Upload mamba_runtime.js with huggingface_hub

mamba_runtime.js

@@ -757,11 +757,11 @@ class MambaRuntime {
   }
 
   // ── Tokenize/detokenize via server ──────────────────────────────────────
-  async tokenize(text) {
+  async tokenize(text, system = '') {
     const resp = await fetch('/tokenize', {
       method: 'POST',
       headers: { 'Content-Type': 'application/json' },
-      body: JSON.stringify({ text }),
+      body: JSON.stringify({ text, system }),
     });
     const data = await resp.json();
     return data.result;
@@ -778,14 +778,14 @@ class MambaRuntime {
   }
 
   // ── Generate text ──────────────────────────────────────────────────────
-  async generate(prompt, maxTokens = 100, temperature = 0.75, onToken = null) {
+  async generate(prompt, maxTokens = 100, temperature = 0.75, onToken = null, system = '') {
     if (!this.ready) throw new Error('Call loadWeights() first');
     this._allocateScratch();
 
     console.log('[mamba] generate:', prompt, 'max_tokens:', maxTokens);
 
-    // Tokenize the prompt
-    const promptTokens = await this.tokenize(prompt);
+    // Tokenize the prompt (with optional system prompt)
+    const promptTokens = await this.tokenize(prompt, system);
     console.log(`[mamba] prompt tokens (${promptTokens.length}):`, promptTokens);
 
     // Process prompt tokens through forward pass to build SSM state