fix cohere & minimize tokenizer requests

src/lib/components/inference-playground/provider-select.svelte

@@ -45,7 +45,7 @@
 		"nebius": "Nebius AI Studio",
 		"hyperbolic": "Hyperbolic",
 		"novita": "Novita",
-		"cohere": "Nohere",
+		"cohere": "Cohere",
 		"hf-inference": "HF Inference API",
 	};
 	const UPPERCASE_WORDS = ["hf", "ai"];

src/lib/components/inference-playground/utils.ts

@@ -180,7 +180,9 @@ export async function handleNonStreamingResponse(
 
 export function isSystemPromptSupported(model: Model | CustomModel) {
 	if (isCustomModel(model)) return true; // OpenAI-compatible models support system messages
-	return model?.config.tokenizer_config?.chat_template?.includes("system");
+	const template = model?.config.tokenizer_config?.chat_template;
+	if (typeof template !== "string") return false;
+	return template.includes("system");
 }
 
 export const defaultSystemMessage: { [key: string]: string } = {
@@ -288,19 +290,25 @@ export function hasInferenceSnippet(
 	return getInferenceSnippet(model, provider, language, "").length > 0;
 }
 
-const tokenizers = new Map<string, PreTrainedTokenizer>();
+const tokenizers = new Map<string, PreTrainedTokenizer | null>();
 
 export async function getTokenizer(model: Model) {
 	if (tokenizers.has(model.id)) return tokenizers.get(model.id)!;
-	const tokenizer = await AutoTokenizer.from_pretrained(model.id);
-	tokenizers.set(model.id, tokenizer);
-	return tokenizer;
+	try {
+		const tokenizer = await AutoTokenizer.from_pretrained(model.id);
+		tokenizers.set(model.id, tokenizer);
+		return tokenizer;
+	} catch {
+		tokenizers.set(model.id, null);
+		return null;
+	}
 }
 
 export async function getTokens(conversation: Conversation): Promise<number> {
 	const model = conversation.model;
 	if (isCustomModel(model)) return 0;
 	const tokenizer = await getTokenizer(model);
+	if (tokenizer === null) return 0;
 
 	// This is a simplified version - you might need to adjust based on your exact needs
 	let formattedText = "";

src/lib/types.ts

@@ -46,7 +46,7 @@ export type Session = {
 };
 
 interface TokenizerConfig {
-	chat_template?: string;
+	chat_template?: string | Array<{ name: string; template: string }>;
 	model_max_length?: number;
 }
 
@@ -156,7 +156,7 @@ export enum UnkTokenEnum {
 }
 
 export type InferenceProviderMapping = {
-	provider: Provider;
+	provider: string;
 	providerId: string;
 	status: Status;
 	task: Task;
@@ -173,6 +173,7 @@ export enum Provider {
 	Replicate = "replicate",
 	Sambanova = "sambanova",
 	Together = "together",
+	Cohere = "cohere",
 }
 
 export enum Status {