@huggingface/tasks 0.20.3 → 0.20.5

package/dist/commonjs/gguf.d.ts

@@ -38,6 +38,7 @@ export declare enum GGMLFileQuantizationType {
     TQ1_0 = 36,
     TQ2_0 = 37,
     MXFP4_MOE = 38,
+    NVFP4 = 39,
     Q2_K_XL = 1000,
     Q3_K_XL = 1001,
     Q4_K_XL = 1002,
@@ -82,6 +83,7 @@ export declare enum GGMLQuantizationType {
     BF16 = 30,
     TQ1_0 = 34,
     TQ2_0 = 35,
-    MXFP4 = 39
+    MXFP4 = 39,
+    NVFP4 = 40
 }
 //# sourceMappingURL=gguf.d.ts.map

package/dist/commonjs/gguf.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"gguf.d.ts","sourceRoot":"","sources":["../../src/gguf.ts"],"names":[],"mappings":"AAGA,oBAAY,wBAAwB;IACnC,GAAG,IAAI;IACP,GAAG,IAAI;IACP,IAAI,IAAI;IACR,IAAI,IAAI;IACR,aAAa,IAAI;IACjB,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,KAAK;IACT,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,IAAI,KAAK;IACT,OAAO,KAAK;IACZ,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,OAAO,KAAK;IACZ,KAAK,KAAK;IACV,MAAM,KAAK;IACX,KAAK,KAAK;IACV,KAAK,KAAK;IACV,KAAK,KAAK;IACV,KAAK,KAAK;IACV,MAAM,KAAK;IACX,KAAK,KAAK;IACV,IAAI,KAAK;IACT,QAAQ,KAAK;IACb,QAAQ,KAAK;IACb,QAAQ,KAAK;IACb,KAAK,KAAK;IACV,KAAK,KAAK;IACV,SAAS,KAAK;IAId,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;CACd;AAGD,eAAO,MAAM,aAAa,QAEzB,CAAC;AACF,eAAO,MAAM,oBAAoB,QAAiC,CAAC;AAEnE,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAGrE;AAKD,eAAO,MAAM,gBAAgB,EAAE,wBAAwB,EA0DtD,CAAC;AAIF,wBAAgB,oBAAoB,CACnC,KAAK,EAAE,wBAAwB,EAC/B,eAAe,EAAE,wBAAwB,EAAE,GACzC,wBAAwB,GAAG,SAAS,CAmCtC;AAGD,oBAAY,oBAAoB;IAC/B,GAAG,IAAI;IACP,GAAG,IAAI;IACP,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,OAAO,KAAK;IACZ,MAAM,KAAK;IACX,OAAO,KAAK;IACZ,KAAK,KAAK;IACV,MAAM,KAAK;IACX,KAAK,KAAK;IACV,KAAK,KAAK;IACV,MAAM,KAAK;IACX,EAAE,KAAK;IACP,GAAG,KAAK;IACR,GAAG,KAAK;IACR,GAAG,KAAK;IACR,GAAG,KAAK;IACR,KAAK,KAAK;IACV,IAAI,KAAK;IACT,KAAK,KAAK;IACV,KAAK,KAAK;IACV,KAAK,KAAK;CACV"}
+{"version":3,"file":"gguf.d.ts","sourceRoot":"","sources":["../../src/gguf.ts"],"names":[],"mappings":"AAGA,oBAAY,wBAAwB;IACnC,GAAG,IAAI;IACP,GAAG,IAAI;IACP,IAAI,IAAI;IACR,IAAI,IAAI;IACR,aAAa,IAAI;IACjB,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,KAAK;IACT,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,IAAI,KAAK;IACT,OAAO,KAAK;IACZ,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,OAAO,KAAK;IACZ,KAAK,KAAK;IACV,MAAM,KAAK;IACX,KAAK,KAAK;IACV,KAAK,KAAK;IACV,KAAK,KAAK;IACV,KAAK,KAAK;IACV,MAAM,KAAK;IACX,KAAK,KAAK;IACV,IAAI,KAAK;IACT,QAAQ,KAAK;IACb,QAAQ,KAAK;IACb,QAAQ,KAAK;IACb,KAAK,KAAK;IACV,KAAK,KAAK;IACV,SAAS,KAAK;IACd,KAAK,KAAK;IAIV,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;CACd;AAGD,eAAO,MAAM,aAAa,QAEzB,CAAC;AACF,eAAO,MAAM,oBAAoB,QAAiC,CAAC;AAEnE,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAGrE;AAKD,eAAO,MAAM,gBAAgB,EAAE,wBAAwB,EA2DtD,CAAC;AAIF,wBAAgB,oBAAoB,CACnC,KAAK,EAAE,wBAAwB,EAC/B,eAAe,EAAE,wBAAwB,EAAE,GACzC,wBAAwB,GAAG,SAAS,CAmCtC;AAGD,oBAAY,oBAAoB;IAC/B,GAAG,IAAI;IACP,GAAG,IAAI;IACP,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,OAAO,KAAK;IACZ,MAAM,KAAK;IACX,OAAO,KAAK;IACZ,KAAK,KAAK;IACV,MAAM,KAAK;IACX,KAAK,KAAK;IACV,KAAK,KAAK;IACV,MAAM,KAAK;IACX,EAAE,KAAK;IACP,GAAG,KAAK;IACR,GAAG,KAAK;IACR,GAAG,KAAK;IACR,GAAG,KAAK;IACR,KAAK,KAAK;IACV,IAAI,KAAK;IACT,KAAK,KAAK;IACV,KAAK,KAAK;IACV,KAAK,KAAK;IACV,KAAK,KAAK;CACV"}

package/dist/commonjs/gguf.js

@@ -47,6 +47,7 @@ var GGMLFileQuantizationType;
     GGMLFileQuantizationType[GGMLFileQuantizationType["TQ1_0"] = 36] = "TQ1_0";
     GGMLFileQuantizationType[GGMLFileQuantizationType["TQ2_0"] = 37] = "TQ2_0";
     GGMLFileQuantizationType[GGMLFileQuantizationType["MXFP4_MOE"] = 38] = "MXFP4_MOE";
+    GGMLFileQuantizationType[GGMLFileQuantizationType["NVFP4"] = 39] = "NVFP4";
     // custom quants used by unsloth
     // they are not officially a scheme enum value in GGUF, but only here for naming
     GGMLFileQuantizationType[GGMLFileQuantizationType["Q2_K_XL"] = 1000] = "Q2_K_XL";
@@ -96,6 +97,7 @@ exports.GGUF_QUANT_ORDER = [
     GGMLFileQuantizationType.Q4_2,
     GGMLFileQuantizationType.Q4_3,
     GGMLFileQuantizationType.MXFP4_MOE,
+    GGMLFileQuantizationType.NVFP4,
     // 3-bit quantizations
     GGMLFileQuantizationType.Q3_K_XL,
     GGMLFileQuantizationType.Q3_K_L,
@@ -187,4 +189,5 @@ var GGMLQuantizationType;
     GGMLQuantizationType[GGMLQuantizationType["TQ1_0"] = 34] = "TQ1_0";
     GGMLQuantizationType[GGMLQuantizationType["TQ2_0"] = 35] = "TQ2_0";
     GGMLQuantizationType[GGMLQuantizationType["MXFP4"] = 39] = "MXFP4";
+    GGMLQuantizationType[GGMLQuantizationType["NVFP4"] = 40] = "NVFP4";
 })(GGMLQuantizationType || (exports.GGMLQuantizationType = GGMLQuantizationType = {}));

package/dist/commonjs/local-apps.d.ts

@@ -56,6 +56,7 @@ declare function isTgiModel(model: ModelData): boolean;
 declare function isLlamaCppGgufModel(model: ModelData): boolean;
 declare function isVllmModel(model: ModelData): boolean;
 declare function isDockerModelRunnerModel(model: ModelData): boolean;
+declare function isUnslothModel(model: ModelData): boolean;
 /**
  * Add your new local app here.
  *
@@ -198,6 +199,13 @@ export declare const LOCAL_APPS: {
         displayOnModelPage: typeof isLlamaCppGgufModel;
         snippet: (model: ModelData, filepath?: string) => string;
     };
+    unsloth: {
+        prettyLabel: string;
+        docsUrl: string;
+        mainTask: "text-generation";
+        displayOnModelPage: typeof isUnslothModel;
+        snippet: (model: ModelData) => LocalAppSnippet[];
+    };
     "docker-model-runner": {
         prettyLabel: string;
         docsUrl: string;

package/dist/commonjs/local-apps.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"local-apps.d.ts","sourceRoot":"","sources":["../../src/local-apps.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;AACjD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAC;AAKnD,MAAM,WAAW,eAAe;IAC/B;;OAEG;IACH,KAAK,EAAE,MAAM,CAAC;IACd;;OAEG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;OAEG;IACH,OAAO,EAAE,MAAM,GAAG,MAAM,EAAE,CAAC;CAC3B;AAED;;GAEG;AACH,MAAM,MAAM,QAAQ,GAAG;IACtB;;OAEG;IACH,WAAW,EAAE,MAAM,CAAC;IACpB;;OAEG;IACH,OAAO,EAAE,MAAM,CAAC;IAChB;;OAEG;IACH,QAAQ,EAAE,YAAY,CAAC;IACvB;;OAEG;IACH,SAAS,CAAC,EAAE,OAAO,CAAC;IAEpB,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB;;OAEG;IACH,kBAAkB,EAAE,CAAC,KAAK,EAAE,SAAS,KAAK,OAAO,CAAC;CAClD,GAAG,CACD;IACA;;OAEG;IACH,QAAQ,EAAE,CAAC,KAAK,EAAE,SAAS,EAAE,QAAQ,CAAC,EAAE,MAAM,KAAK,GAAG,CAAC;CACtD,GACD;IACA;;;;OAIG;IACH,OAAO,EAAE,CAAC,KAAK,EAAE,SAAS,EAAE,QAAQ,CAAC,EAAE,MAAM,KAAK,MAAM,GAAG,MAAM,EAAE,GAAG,eAAe,GAAG,eAAe,EAAE,CAAC;CACzG,CACH,CAAC;AAsBF,iBAAS,UAAU,CAAC,KAAK,EAAE,SAAS,GAAG,OAAO,CAE7C;AAED,iBAAS,mBAAmB,CAAC,KAAK,EAAE,SAAS,WAE5C;AAED,iBAAS,WAAW,CAAC,KAAK,EAAE,SAAS,GAAG,OAAO,CAU9C;AAED,iBAAS,wBAAwB,CAAC,KAAK,EAAE,SAAS,GAAG,OAAO,CAE3D;AAiXD;;;;;;;;;;GAUG;AACH,eAAO,MAAM,UAAU;;;;;;yBAvWS,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;yBAiDzC,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;yBAyC3D,SAAS,KAAG,eAAe,EAAE;;;;;;oCAuS3B,SAAS;yBAjPT,SAAS,KAAG,eAAe,EAAE;;;;;;;yBAoF9B,SAAS,KAAG,eAAe,EAAE;;;;;;;yBA7B/B,SAAS,KAAG,eAAe,EAAE;;;;;;;;;;;;;;yBApIzB,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;yBAJjD,SAAS,aAAa,MAAM,KAAG,MAAM;;;;;;;yBA6O1B,SAAS,aAAa,MAAM,KAAG,MAAM;;;;;;;yBAM9C,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;yBA3CtD,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;CAmRtC,CAAC;AAErC,MAAM,MAAM,WAAW,GAAG,MAAM,OAAO,UAAU,CAAC"}
+{"version":3,"file":"local-apps.d.ts","sourceRoot":"","sources":["../../src/local-apps.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;AACjD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAC;AAKnD,MAAM,WAAW,eAAe;IAC/B;;OAEG;IACH,KAAK,EAAE,MAAM,CAAC;IACd;;OAEG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;OAEG;IACH,OAAO,EAAE,MAAM,GAAG,MAAM,EAAE,CAAC;CAC3B;AAED;;GAEG;AACH,MAAM,MAAM,QAAQ,GAAG;IACtB;;OAEG;IACH,WAAW,EAAE,MAAM,CAAC;IACpB;;OAEG;IACH,OAAO,EAAE,MAAM,CAAC;IAChB;;OAEG;IACH,QAAQ,EAAE,YAAY,CAAC;IACvB;;OAEG;IACH,SAAS,CAAC,EAAE,OAAO,CAAC;IAEpB,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB;;OAEG;IACH,kBAAkB,EAAE,CAAC,KAAK,EAAE,SAAS,KAAK,OAAO,CAAC;CAClD,GAAG,CACD;IACA;;OAEG;IACH,QAAQ,EAAE,CAAC,KAAK,EAAE,SAAS,EAAE,QAAQ,CAAC,EAAE,MAAM,KAAK,GAAG,CAAC;CACtD,GACD;IACA;;;;OAIG;IACH,OAAO,EAAE,CAAC,KAAK,EAAE,SAAS,EAAE,QAAQ,CAAC,EAAE,MAAM,KAAK,MAAM,GAAG,MAAM,EAAE,GAAG,eAAe,GAAG,eAAe,EAAE,CAAC;CACzG,CACH,CAAC;AAsBF,iBAAS,UAAU,CAAC,KAAK,EAAE,SAAS,GAAG,OAAO,CAE7C;AAED,iBAAS,mBAAmB,CAAC,KAAK,EAAE,SAAS,WAE5C;AAED,iBAAS,WAAW,CAAC,KAAK,EAAE,SAAS,GAAG,OAAO,CAU9C;AAED,iBAAS,wBAAwB,CAAC,KAAK,EAAE,SAAS,GAAG,OAAO,CAE3D;AA0BD,iBAAS,cAAc,CAAC,KAAK,EAAE,SAAS,WAEvC;AA6ZD;;;;;;;;;;GAUG;AACH,eAAO,MAAM,UAAU;;;;;;yBA3ZS,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;yBAiDzC,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;yBAmF3D,SAAS,KAAG,eAAe,EAAE;;;;;;oCAiT3B,SAAS;yBA3PT,SAAS,KAAG,eAAe,EAAE;;;;;;;yBAoF9B,SAAS,KAAG,eAAe,EAAE;;;;;;;yBA7B/B,SAAS,KAAG,eAAe,EAAE;;;;;;;;;;;;;;yBApIzB,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;yBA9CjD,SAAS,aAAa,MAAM,KAAG,MAAM;;;;;;;yBAIpC,SAAS,KAAG,eAAe,EAAE;;;;;;;yBA6RnB,SAAS,aAAa,MAAM,KAAG,MAAM;;;;;;;yBAM9C,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;yBArDtD,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;CAuStC,CAAC;AAErC,MAAM,MAAM,WAAW,GAAG,MAAM,OAAO,UAAU,CAAC"}

package/dist/commonjs/local-apps.js

@@ -43,6 +43,23 @@ function isAmdRyzenModel(model) {
 function isMlxModel(model) {
     return model.tags.includes("mlx");
 }
+/**
+ * Returns the model's chat template string, coalescing across sources:
+ * GGUF metadata > chat_template_jinja file > tokenizer_config.json
+ */
+function getChatTemplate(model) {
+    const ct = model.gguf?.chat_template ?? model.config?.chat_template_jinja ?? model.config?.tokenizer_config?.chat_template;
+    if (typeof ct === "string") {
+        return ct;
+    }
+    if (Array.isArray(ct)) {
+        return ct[0]?.template;
+    }
+    return undefined;
+}
+function isUnslothModel(model) {
+    return model.tags.includes("unsloth") || isLlamaCppGgufModel(model);
+}
 function getQuantTag(filepath) {
     const defaultTag = ":{{QUANT_TAG}}";
     if (!filepath) {
@@ -115,6 +132,43 @@ const snippetNodeLlamaCppCli = (model, filepath) => {
 const snippetOllama = (model, filepath) => {
     return `ollama run hf.co/${model.id}${getQuantTag(filepath)}`;
 };
+const snippetUnsloth = (model) => {
+    const isGguf = isLlamaCppGgufModel(model);
+    const studio_instructions = {
+        title: "Open model in Unsloth Studio",
+        setup: ["pip install unsloth", "unsloth studio setup"].join("\n"),
+        content: [
+            "# Run unsloth studio",
+            "unsloth studio -H 0.0.0.0 -p 8000",
+            "# Then open http://localhost:8000/chat in your browser",
+            "# Search for " + model.id + " to start chatting",
+        ].join("\n"),
+    };
+    const hf_spaces_instructions = {
+        title: "Using HuggingFace Spaces for Unsloth",
+        setup: "# No setup required",
+        content: "# Open https://huggingface.co/spaces/unsloth/studio in your browser\n# Search for " +
+            model.id +
+            " to start chatting",
+    };
+    const fastmodel_instructions = {
+        title: "Load model with FastModel",
+        setup: "pip install unsloth",
+        content: [
+            "from unsloth import FastModel",
+            "model, tokenizer = FastModel.from_pretrained(",
+            '    model_name="' + model.id + '",',
+            "    max_seq_length=2048,",
+            ")",
+        ].join("\n"),
+    };
+    if (isGguf) {
+        return [studio_instructions, hf_spaces_instructions];
+    }
+    else {
+        return [studio_instructions, hf_spaces_instructions, fastmodel_instructions];
+    }
+};
 const snippetLocalAI = (model, filepath) => {
     const command = (binary) => ["# Load and run the model:", `${binary} huggingface://${model.id}/${filepath ?? "{{GGUF_FILE}}"}`].join("\n");
     return [
@@ -298,12 +352,25 @@ const snippetMlxLm = (model) => {
     ];
 };
 const snippetPi = (model, filepath) => {
-    const quantTag = getQuantTag(filepath);
     const modelName = model.id.split("/").pop() ?? model.id;
+    const isMLX = isMlxModel(model);
+    // Step 1: Server — differs by backend
+    const serverStep = isMLX
+        ? {
+            title: "Start the MLX server",
+            setup: "# Install MLX LM:\nuv tool install mlx-lm",
+            content: `# Start a local OpenAI-compatible server:\nmlx_lm.server --model "${model.id}"`,
+        }
+        : {
+            title: "Start the llama.cpp server",
+            setup: "# Install llama.cpp:\nbrew install llama.cpp",
+            content: `# Start a local OpenAI-compatible server:\nllama-server -hf ${model.id}${getQuantTag(filepath)} --jinja`,
+        };
+    // Step 2: Pi config — port and provider name differ
     const modelsJson = JSON.stringify({
         providers: {
-            "llama-cpp": {
-                baseUrl: "http://localhost:8080/v1",
+            [isMLX ? "mlx-lm" : "llama-cpp"]: {
+                baseUrl: isMLX ? "http://localhost:8000/v1" : "http://localhost:8080/v1",
                 api: "openai-completions",
                 apiKey: "none",
                 models: [{ id: modelName }],
@@ -311,11 +378,7 @@ const snippetPi = (model, filepath) => {
         },
     }, null, 2);
     return [
-        {
-            title: "Start the llama.cpp server",
-            setup: "# Install llama.cpp:\nbrew install llama.cpp",
-            content: `# Start a local OpenAI-compatible server:\nllama-server -hf ${model.id}${quantTag} --jinja`,
-        },
+        serverStep,
         {
             title: "Configure the model in Pi",
             setup: "# Install Pi:\nnpm install -g @mariozechner/pi-coding-agent",
@@ -323,7 +386,7 @@ const snippetPi = (model, filepath) => {
         },
         {
             title: "Run Pi",
-            content: `# Start Pi in your project directory:\npi`,
+            content: "# Start Pi in your project directory:\npi",
         },
     ];
 };
@@ -539,6 +602,13 @@ exports.LOCAL_APPS = {
         displayOnModelPage: isLlamaCppGgufModel,
         snippet: snippetOllama,
     },
+    unsloth: {
+        prettyLabel: "Unsloth",
+        docsUrl: "https://unsloth.ai/docs",
+        mainTask: "text-generation",
+        displayOnModelPage: isUnslothModel,
+        snippet: snippetUnsloth,
+    },
     "docker-model-runner": {
         prettyLabel: "Docker Model Runner",
         docsUrl: "https://docs.docker.com/ai/model-runner/",
@@ -557,7 +627,9 @@ exports.LOCAL_APPS = {
         prettyLabel: "Pi",
         docsUrl: "https://github.com/badlogic/pi-mono",
         mainTask: "text-generation",
-        displayOnModelPage: (model) => isLlamaCppGgufModel(model) && !!model.gguf?.chat_template?.includes("tools"),
+        displayOnModelPage: (model) => (isLlamaCppGgufModel(model) || isMlxModel(model)) &&
+            model.pipeline_tag === "text-generation" &&
+            !!getChatTemplate(model)?.includes("tools"),
         snippet: snippetPi,
     },
 };

package/dist/commonjs/local-apps.spec.js

@@ -126,6 +126,27 @@ curl -X POST "http://localhost:8000/v1/chat/completions" \\
         (0, vitest_1.expect)(snippet[1].content).toContain(`"id": "Llama-3.2-3B-Instruct-GGUF"`);
         (0, vitest_1.expect)(snippet[2].content).toContain("pi");
     });
+    (0, vitest_1.it)("pi - mlx", async () => {
+        const { snippet: snippetFunc } = local_apps_js_1.LOCAL_APPS["pi"];
+        const model = {
+            id: "mlx-community/Llama-3.2-3B-Instruct-mlx",
+            tags: ["mlx", "conversational"],
+            pipeline_tag: "text-generation",
+            config: {
+                tokenizer_config: {
+                    chat_template: "{% if tools %}...{% endif %}",
+                },
+            },
+            inference: "",
+        };
+        const snippet = snippetFunc(model);
+        (0, vitest_1.expect)(snippet[0].setup).toContain("uv tool install mlx-lm");
+        (0, vitest_1.expect)(snippet[0].content).toContain('mlx_lm.server --model "mlx-community/Llama-3.2-3B-Instruct-mlx"');
+        (0, vitest_1.expect)(snippet[1].setup).toContain("npm install -g @mariozechner/pi-coding-agent");
+        (0, vitest_1.expect)(snippet[1].content).toContain('"baseUrl": "http://localhost:8000/v1"');
+        (0, vitest_1.expect)(snippet[1].content).toContain('"id": "Llama-3.2-3B-Instruct-mlx"');
+        (0, vitest_1.expect)(snippet[2].content).toContain("pi");
+    });
     (0, vitest_1.it)("docker model runner", async () => {
         const { snippet: snippetFunc } = local_apps_js_1.LOCAL_APPS["docker-model-runner"];
         const model = {
@@ -137,4 +158,55 @@ curl -X POST "http://localhost:8000/v1/chat/completions" \\
         const snippet = snippetFunc(model);
         (0, vitest_1.expect)(snippet).toEqual(`docker model run hf.co/bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
     });
+    (0, vitest_1.it)("unsloth tagged model", async () => {
+        const { displayOnModelPage, snippet: snippetFunc } = local_apps_js_1.LOCAL_APPS.unsloth;
+        const model = {
+            id: "some-user/my-unsloth-finetune",
+            tags: ["unsloth", "conversational"],
+            inference: "",
+        };
+        (0, vitest_1.expect)(displayOnModelPage(model)).toBe(true);
+        const snippet = snippetFunc(model);
+        (0, vitest_1.expect)(snippet[0].setup).toBe("pip install unsloth\nunsloth studio setup");
+        (0, vitest_1.expect)(snippet[0].content).toBe("# Run unsloth studio\nunsloth studio -H 0.0.0.0 -p 8000\n# Then open http://localhost:8000/chat in your browser\n# Search for some-user/my-unsloth-finetune to start chatting");
+        (0, vitest_1.expect)(snippet[1].setup).toBe("# No setup required");
+        (0, vitest_1.expect)(snippet[1].content).toBe("# Open https://huggingface.co/spaces/unsloth/studio in your browser\n# Search for some-user/my-unsloth-finetune to start chatting");
+        (0, vitest_1.expect)(snippet[2].setup).toBe("pip install unsloth");
+        (0, vitest_1.expect)(snippet[2].content).toBe('from unsloth import FastModel\nmodel, tokenizer = FastModel.from_pretrained(\n    model_name="some-user/my-unsloth-finetune",\n    max_seq_length=2048,\n)');
+    });
+    (0, vitest_1.it)("unsloth namespace gguf model", async () => {
+        const { displayOnModelPage, snippet: snippetFunc } = local_apps_js_1.LOCAL_APPS.unsloth;
+        const model = {
+            id: "unsloth/Llama-3.2-3B-Instruct-GGUF",
+            tags: ["conversational"],
+            gguf: { total: 1, context_length: 4096 },
+            inference: "",
+        };
+        (0, vitest_1.expect)(displayOnModelPage(model)).toBe(true);
+        const snippet = snippetFunc(model);
+        (0, vitest_1.expect)(snippet[0].setup).toBe("pip install unsloth\nunsloth studio setup");
+        (0, vitest_1.expect)(snippet[0].content).toBe("# Run unsloth studio\nunsloth studio -H 0.0.0.0 -p 8000\n# Then open http://localhost:8000/chat in your browser\n# Search for unsloth/Llama-3.2-3B-Instruct-GGUF to start chatting");
+        (0, vitest_1.expect)(snippet[1].setup).toBe("# No setup required");
+        (0, vitest_1.expect)(snippet[1].content).toBe("# Open https://huggingface.co/spaces/unsloth/studio in your browser\n# Search for unsloth/Llama-3.2-3B-Instruct-GGUF to start chatting");
+        (0, vitest_1.expect)(snippet).toHaveLength(2); // GGUF models only get 2 snippets
+    });
+    (0, vitest_1.it)("non unsloth namespace gguf model", async () => {
+        const { displayOnModelPage } = local_apps_js_1.LOCAL_APPS.unsloth;
+        const model = {
+            id: "dummy/Llama-3.2-3B-Instruct-GGUF",
+            tags: ["conversational"],
+            gguf: { total: 1, context_length: 4096 },
+            inference: "",
+        };
+        (0, vitest_1.expect)(displayOnModelPage(model)).toBe(true);
+    });
+    (0, vitest_1.it)("unsloth not shown for unrelated model", async () => {
+        const { displayOnModelPage } = local_apps_js_1.LOCAL_APPS.unsloth;
+        const model = {
+            id: "meta-llama/Llama-3.2-3B-Instruct",
+            tags: ["conversational"],
+            inference: "",
+        };
+        (0, vitest_1.expect)(displayOnModelPage(model)).toBe(false);
+    });
 });

package/dist/esm/gguf.d.ts

@@ -38,6 +38,7 @@ export declare enum GGMLFileQuantizationType {
     TQ1_0 = 36,
     TQ2_0 = 37,
     MXFP4_MOE = 38,
+    NVFP4 = 39,
     Q2_K_XL = 1000,
     Q3_K_XL = 1001,
     Q4_K_XL = 1002,
@@ -82,6 +83,7 @@ export declare enum GGMLQuantizationType {
     BF16 = 30,
     TQ1_0 = 34,
     TQ2_0 = 35,
-    MXFP4 = 39
+    MXFP4 = 39,
+    NVFP4 = 40
 }
 //# sourceMappingURL=gguf.d.ts.map

package/dist/esm/gguf.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"gguf.d.ts","sourceRoot":"","sources":["../../src/gguf.ts"],"names":[],"mappings":"AAGA,oBAAY,wBAAwB;IACnC,GAAG,IAAI;IACP,GAAG,IAAI;IACP,IAAI,IAAI;IACR,IAAI,IAAI;IACR,aAAa,IAAI;IACjB,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,KAAK;IACT,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,IAAI,KAAK;IACT,OAAO,KAAK;IACZ,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,OAAO,KAAK;IACZ,KAAK,KAAK;IACV,MAAM,KAAK;IACX,KAAK,KAAK;IACV,KAAK,KAAK;IACV,KAAK,KAAK;IACV,KAAK,KAAK;IACV,MAAM,KAAK;IACX,KAAK,KAAK;IACV,IAAI,KAAK;IACT,QAAQ,KAAK;IACb,QAAQ,KAAK;IACb,QAAQ,KAAK;IACb,KAAK,KAAK;IACV,KAAK,KAAK;IACV,SAAS,KAAK;IAId,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;CACd;AAGD,eAAO,MAAM,aAAa,QAEzB,CAAC;AACF,eAAO,MAAM,oBAAoB,QAAiC,CAAC;AAEnE,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAGrE;AAKD,eAAO,MAAM,gBAAgB,EAAE,wBAAwB,EA0DtD,CAAC;AAIF,wBAAgB,oBAAoB,CACnC,KAAK,EAAE,wBAAwB,EAC/B,eAAe,EAAE,wBAAwB,EAAE,GACzC,wBAAwB,GAAG,SAAS,CAmCtC;AAGD,oBAAY,oBAAoB;IAC/B,GAAG,IAAI;IACP,GAAG,IAAI;IACP,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,OAAO,KAAK;IACZ,MAAM,KAAK;IACX,OAAO,KAAK;IACZ,KAAK,KAAK;IACV,MAAM,KAAK;IACX,KAAK,KAAK;IACV,KAAK,KAAK;IACV,MAAM,KAAK;IACX,EAAE,KAAK;IACP,GAAG,KAAK;IACR,GAAG,KAAK;IACR,GAAG,KAAK;IACR,GAAG,KAAK;IACR,KAAK,KAAK;IACV,IAAI,KAAK;IACT,KAAK,KAAK;IACV,KAAK,KAAK;IACV,KAAK,KAAK;CACV"}
+{"version":3,"file":"gguf.d.ts","sourceRoot":"","sources":["../../src/gguf.ts"],"names":[],"mappings":"AAGA,oBAAY,wBAAwB;IACnC,GAAG,IAAI;IACP,GAAG,IAAI;IACP,IAAI,IAAI;IACR,IAAI,IAAI;IACR,aAAa,IAAI;IACjB,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,KAAK;IACT,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,IAAI,KAAK;IACT,OAAO,KAAK;IACZ,MAAM,KAAK;IACX,MAAM,KAAK;IACX,MAAM,KAAK;IACX,OAAO,KAAK;IACZ,KAAK,KAAK;IACV,MAAM,KAAK;IACX,KAAK,KAAK;IACV,KAAK,KAAK;IACV,KAAK,KAAK;IACV,KAAK,KAAK;IACV,MAAM,KAAK;IACX,KAAK,KAAK;IACV,IAAI,KAAK;IACT,QAAQ,KAAK;IACb,QAAQ,KAAK;IACb,QAAQ,KAAK;IACb,KAAK,KAAK;IACV,KAAK,KAAK;IACV,SAAS,KAAK;IACd,KAAK,KAAK;IAIV,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;IACd,OAAO,OAAO;CACd;AAGD,eAAO,MAAM,aAAa,QAEzB,CAAC;AACF,eAAO,MAAM,oBAAoB,QAAiC,CAAC;AAEnE,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAGrE;AAKD,eAAO,MAAM,gBAAgB,EAAE,wBAAwB,EA2DtD,CAAC;AAIF,wBAAgB,oBAAoB,CACnC,KAAK,EAAE,wBAAwB,EAC/B,eAAe,EAAE,wBAAwB,EAAE,GACzC,wBAAwB,GAAG,SAAS,CAmCtC;AAGD,oBAAY,oBAAoB;IAC/B,GAAG,IAAI;IACP,GAAG,IAAI;IACP,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,IAAI;IACR,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,IAAI,KAAK;IACT,OAAO,KAAK;IACZ,MAAM,KAAK;IACX,OAAO,KAAK;IACZ,KAAK,KAAK;IACV,MAAM,KAAK;IACX,KAAK,KAAK;IACV,KAAK,KAAK;IACV,MAAM,KAAK;IACX,EAAE,KAAK;IACP,GAAG,KAAK;IACR,GAAG,KAAK;IACR,GAAG,KAAK;IACR,GAAG,KAAK;IACR,KAAK,KAAK;IACV,IAAI,KAAK;IACT,KAAK,KAAK;IACV,KAAK,KAAK;IACV,KAAK,KAAK;IACV,KAAK,KAAK;CACV"}

package/dist/esm/gguf.js

@@ -42,6 +42,7 @@ export var GGMLFileQuantizationType;
     GGMLFileQuantizationType[GGMLFileQuantizationType["TQ1_0"] = 36] = "TQ1_0";
     GGMLFileQuantizationType[GGMLFileQuantizationType["TQ2_0"] = 37] = "TQ2_0";
     GGMLFileQuantizationType[GGMLFileQuantizationType["MXFP4_MOE"] = 38] = "MXFP4_MOE";
+    GGMLFileQuantizationType[GGMLFileQuantizationType["NVFP4"] = 39] = "NVFP4";
     // custom quants used by unsloth
     // they are not officially a scheme enum value in GGUF, but only here for naming
     GGMLFileQuantizationType[GGMLFileQuantizationType["Q2_K_XL"] = 1000] = "Q2_K_XL";
@@ -91,6 +92,7 @@ export const GGUF_QUANT_ORDER = [
     GGMLFileQuantizationType.Q4_2,
     GGMLFileQuantizationType.Q4_3,
     GGMLFileQuantizationType.MXFP4_MOE,
+    GGMLFileQuantizationType.NVFP4,
     // 3-bit quantizations
     GGMLFileQuantizationType.Q3_K_XL,
     GGMLFileQuantizationType.Q3_K_L,
@@ -182,4 +184,5 @@ export var GGMLQuantizationType;
     GGMLQuantizationType[GGMLQuantizationType["TQ1_0"] = 34] = "TQ1_0";
     GGMLQuantizationType[GGMLQuantizationType["TQ2_0"] = 35] = "TQ2_0";
     GGMLQuantizationType[GGMLQuantizationType["MXFP4"] = 39] = "MXFP4";
+    GGMLQuantizationType[GGMLQuantizationType["NVFP4"] = 40] = "NVFP4";
 })(GGMLQuantizationType || (GGMLQuantizationType = {}));

package/dist/esm/local-apps.d.ts

@@ -56,6 +56,7 @@ declare function isTgiModel(model: ModelData): boolean;
 declare function isLlamaCppGgufModel(model: ModelData): boolean;
 declare function isVllmModel(model: ModelData): boolean;
 declare function isDockerModelRunnerModel(model: ModelData): boolean;
+declare function isUnslothModel(model: ModelData): boolean;
 /**
  * Add your new local app here.
  *
@@ -198,6 +199,13 @@ export declare const LOCAL_APPS: {
         displayOnModelPage: typeof isLlamaCppGgufModel;
         snippet: (model: ModelData, filepath?: string) => string;
     };
+    unsloth: {
+        prettyLabel: string;
+        docsUrl: string;
+        mainTask: "text-generation";
+        displayOnModelPage: typeof isUnslothModel;
+        snippet: (model: ModelData) => LocalAppSnippet[];
+    };
     "docker-model-runner": {
         prettyLabel: string;
         docsUrl: string;

package/dist/esm/local-apps.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"local-apps.d.ts","sourceRoot":"","sources":["../../src/local-apps.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;AACjD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAC;AAKnD,MAAM,WAAW,eAAe;IAC/B;;OAEG;IACH,KAAK,EAAE,MAAM,CAAC;IACd;;OAEG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;OAEG;IACH,OAAO,EAAE,MAAM,GAAG,MAAM,EAAE,CAAC;CAC3B;AAED;;GAEG;AACH,MAAM,MAAM,QAAQ,GAAG;IACtB;;OAEG;IACH,WAAW,EAAE,MAAM,CAAC;IACpB;;OAEG;IACH,OAAO,EAAE,MAAM,CAAC;IAChB;;OAEG;IACH,QAAQ,EAAE,YAAY,CAAC;IACvB;;OAEG;IACH,SAAS,CAAC,EAAE,OAAO,CAAC;IAEpB,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB;;OAEG;IACH,kBAAkB,EAAE,CAAC,KAAK,EAAE,SAAS,KAAK,OAAO,CAAC;CAClD,GAAG,CACD;IACA;;OAEG;IACH,QAAQ,EAAE,CAAC,KAAK,EAAE,SAAS,EAAE,QAAQ,CAAC,EAAE,MAAM,KAAK,GAAG,CAAC;CACtD,GACD;IACA;;;;OAIG;IACH,OAAO,EAAE,CAAC,KAAK,EAAE,SAAS,EAAE,QAAQ,CAAC,EAAE,MAAM,KAAK,MAAM,GAAG,MAAM,EAAE,GAAG,eAAe,GAAG,eAAe,EAAE,CAAC;CACzG,CACH,CAAC;AAsBF,iBAAS,UAAU,CAAC,KAAK,EAAE,SAAS,GAAG,OAAO,CAE7C;AAED,iBAAS,mBAAmB,CAAC,KAAK,EAAE,SAAS,WAE5C;AAED,iBAAS,WAAW,CAAC,KAAK,EAAE,SAAS,GAAG,OAAO,CAU9C;AAED,iBAAS,wBAAwB,CAAC,KAAK,EAAE,SAAS,GAAG,OAAO,CAE3D;AAiXD;;;;;;;;;;GAUG;AACH,eAAO,MAAM,UAAU;;;;;;yBAvWS,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;yBAiDzC,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;yBAyC3D,SAAS,KAAG,eAAe,EAAE;;;;;;oCAuS3B,SAAS;yBAjPT,SAAS,KAAG,eAAe,EAAE;;;;;;;yBAoF9B,SAAS,KAAG,eAAe,EAAE;;;;;;;yBA7B/B,SAAS,KAAG,eAAe,EAAE;;;;;;;;;;;;;;yBApIzB,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;yBAJjD,SAAS,aAAa,MAAM,KAAG,MAAM;;;;;;;yBA6O1B,SAAS,aAAa,MAAM,KAAG,MAAM;;;;;;;yBAM9C,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;yBA3CtD,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;CAmRtC,CAAC;AAErC,MAAM,MAAM,WAAW,GAAG,MAAM,OAAO,UAAU,CAAC"}
+{"version":3,"file":"local-apps.d.ts","sourceRoot":"","sources":["../../src/local-apps.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;AACjD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAC;AAKnD,MAAM,WAAW,eAAe;IAC/B;;OAEG;IACH,KAAK,EAAE,MAAM,CAAC;IACd;;OAEG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;OAEG;IACH,OAAO,EAAE,MAAM,GAAG,MAAM,EAAE,CAAC;CAC3B;AAED;;GAEG;AACH,MAAM,MAAM,QAAQ,GAAG;IACtB;;OAEG;IACH,WAAW,EAAE,MAAM,CAAC;IACpB;;OAEG;IACH,OAAO,EAAE,MAAM,CAAC;IAChB;;OAEG;IACH,QAAQ,EAAE,YAAY,CAAC;IACvB;;OAEG;IACH,SAAS,CAAC,EAAE,OAAO,CAAC;IAEpB,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB;;OAEG;IACH,kBAAkB,EAAE,CAAC,KAAK,EAAE,SAAS,KAAK,OAAO,CAAC;CAClD,GAAG,CACD;IACA;;OAEG;IACH,QAAQ,EAAE,CAAC,KAAK,EAAE,SAAS,EAAE,QAAQ,CAAC,EAAE,MAAM,KAAK,GAAG,CAAC;CACtD,GACD;IACA;;;;OAIG;IACH,OAAO,EAAE,CAAC,KAAK,EAAE,SAAS,EAAE,QAAQ,CAAC,EAAE,MAAM,KAAK,MAAM,GAAG,MAAM,EAAE,GAAG,eAAe,GAAG,eAAe,EAAE,CAAC;CACzG,CACH,CAAC;AAsBF,iBAAS,UAAU,CAAC,KAAK,EAAE,SAAS,GAAG,OAAO,CAE7C;AAED,iBAAS,mBAAmB,CAAC,KAAK,EAAE,SAAS,WAE5C;AAED,iBAAS,WAAW,CAAC,KAAK,EAAE,SAAS,GAAG,OAAO,CAU9C;AAED,iBAAS,wBAAwB,CAAC,KAAK,EAAE,SAAS,GAAG,OAAO,CAE3D;AA0BD,iBAAS,cAAc,CAAC,KAAK,EAAE,SAAS,WAEvC;AA6ZD;;;;;;;;;;GAUG;AACH,eAAO,MAAM,UAAU;;;;;;yBA3ZS,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;yBAiDzC,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;yBAmF3D,SAAS,KAAG,eAAe,EAAE;;;;;;oCAiT3B,SAAS;yBA3PT,SAAS,KAAG,eAAe,EAAE;;;;;;;yBAoF9B,SAAS,KAAG,eAAe,EAAE;;;;;;;yBA7B/B,SAAS,KAAG,eAAe,EAAE;;;;;;;;;;;;;;yBApIzB,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;yBA9CjD,SAAS,aAAa,MAAM,KAAG,MAAM;;;;;;;yBAIpC,SAAS,KAAG,eAAe,EAAE;;;;;;;yBA6RnB,SAAS,aAAa,MAAM,KAAG,MAAM;;;;;;;yBAM9C,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;;;;;;yBArDtD,SAAS,aAAa,MAAM,KAAG,eAAe,EAAE;;CAuStC,CAAC;AAErC,MAAM,MAAM,WAAW,GAAG,MAAM,OAAO,UAAU,CAAC"}

package/dist/esm/local-apps.js

@@ -40,6 +40,23 @@ function isAmdRyzenModel(model) {
 function isMlxModel(model) {
     return model.tags.includes("mlx");
 }
+/**
+ * Returns the model's chat template string, coalescing across sources:
+ * GGUF metadata > chat_template_jinja file > tokenizer_config.json
+ */
+function getChatTemplate(model) {
+    const ct = model.gguf?.chat_template ?? model.config?.chat_template_jinja ?? model.config?.tokenizer_config?.chat_template;
+    if (typeof ct === "string") {
+        return ct;
+    }
+    if (Array.isArray(ct)) {
+        return ct[0]?.template;
+    }
+    return undefined;
+}
+function isUnslothModel(model) {
+    return model.tags.includes("unsloth") || isLlamaCppGgufModel(model);
+}
 function getQuantTag(filepath) {
     const defaultTag = ":{{QUANT_TAG}}";
     if (!filepath) {
@@ -112,6 +129,43 @@ const snippetNodeLlamaCppCli = (model, filepath) => {
 const snippetOllama = (model, filepath) => {
     return `ollama run hf.co/${model.id}${getQuantTag(filepath)}`;
 };
+const snippetUnsloth = (model) => {
+    const isGguf = isLlamaCppGgufModel(model);
+    const studio_instructions = {
+        title: "Open model in Unsloth Studio",
+        setup: ["pip install unsloth", "unsloth studio setup"].join("\n"),
+        content: [
+            "# Run unsloth studio",
+            "unsloth studio -H 0.0.0.0 -p 8000",
+            "# Then open http://localhost:8000/chat in your browser",
+            "# Search for " + model.id + " to start chatting",
+        ].join("\n"),
+    };
+    const hf_spaces_instructions = {
+        title: "Using HuggingFace Spaces for Unsloth",
+        setup: "# No setup required",
+        content: "# Open https://huggingface.co/spaces/unsloth/studio in your browser\n# Search for " +
+            model.id +
+            " to start chatting",
+    };
+    const fastmodel_instructions = {
+        title: "Load model with FastModel",
+        setup: "pip install unsloth",
+        content: [
+            "from unsloth import FastModel",
+            "model, tokenizer = FastModel.from_pretrained(",
+            '    model_name="' + model.id + '",',
+            "    max_seq_length=2048,",
+            ")",
+        ].join("\n"),
+    };
+    if (isGguf) {
+        return [studio_instructions, hf_spaces_instructions];
+    }
+    else {
+        return [studio_instructions, hf_spaces_instructions, fastmodel_instructions];
+    }
+};
 const snippetLocalAI = (model, filepath) => {
     const command = (binary) => ["# Load and run the model:", `${binary} huggingface://${model.id}/${filepath ?? "{{GGUF_FILE}}"}`].join("\n");
     return [
@@ -295,12 +349,25 @@ const snippetMlxLm = (model) => {
     ];
 };
 const snippetPi = (model, filepath) => {
-    const quantTag = getQuantTag(filepath);
     const modelName = model.id.split("/").pop() ?? model.id;
+    const isMLX = isMlxModel(model);
+    // Step 1: Server — differs by backend
+    const serverStep = isMLX
+        ? {
+            title: "Start the MLX server",
+            setup: "# Install MLX LM:\nuv tool install mlx-lm",
+            content: `# Start a local OpenAI-compatible server:\nmlx_lm.server --model "${model.id}"`,
+        }
+        : {
+            title: "Start the llama.cpp server",
+            setup: "# Install llama.cpp:\nbrew install llama.cpp",
+            content: `# Start a local OpenAI-compatible server:\nllama-server -hf ${model.id}${getQuantTag(filepath)} --jinja`,
+        };
+    // Step 2: Pi config — port and provider name differ
     const modelsJson = JSON.stringify({
         providers: {
-            "llama-cpp": {
-                baseUrl: "http://localhost:8080/v1",
+            [isMLX ? "mlx-lm" : "llama-cpp"]: {
+                baseUrl: isMLX ? "http://localhost:8000/v1" : "http://localhost:8080/v1",
                 api: "openai-completions",
                 apiKey: "none",
                 models: [{ id: modelName }],
@@ -308,11 +375,7 @@ const snippetPi = (model, filepath) => {
         },
     }, null, 2);
     return [
-        {
-            title: "Start the llama.cpp server",
-            setup: "# Install llama.cpp:\nbrew install llama.cpp",
-            content: `# Start a local OpenAI-compatible server:\nllama-server -hf ${model.id}${quantTag} --jinja`,
-        },
+        serverStep,
         {
             title: "Configure the model in Pi",
             setup: "# Install Pi:\nnpm install -g @mariozechner/pi-coding-agent",
@@ -320,7 +383,7 @@ const snippetPi = (model, filepath) => {
         },
         {
             title: "Run Pi",
-            content: `# Start Pi in your project directory:\npi`,
+            content: "# Start Pi in your project directory:\npi",
         },
     ];
 };
@@ -536,6 +599,13 @@ export const LOCAL_APPS = {
         displayOnModelPage: isLlamaCppGgufModel,
         snippet: snippetOllama,
     },
+    unsloth: {
+        prettyLabel: "Unsloth",
+        docsUrl: "https://unsloth.ai/docs",
+        mainTask: "text-generation",
+        displayOnModelPage: isUnslothModel,
+        snippet: snippetUnsloth,
+    },
     "docker-model-runner": {
         prettyLabel: "Docker Model Runner",
         docsUrl: "https://docs.docker.com/ai/model-runner/",
@@ -554,7 +624,9 @@ export const LOCAL_APPS = {
         prettyLabel: "Pi",
         docsUrl: "https://github.com/badlogic/pi-mono",
         mainTask: "text-generation",
-        displayOnModelPage: (model) => isLlamaCppGgufModel(model) && !!model.gguf?.chat_template?.includes("tools"),
+        displayOnModelPage: (model) => (isLlamaCppGgufModel(model) || isMlxModel(model)) &&
+            model.pipeline_tag === "text-generation" &&
+            !!getChatTemplate(model)?.includes("tools"),
         snippet: snippetPi,
     },
 };

package/dist/esm/local-apps.spec.js

@@ -124,6 +124,27 @@ curl -X POST "http://localhost:8000/v1/chat/completions" \\
         expect(snippet[1].content).toContain(`"id": "Llama-3.2-3B-Instruct-GGUF"`);
         expect(snippet[2].content).toContain("pi");
     });
+    it("pi - mlx", async () => {
+        const { snippet: snippetFunc } = LOCAL_APPS["pi"];
+        const model = {
+            id: "mlx-community/Llama-3.2-3B-Instruct-mlx",
+            tags: ["mlx", "conversational"],
+            pipeline_tag: "text-generation",
+            config: {
+                tokenizer_config: {
+                    chat_template: "{% if tools %}...{% endif %}",
+                },
+            },
+            inference: "",
+        };
+        const snippet = snippetFunc(model);
+        expect(snippet[0].setup).toContain("uv tool install mlx-lm");
+        expect(snippet[0].content).toContain('mlx_lm.server --model "mlx-community/Llama-3.2-3B-Instruct-mlx"');
+        expect(snippet[1].setup).toContain("npm install -g @mariozechner/pi-coding-agent");
+        expect(snippet[1].content).toContain('"baseUrl": "http://localhost:8000/v1"');
+        expect(snippet[1].content).toContain('"id": "Llama-3.2-3B-Instruct-mlx"');
+        expect(snippet[2].content).toContain("pi");
+    });
     it("docker model runner", async () => {
         const { snippet: snippetFunc } = LOCAL_APPS["docker-model-runner"];
         const model = {
@@ -135,4 +156,55 @@ curl -X POST "http://localhost:8000/v1/chat/completions" \\
         const snippet = snippetFunc(model);
         expect(snippet).toEqual(`docker model run hf.co/bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
     });
+    it("unsloth tagged model", async () => {
+        const { displayOnModelPage, snippet: snippetFunc } = LOCAL_APPS.unsloth;
+        const model = {
+            id: "some-user/my-unsloth-finetune",
+            tags: ["unsloth", "conversational"],
+            inference: "",
+        };
+        expect(displayOnModelPage(model)).toBe(true);
+        const snippet = snippetFunc(model);
+        expect(snippet[0].setup).toBe("pip install unsloth\nunsloth studio setup");
+        expect(snippet[0].content).toBe("# Run unsloth studio\nunsloth studio -H 0.0.0.0 -p 8000\n# Then open http://localhost:8000/chat in your browser\n# Search for some-user/my-unsloth-finetune to start chatting");
+        expect(snippet[1].setup).toBe("# No setup required");
+        expect(snippet[1].content).toBe("# Open https://huggingface.co/spaces/unsloth/studio in your browser\n# Search for some-user/my-unsloth-finetune to start chatting");
+        expect(snippet[2].setup).toBe("pip install unsloth");
+        expect(snippet[2].content).toBe('from unsloth import FastModel\nmodel, tokenizer = FastModel.from_pretrained(\n    model_name="some-user/my-unsloth-finetune",\n    max_seq_length=2048,\n)');
+    });
+    it("unsloth namespace gguf model", async () => {
+        const { displayOnModelPage, snippet: snippetFunc } = LOCAL_APPS.unsloth;
+        const model = {
+            id: "unsloth/Llama-3.2-3B-Instruct-GGUF",
+            tags: ["conversational"],
+            gguf: { total: 1, context_length: 4096 },
+            inference: "",
+        };
+        expect(displayOnModelPage(model)).toBe(true);
+        const snippet = snippetFunc(model);
+        expect(snippet[0].setup).toBe("pip install unsloth\nunsloth studio setup");
+        expect(snippet[0].content).toBe("# Run unsloth studio\nunsloth studio -H 0.0.0.0 -p 8000\n# Then open http://localhost:8000/chat in your browser\n# Search for unsloth/Llama-3.2-3B-Instruct-GGUF to start chatting");
+        expect(snippet[1].setup).toBe("# No setup required");
+        expect(snippet[1].content).toBe("# Open https://huggingface.co/spaces/unsloth/studio in your browser\n# Search for unsloth/Llama-3.2-3B-Instruct-GGUF to start chatting");
+        expect(snippet).toHaveLength(2); // GGUF models only get 2 snippets
+    });
+    it("non unsloth namespace gguf model", async () => {
+        const { displayOnModelPage } = LOCAL_APPS.unsloth;
+        const model = {
+            id: "dummy/Llama-3.2-3B-Instruct-GGUF",
+            tags: ["conversational"],
+            gguf: { total: 1, context_length: 4096 },
+            inference: "",
+        };
+        expect(displayOnModelPage(model)).toBe(true);
+    });
+    it("unsloth not shown for unrelated model", async () => {
+        const { displayOnModelPage } = LOCAL_APPS.unsloth;
+        const model = {
+            id: "meta-llama/Llama-3.2-3B-Instruct",
+            tags: ["conversational"],
+            inference: "",
+        };
+        expect(displayOnModelPage(model)).toBe(false);
+    });
 });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@huggingface/tasks",
-  "version": "0.20.3",
+  "version": "0.20.5",
   "description": "List of ML tasks for huggingface.co/tasks",
   "keywords": [
     "hub",

package/src/gguf.ts

@@ -41,6 +41,7 @@ export enum GGMLFileQuantizationType {
 	TQ1_0 = 36,
 	TQ2_0 = 37,
 	MXFP4_MOE = 38,
+	NVFP4 = 39,
 
 	// custom quants used by unsloth
 	// they are not officially a scheme enum value in GGUF, but only here for naming
@@ -99,6 +100,7 @@ export const GGUF_QUANT_ORDER: GGMLFileQuantizationType[] = [
 	GGMLFileQuantizationType.Q4_2,
 	GGMLFileQuantizationType.Q4_3,
 	GGMLFileQuantizationType.MXFP4_MOE,
+	GGMLFileQuantizationType.NVFP4,
 
 	// 3-bit quantizations
 	GGMLFileQuantizationType.Q3_K_XL,
@@ -202,4 +204,5 @@ export enum GGMLQuantizationType {
 	TQ1_0 = 34,
 	TQ2_0 = 35,
 	MXFP4 = 39,
+	NVFP4 = 40,
 }

package/src/local-apps.spec.ts

@@ -138,6 +138,29 @@ curl -X POST "http://localhost:8000/v1/chat/completions" \\
 		expect(snippet[2].content).toContain("pi");
 	});
 
+	it("pi - mlx", async () => {
+		const { snippet: snippetFunc } = LOCAL_APPS["pi"];
+		const model: ModelData = {
+			id: "mlx-community/Llama-3.2-3B-Instruct-mlx",
+			tags: ["mlx", "conversational"],
+			pipeline_tag: "text-generation",
+			config: {
+				tokenizer_config: {
+					chat_template: "{% if tools %}...{% endif %}",
+				},
+			},
+			inference: "",
+		};
+		const snippet = snippetFunc(model);
+
+		expect(snippet[0].setup).toContain("uv tool install mlx-lm");
+		expect(snippet[0].content).toContain('mlx_lm.server --model "mlx-community/Llama-3.2-3B-Instruct-mlx"');
+		expect(snippet[1].setup).toContain("npm install -g @mariozechner/pi-coding-agent");
+		expect(snippet[1].content).toContain('"baseUrl": "http://localhost:8000/v1"');
+		expect(snippet[1].content).toContain('"id": "Llama-3.2-3B-Instruct-mlx"');
+		expect(snippet[2].content).toContain("pi");
+	});
+
 	it("docker model runner", async () => {
 		const { snippet: snippetFunc } = LOCAL_APPS["docker-model-runner"];
 		const model: ModelData = {
@@ -150,4 +173,73 @@ curl -X POST "http://localhost:8000/v1/chat/completions" \\
 
 		expect(snippet).toEqual(`docker model run hf.co/bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
 	});
+
+	it("unsloth tagged model", async () => {
+		const { displayOnModelPage, snippet: snippetFunc } = LOCAL_APPS.unsloth;
+		const model: ModelData = {
+			id: "some-user/my-unsloth-finetune",
+			tags: ["unsloth", "conversational"],
+			inference: "",
+		};
+
+		expect(displayOnModelPage(model)).toBe(true);
+		const snippet = snippetFunc(model);
+		expect(snippet[0].setup).toBe("pip install unsloth\nunsloth studio setup");
+		expect(snippet[0].content).toBe(
+			"# Run unsloth studio\nunsloth studio -H 0.0.0.0 -p 8000\n# Then open http://localhost:8000/chat in your browser\n# Search for some-user/my-unsloth-finetune to start chatting",
+		);
+		expect(snippet[1].setup).toBe("# No setup required");
+		expect(snippet[1].content).toBe(
+			"# Open https://huggingface.co/spaces/unsloth/studio in your browser\n# Search for some-user/my-unsloth-finetune to start chatting",
+		);
+		expect(snippet[2].setup).toBe("pip install unsloth");
+		expect(snippet[2].content).toBe(
+			'from unsloth import FastModel\nmodel, tokenizer = FastModel.from_pretrained(\n    model_name="some-user/my-unsloth-finetune",\n    max_seq_length=2048,\n)',
+		);
+	});
+
+	it("unsloth namespace gguf model", async () => {
+		const { displayOnModelPage, snippet: snippetFunc } = LOCAL_APPS.unsloth;
+		const model: ModelData = {
+			id: "unsloth/Llama-3.2-3B-Instruct-GGUF",
+			tags: ["conversational"],
+			gguf: { total: 1, context_length: 4096 },
+			inference: "",
+		};
+
+		expect(displayOnModelPage(model)).toBe(true);
+		const snippet = snippetFunc(model);
+		expect(snippet[0].setup).toBe("pip install unsloth\nunsloth studio setup");
+		expect(snippet[0].content).toBe(
+			"# Run unsloth studio\nunsloth studio -H 0.0.0.0 -p 8000\n# Then open http://localhost:8000/chat in your browser\n# Search for unsloth/Llama-3.2-3B-Instruct-GGUF to start chatting",
+		);
+		expect(snippet[1].setup).toBe("# No setup required");
+		expect(snippet[1].content).toBe(
+			"# Open https://huggingface.co/spaces/unsloth/studio in your browser\n# Search for unsloth/Llama-3.2-3B-Instruct-GGUF to start chatting",
+		);
+		expect(snippet).toHaveLength(2); // GGUF models only get 2 snippets
+	});
+
+	it("non unsloth namespace gguf model", async () => {
+		const { displayOnModelPage } = LOCAL_APPS.unsloth;
+		const model: ModelData = {
+			id: "dummy/Llama-3.2-3B-Instruct-GGUF",
+			tags: ["conversational"],
+			gguf: { total: 1, context_length: 4096 },
+			inference: "",
+		};
+
+		expect(displayOnModelPage(model)).toBe(true);
+	});
+
+	it("unsloth not shown for unrelated model", async () => {
+		const { displayOnModelPage } = LOCAL_APPS.unsloth;
+		const model: ModelData = {
+			id: "meta-llama/Llama-3.2-3B-Instruct",
+			tags: ["conversational"],
+			inference: "",
+		};
+
+		expect(displayOnModelPage(model)).toBe(false);
+	});
 });

package/src/local-apps.ts

@@ -115,6 +115,26 @@ function isMlxModel(model: ModelData) {
 	return model.tags.includes("mlx");
 }
 
+/**
+ * Returns the model's chat template string, coalescing across sources:
+ * GGUF metadata > chat_template_jinja file > tokenizer_config.json
+ */
+function getChatTemplate(model: ModelData): string | undefined {
+	const ct =
+		model.gguf?.chat_template ?? model.config?.chat_template_jinja ?? model.config?.tokenizer_config?.chat_template;
+	if (typeof ct === "string") {
+		return ct;
+	}
+	if (Array.isArray(ct)) {
+		return ct[0]?.template;
+	}
+	return undefined;
+}
+
+function isUnslothModel(model: ModelData) {
+	return model.tags.includes("unsloth") || isLlamaCppGgufModel(model);
+}
+
 function getQuantTag(filepath?: string): string {
 	const defaultTag = ":{{QUANT_TAG}}";
 
@@ -193,6 +213,48 @@ const snippetOllama = (model: ModelData, filepath?: string): string => {
 	return `ollama run hf.co/${model.id}${getQuantTag(filepath)}`;
 };
 
+const snippetUnsloth = (model: ModelData): LocalAppSnippet[] => {
+	const isGguf = isLlamaCppGgufModel(model);
+
+	const studio_instructions: LocalAppSnippet = {
+		title: "Open model in Unsloth Studio",
+		setup: ["pip install unsloth", "unsloth studio setup"].join("\n"),
+		content: [
+			"# Run unsloth studio",
+			"unsloth studio -H 0.0.0.0 -p 8000",
+			"# Then open http://localhost:8000/chat in your browser",
+			"# Search for " + model.id + " to start chatting",
+		].join("\n"),
+	};
+
+	const hf_spaces_instructions: LocalAppSnippet = {
+		title: "Using HuggingFace Spaces for Unsloth",
+		setup: "# No setup required",
+		content:
+			"# Open https://huggingface.co/spaces/unsloth/studio in your browser\n# Search for " +
+			model.id +
+			" to start chatting",
+	};
+
+	const fastmodel_instructions: LocalAppSnippet = {
+		title: "Load model with FastModel",
+		setup: "pip install unsloth",
+		content: [
+			"from unsloth import FastModel",
+			"model, tokenizer = FastModel.from_pretrained(",
+			'    model_name="' + model.id + '",',
+			"    max_seq_length=2048,",
+			")",
+		].join("\n"),
+	};
+
+	if (isGguf) {
+		return [studio_instructions, hf_spaces_instructions];
+	} else {
+		return [studio_instructions, hf_spaces_instructions, fastmodel_instructions];
+	}
+};
+
 const snippetLocalAI = (model: ModelData, filepath?: string): LocalAppSnippet[] => {
 	const command = (binary: string) =>
 		["# Load and run the model:", `${binary} huggingface://${model.id}/${filepath ?? "{{GGUF_FILE}}"}`].join("\n");
@@ -390,14 +452,28 @@ const snippetMlxLm = (model: ModelData): LocalAppSnippet[] => {
 };
 
 const snippetPi = (model: ModelData, filepath?: string): LocalAppSnippet[] => {
-	const quantTag = getQuantTag(filepath);
 	const modelName = model.id.split("/").pop() ?? model.id;
+	const isMLX = isMlxModel(model);
+
+	// Step 1: Server — differs by backend
+	const serverStep: LocalAppSnippet = isMLX
+		? {
+				title: "Start the MLX server",
+				setup: "# Install MLX LM:\nuv tool install mlx-lm",
+				content: `# Start a local OpenAI-compatible server:\nmlx_lm.server --model "${model.id}"`,
+			}
+		: {
+				title: "Start the llama.cpp server",
+				setup: "# Install llama.cpp:\nbrew install llama.cpp",
+				content: `# Start a local OpenAI-compatible server:\nllama-server -hf ${model.id}${getQuantTag(filepath)} --jinja`,
+			};
 
+	// Step 2: Pi config — port and provider name differ
 	const modelsJson = JSON.stringify(
 		{
 			providers: {
-				"llama-cpp": {
-					baseUrl: "http://localhost:8080/v1",
+				[isMLX ? "mlx-lm" : "llama-cpp"]: {
+					baseUrl: isMLX ? "http://localhost:8000/v1" : "http://localhost:8080/v1",
 					api: "openai-completions",
 					apiKey: "none",
 					models: [{ id: modelName }],
@@ -409,11 +485,7 @@ const snippetPi = (model: ModelData, filepath?: string): LocalAppSnippet[] => {
 	);
 
 	return [
-		{
-			title: "Start the llama.cpp server",
-			setup: "# Install llama.cpp:\nbrew install llama.cpp",
-			content: `# Start a local OpenAI-compatible server:\nllama-server -hf ${model.id}${quantTag} --jinja`,
-		},
+		serverStep,
 		{
 			title: "Configure the model in Pi",
 			setup: "# Install Pi:\nnpm install -g @mariozechner/pi-coding-agent",
@@ -421,7 +493,7 @@ const snippetPi = (model: ModelData, filepath?: string): LocalAppSnippet[] => {
 		},
 		{
 			title: "Run Pi",
-			content: `# Start Pi in your project directory:\npi`,
+			content: "# Start Pi in your project directory:\npi",
 		},
 	];
 };
@@ -643,6 +715,13 @@ export const LOCAL_APPS = {
 		displayOnModelPage: isLlamaCppGgufModel,
 		snippet: snippetOllama,
 	},
+	unsloth: {
+		prettyLabel: "Unsloth",
+		docsUrl: "https://unsloth.ai/docs",
+		mainTask: "text-generation",
+		displayOnModelPage: isUnslothModel,
+		snippet: snippetUnsloth,
+	},
 	"docker-model-runner": {
 		prettyLabel: "Docker Model Runner",
 		docsUrl: "https://docs.docker.com/ai/model-runner/",
@@ -661,7 +740,10 @@ export const LOCAL_APPS = {
 		prettyLabel: "Pi",
 		docsUrl: "https://github.com/badlogic/pi-mono",
 		mainTask: "text-generation",
-		displayOnModelPage: (model) => isLlamaCppGgufModel(model) && !!model.gguf?.chat_template?.includes("tools"),
+		displayOnModelPage: (model) =>
+			(isLlamaCppGgufModel(model) || isMlxModel(model)) &&
+			model.pipeline_tag === "text-generation" &&
+			!!getChatTemplate(model)?.includes("tools"),
 		snippet: snippetPi,
 	},
 } satisfies Record<string, LocalApp>;