js/ONNX or bitgpu) | Hugging Face Git LFS SHA-256 hash | Browser Cache API & IndexedDB (`transformers-cache`) |
+| **WebGPU Text / Multimodal Models** | `huggingface.co` / Hugging Face CDN | Eight shipped presets: LiquidAI LFM2.5 2.6B, 1.2B Instruct, 1.2B Thinking, VL 1.6B, VL 3B, Michionlion Nanbeige4.2-3B, RASMUS MiniCPM5-2B, and prism-ml Bonsai 27B | ~0.76 – 4.0 GB per selected preset | HTTPS GET (Transformers.js/ONNX or bitgpu) | Hugging Face Git LFS SHA-256 hash | Browser Cache API & IndexedDB (`transformers-cache`) |
| **Local Vision Model** | `huggingface.co` / Hugging Face CDN | `webbrain-one/webbrain-vl-2-450M-onnx` weights for local screenshot description and UI analysis | ~810 MB | HTTPS GET (Transformers.js pipeline) | Hugging Face Git LFS SHA-256 hash | Browser Cache API & IndexedDB (`transformers-cache`) |
| **Emergency Text Pack & SQLite Index** | `github.com/webbrain-one/emergency-box-corpus` (GitHub Releases CDN) | Release assets for curated public-domain field references, prebuilt SQLite FTS5 database, and precomputed E5 embeddings | ~245 MB (compressed ZIP) | Resumable HTTP `Range: bytes={offset}-` streaming fetch | Strict **SHA-256** hash comparison against hardcoded release descriptor before activation | OPFS (`webbrain-offline-rag/emergency-box-text/`) & IndexedDB (`webbrain_offline_rag`) |
| **Multilingual Semantic Model** | `huggingface.co` / Hugging Face CDN (`Xenova/multilingual-e5-small`) | ONNX weights for multilingual query embedding and vector search / candidate reranking | ~134 MB | HTTPS GET (ONNX Runtime Web / Transformers.js) | SHA-256 verification via Transformers.js manifest | Browser Cache API & IndexedDB (`transformers-cache`) |
diff --git a/src/chrome/src/agent/tool-call-parser.js b/src/chrome/src/agent/tool-call-parser.js
index f55ab7bb8..42d680b97 100644
--- a/src/chrome/src/agent/tool-call-parser.js
+++ b/src/chrome/src/agent/tool-call-parser.js
@@ -357,7 +357,12 @@ export function parseToolCallsFromText(text, allowedNames) {
const results = [];
const parseXmlParamValue = (value) => {
- const cleaned = String(value || '')
+ const raw = String(value || '');
+ // MiniCPM5 wraps values containing <, &, or newlines in CDATA. Extract
+ // the literal content first so the tag strip below does not eat it.
+ const cdataMatch = /^\s*<!\[CDATA\[([\s\S]*?)\]\]>\s*$/.exec(raw);
+ if (cdataMatch) return cdataMatch[1];
+ const cleaned = raw
.replace(/<[^>]+>/g, '')
.trim();
if (!cleaned) return '';
@@ -435,13 +440,36 @@ export function parseToolCallsFromText(text, allowedNames) {
// XML-ish tool-call format used by some local/chat-template models:
// <tool_call><function=click_ax><parameter=ref_id>ref_6</parameter>...
const xmlToolRe = /<tool_call>\s*<function(?:\s*=\s*["']?([A-Za-z_]\w*)["']?|\s+name\s*=\s*["']?([A-Za-z_]\w*)["']?)\s*>\s*([\s\S]*?)\s*<\/function>\s*<\/tool_call>/gi;
+ const xmlToolSpans = [];
let xmlMatch;
while ((xmlMatch = xmlToolRe.exec(text)) !== null) {
+ xmlToolSpans.push({ start: xmlMatch.index, end: xmlMatch.index + xmlMatch[0].length });
const toolName = xmlMatch[1] || xmlMatch[2];
if (!allowedNames.has(toolName)) continue;
const body = xmlMatch[3] || '';
const args = {};
- const paramRe = /<parameter(?:\s*=\s*["']?([A-Za-z_]\w*)["']?|\s+name\s*=\s*["']?([A-Za-z_]\w*)["']?)\s*>\s*([\s\S]*?)\s*<\/parameter>/gi;
+ const paramRe = /<(?:param|parameter)(?:\s*=\s*["']?([A-Za-z_]\w*)["']?|\s+name\s*=\s*["']?([A-Za-z_]\w*)["']?)\s*>\s*([\s\S]*?)\s*<\/(?:param|parameter)>/gi;
+ let paramMatch;
+ while ((paramMatch = paramRe.exec(body)) !== null) {
+ const key = paramMatch[1] || paramMatch[2];
+ if (!key) continue;
+ args[key] = parseXmlParamValue(paramMatch[3]);
+ }
+ results.push({ name: toolName, arguments: args });
+ }
+
+ // MiniCPM5-2B native tool format (no outer <tool_call> wrapper):
+ // <function name="click"><param name="ref_id">ref_6</param>...</function>
+ const minicpmFunctionRe = /<function(?:\s+name\s*=\s*["']([A-Za-z_]\w*)["']|\s*=\s*["']?([A-Za-z_]\w*)["']?)\s*>\s*([\s\S]*?)\s*<\/function>/gi;
+ let minicpmMatch;
+ while ((minicpmMatch = minicpmFunctionRe.exec(text)) !== null) {
+ // Skip functions already consumed inside a <tool_call> wrapper above.
+ if (xmlToolSpans.some(span => minicpmMatch.index >= span.start && minicpmMatch.index < span.end)) continue;
+ const toolName = minicpmMatch[1] || minicpmMatch[2];
+ if (!allowedNames.has(toolName)) continue;
+ const body = minicpmMatch[3] || '';
+ const args = {};
+ const paramRe = /<(?:param|parameter)(?:\s*=\s*["']?([A-Za-z_]\w*)["']?|\s+name\s*=\s*["']?([A-Za-z_]\w*)["']?)\s*>\s*([\s\S]*?)\s*<\/(?:param|parameter)>/gi;
let paramMatch;
while ((paramMatch = paramRe.exec(body)) !== null) {
const key = paramMatch[1] || paramMatch[2];
diff --git a/src/chrome/src/offscreen/inference-worker.js b/src/chrome/src/offscreen/inference-worker.js
index af8ff7ba2..cc657e1ab 100644
--- a/src/chrome/src/offscreen/inference-worker.js
+++ b/src/chrome/src/offscreen/inference-worker.js
@@ -31,6 +31,7 @@ const WEBGPU_LFM25_12B_THINKING_MODEL_ID = 'LiquidAI/LFM2.5-1.2B-Thinking-ONNX';
const WEBGPU_LFM25_VL_16B_MODEL_ID = 'LiquidAI/LFM2.5-VL-1.6B-ONNX';
const WEBGPU_LFM25_VL_3B_MODEL_ID = 'LiquidAI/LFM2.5-VL-3B-ONNX';
const WEBGPU_NANBEIGE42_3B_MODEL_ID = 'Michionlion/Nanbeige4.2-3B-ONNX-WebGPU';
+const WEBGPU_MINICPM5_2B_MODEL_ID = 'RASMUS/MiniCPM5-2B-ONNX';
const WEBGPU_BONSAI27_MODEL_ID = 'prism-ml/Bonsai-27B-gguf';
const WEBGPU_LFM25_MAX_NEW_TOKENS = 2048;
const WEBGPU_LFM25_TEXT_MODEL_IDS = new Set([
@@ -48,6 +49,7 @@ const WEBGPU_REASONING_MODEL_IDS = new Set([
WEBGPU_LFM25_MODEL_ID,
WEBGPU_LFM25_12B_THINKING_MODEL_ID,
WEBGPU_NANBEIGE42_3B_MODEL_ID,
+ WEBGPU_MINICPM5_2B_MODEL_ID,
]);
// Chat templates that emit the opening `<think>` themselves, so the runtime
// only returns the reasoning suffix.
@@ -58,6 +60,7 @@ const WEBGPU_OPEN_THINKING_MODEL_IDS = new Set([
const WEBGPU_LONG_OUTPUT_MODEL_IDS = new Set([
...WEBGPU_LFM25_TEXT_MODEL_IDS,
WEBGPU_NANBEIGE42_3B_MODEL_ID,
+ WEBGPU_MINICPM5_2B_MODEL_ID,
]);
// Publisher-recommended decoding for the shipped reasoning presets. Custom
// repositories keep Transformers.js greedy defaults.
@@ -66,6 +69,8 @@ const WEBGPU_TEXT_SAMPLING = new Map([
[WEBGPU_LFM25_12B_THINKING_MODEL_ID, { temperature: 0.05, top_k: 50, repetition_penalty: 1.05 }],
// Nanbeige4.2-3B's own generation_config.json.
[WEBGPU_NANBEIGE42_3B_MODEL_ID, { temperature: 0.6, top_k: 20, top_p: 0.95 }],
+ // MiniCPM5-2B quickstart: temperature=1.0, top_p=0.95.
+ [WEBGPU_MINICPM5_2B_MODEL_ID, { temperature: 1.0, top_p: 0.95 }],
]);
// Nanbeige ships a single WebGPU-fused graph under a non-default file name.
// Without this override Transformers.js looks for `onnx/model_q4f16.onnx`,
diff --git a/src/chrome/src/providers/webgpu.js b/src/chrome/src/providers/webgpu.js
index 2e7f6f9bd..b89c0c253 100644
--- a/src/chrome/src/providers/webgpu.js
+++ b/src/chrome/src/providers/webgpu.js
@@ -11,6 +11,7 @@ export const WEBGPU_LFM25_12B_THINKING_MODEL_ID = 'LiquidAI/LFM2.5-1.2B-Thinking
export const WEBGPU_LFM25_VL_16B_MODEL_ID = 'LiquidAI/LFM2.5-VL-1.6B-ONNX';
export const WEBGPU_LFM25_VL_3B_MODEL_ID = 'LiquidAI/LFM2.5-VL-3B-ONNX';
export const WEBGPU_NANBEIGE42_3B_MODEL_ID = 'Michionlion/Nanbeige4.2-3B-ONNX-WebGPU';
+export const WEBGPU_MINICPM5_2B_MODEL_ID = 'RASMUS/MiniCPM5-2B-ONNX';
export const WEBGPU_BONSAI27_MODEL_ID = 'prism-ml/Bonsai-27B-gguf';
export const WEBGPU_DTYPE = 'q4f16';
export const WEBGPU_BONSAI27_DTYPE = 'q1';
@@ -90,6 +91,19 @@ export const WEBGPU_MODEL_PRESETS = Object.freeze([
contextWindow: 4096,
supportsVision: false,
}),
+ Object.freeze({
+ id: WEBGPU_MINICPM5_2B_MODEL_ID,
+ runtime: WEBGPU_RUNTIME_ONNX,
+ label: 'MiniCPM5-2B',
+ size: '1.83 GB',
+ dtype: WEBGPU_DTYPE,
+ dtypeLabel: WEBGPU_DTYPE,
+ // OpenBMB's base context is 131k, but the practical browser setting stays
+ // at 16k like the LFM2.5 text presets: 42 GQA layers cost roughly 42 KB
+ // per token in FP16, so 16k needs about 672 MB of cache on top of weights.
+ contextWindow: 16384,
+ supportsVision: false,
+ }),
Object.freeze({
id: WEBGPU_BONSAI27_MODEL_ID,
runtime: WEBGPU_RUNTIME_BITGPU,
diff --git a/src/chrome/src/ui/apocalypse-mode.html b/src/chrome/src/ui/apocalypse-mode.html
index 25cd60b44..d0a099f57 100644
--- a/src/chrome/src/ui/apocalypse-mode.html
+++ b/src/chrome/src/ui/apocalypse-mode.html
@@ -167,6 +167,10 @@
<input type="radio" name="webgpu-text-preset" value="Michionlion/Nanbeige4.2-3B-ONNX-WebGPU" data-webgpu-text-preset>
<span>Nanbeige4.2-3B</span>
</label>
+ <label class="webgpu-text-preset">
+ <input type="radio" name="webgpu-text-preset" value="RASMUS/MiniCPM5-2B-ONNX" data-webgpu-text-preset>
+ <span>MiniCPM5-2B</span>
+ </label>
<label class="webgpu-text-preset">
<input type="radio" name="webgpu-text-preset" value="prism-ml/Bonsai-27B-gguf" data-webgpu-text-preset>
<span data-i18n="ap.models.text.bonsai">Basic text model</span>
diff --git a/src/firefox/src/agent/tool-call-parser.js b/src/firefox/src/agent/tool-call-parser.js
index f55ab7bb8..42d680b97 100644
--- a/src/firefox/src/agent/tool-call-parser.js
+++ b/src/firefox/src/agent/tool-call-parser.js
@@ -357,7 +357,12 @@ export function parseToolCallsFromText(text, allowedNames) {
const results = [];
const parseXmlParamValue = (value) => {
- const cleaned = String(value || '')
+ const raw = String(value || '');
+ // MiniCPM5 wraps values containing <, &, or newlines in CDATA. Extract
+ // the literal content first so the tag strip below does not eat it.
+ const cdataMatch = /^\s*<!\[CDATA\[([\s\S]*?)\]\]>\s*$/.exec(raw);
+ if (cdataMatch) return cdataMatch[1];
+ const cleaned = raw
.replace(/<[^>]+>/g, '')
.trim();
if (!cleaned) return '';
@@ -435,13 +440,36 @@ export function parseToolCallsFromText(text, allowedNames) {
// XML-ish tool-call format used by some local/chat-template models:
// <tool_call><function=click_ax><parameter=ref_id>ref_6</parameter>...
const xmlToolRe = /<tool_call>\s*<function(?:\s*=\s*["']?([A-Za-z_]\w*)["']?|\s+name\s*=\s*["']?([A-Za-z_]\w*)["']?)\s*>\s*([\s\S]*?)\s*<\/function>\s*<\/tool_call>/gi;
+ const xmlToolSpans = [];
let xmlMatch;
while ((xmlMatch = xmlToolRe.exec(text)) !== null) {
+ xmlToolSpans.push({ start: xmlMatch.index, end: xmlMatch.index + xmlMatch[0].length });
const toolName = xmlMatch[1] || xmlMatch[2];
if (!allowedNames.has(toolName)) continue;
const body = xmlMatch[3] || '';
const args = {};
- const paramRe = /<parameter(?:\s*=\s*["']?([A-Za-z_]\w*)["']?|\s+name\s*=\s*["']?([A-Za-z_]\w*)["']?)\s*>\s*([\s\S]*?)\s*<\/parameter>/gi;
+ const paramRe = /<(?:param|parameter)(?:\s*=\s*["']?([A-Za-z_]\w*)["']?|\s+name\s*=\s*["']?([A-Za-z_]\w*)["']?)\s*>\s*([\s\S]*?)\s*<\/(?:param|parameter)>/gi;
+ let paramMatch;
+ while ((paramMatch = paramRe.exec(body)) !== null) {
+ const key = paramMatch[1] || paramMatch[2];
+ if (!key) continue;
+ args[key] = parseXmlParamValue(paramMatch[3]);
+ }
+ results.push({ name: toolName, arguments: args });
+ }
+
+ // MiniCPM5-2B native tool format (no outer <tool_call> wrapper):
+ // <function name="click"><param name="ref_id">ref_6</param>...</function>
+ const minicpmFunctionRe = /<function(?:\s+name\s*=\s*["']([A-Za-z_]\w*)["']|\s*=\s*["']?([A-Za-z_]\w*)["']?)\s*>\s*([\s\S]*?)\s*<\/function>/gi;
+ let minicpmMatch;
+ while ((minicpmMatch = minicpmFunctionRe.exec(text)) !== null) {
+ // Skip functions already consumed inside a <tool_call> wrapper above.
+ if (xmlToolSpans.some(span => minicpmMatch.index >= span.start && minicpmMatch.index < span.end)) continue;
+ const toolName = minicpmMatch[1] || minicpmMatch[2];
+ if (!allowedNames.has(toolName)) continue;
+ const body = minicpmMatch[3] || '';
+ const args = {};
+ const paramRe = /<(?:param|parameter)(?:\s*=\s*["']?([A-Za-z_]\w*)["']?|\s+name\s*=\s*["']?([A-Za-z_]\w*)["']?)\s*>\s*([\s\S]*?)\s*<\/(?:param|parameter)>/gi;
let paramMatch;
while ((paramMatch = paramRe.exec(body)) !== null) {
const key = paramMatch[1] || paramMatch[2];
diff --git a/test/run.js b/test/run.js
index 449984905..1cb1ccade 100644
--- a/test/run.js
+++ b/test/run.js
@@ -975,6 +975,7 @@ const {
WEBGPU_LFM25_VL_16B_MODEL_ID,
WEBGPU_LFM25_VL_3B_MODEL_ID,
WEBGPU_NANBEIGE42_3B_MODEL_ID,
+ WEBGPU_MINICPM5_2B_MODEL_ID,
WEBGPU_BONSAI27_MODEL_ID,
WEBGPU_MODEL_ID,
WEBGPU_MODEL_PRESETS,
@@ -62326,6 +62327,7 @@ test('Chrome exposes separate endpoint-free WebGPU text and vision providers', a
{ id: WEBGPU_LFM25_VL_16B_MODEL_ID, label: 'LFM2.5-VL-1.6B', runtime: 'onnx-vl', contextWindow: 16384, supportsVision: true },
{ id: WEBGPU_LFM25_VL_3B_MODEL_ID, label: 'LFM2.5-VL-3B', runtime: 'onnx-vl', contextWindow: 16384, supportsVision: true },
{ id: WEBGPU_NANBEIGE42_3B_MODEL_ID, label: 'Nanbeige4.2-3B', runtime: 'onnx', contextWindow: 4096, supportsVision: false },
+ { id: WEBGPU_MINICPM5_2B_MODEL_ID, label: 'MiniCPM5-2B', runtime: 'onnx', contextWindow: 16384, supportsVision: false },
{ id: WEBGPU_BONSAI27_MODEL_ID, label: 'Basic text model', runtime: 'bitgpu', contextWindow: 4096, supportsVision: false },
]);
assert.equal(new WebGPUProvider({ model: WEBGPU_BONSAI27_MODEL_ID }).dtype, 'q1');
@@ -63155,6 +63157,7 @@ test('WebGPU worker follows local text-generation and WebBrain VL vision contrac
WEBGPU_LFM25_VL_16B_MODEL_ID,
WEBGPU_LFM25_VL_3B_MODEL_ID,
WEBGPU_NANBEIGE42_3B_MODEL_ID,
+ WEBGPU_MINICPM5_2B_MODEL_ID,
]) {
assert.match(apocalypseHtml, new RegExp(modelId.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')),
`${modelId} is missing from the Apocalypse WebGPU picker`);
@@ -64063,7 +64066,9 @@ test('WebGPU worker replays text tool history and applies model-specific generat
const content = modelId === 'LiquidAI/LFM2.5-2.6B-ONNX'
|| modelId === 'Michionlion/Nanbeige4.2-3B-ONNX-WebGPU'
? 'private model reasoning</think>Hello!'
- : 'text answer';
+ : modelId === 'RASMUS/MiniCPM5-2B-ONNX'
+ ? '<think>private model reasoning</think>Hello!'
+ : 'text answer';
return [{ generated_text: [...input, { role: 'assistant', content }] }];
};
instance.model = {};
@@ -64371,6 +64376,26 @@ test('WebGPU worker replays text tool history and applies model-specific generat
'the chat path must load the same overridden graph file name as the download path',
);
+ const minicpmPayload = {
+ ...textPayload,
+ modelId: WEBGPU_MINICPM5_2B_MODEL_ID,
+ };
+ await dispatch('download-text', minicpmPayload);
+ assert.equal(globalThis.__webgpuPipelineOptions.options.model_file_name, undefined,
+ 'MiniCPM5 publishes the default model_q4f16.onnx graph name');
+ const minicpmResponse = await dispatch('text-chat', minicpmPayload);
+ assert.equal(minicpmResponse.content, 'Hello!');
+ assert.equal(minicpmResponse.reasoningContent, 'private model reasoning',
+ 'MiniCPM5 emits a full <think> wrapper, so the closed-think branch is the reasoning path');
+ assert.deepEqual(globalThis.__webgpuGenerationOptions, {
+ do_sample: true,
+ temperature: 1.0,
+ top_p: 0.95,
+ max_new_tokens: 2048,
+ tools: undefined,
+ tokenizer_encode_kwargs: { preserve_thinking: false },
+ }, 'MiniCPM5 must use its quickstart sampling and the reasoning-template argument');
+
const lfmInstructPayload = {
...textPayload,
modelId: WEBGPU_LFM25_12B_INSTRUCT_MODEL_ID,
@@ -100344,6 +100369,38 @@ test('text tool-call parser is production code with format and allowlist coverag
{ name: 'click_ax', args: { ref_id: 'ref_7' } },
],
},
+ {
+ label: 'MiniCPM5 bare function with param tags',
+ raw: [
+ '<function name="click_ax">',
+ '<param name="ref_id">ref_7</param>',
+ '<param name="force">true</param>',
+ '</function>',
+ ].join(''),
+ expected: [{
+ name: 'click_ax',
+ args: { ref_id: 'ref_7', force: true },
+ }],
+ },
+ {
+ label: 'MiniCPM5 bare function with CDATA value',
+ raw: '<function name="read_page"><param name="text"><![CDATA[Keep <b>this</b> & that]]></param></function>',
+ expected: [{
+ name: 'read_page',
+ args: { text: 'Keep <b>this</b> & that' },
+ }],
+ },
+ {
+ label: 'MiniCPM5 bare functions preserve order alongside wrapper',
+ raw: [
+ '<tool_call>{"name":"read_page","arguments":{}}</tool_call>',
+ '<function name="click_ax"><param name="ref_id">ref_7</param></function>',
+ ].join('\n'),
+ expected: [
+ { name: 'read_page', args: {} },
+ { name: 'click_ax', args: { ref_id: 'ref_7' } },
+ ],
+ },
];
for (const parser of [ToolCallParserCh, ToolCallParserFx]) {