Upload serve_mamba.js with huggingface_hub

serve_mamba.js

@@ -19,7 +19,9 @@ function handleAPI(req, res) {
     const data = JSON.parse(body);
     const p = req.url;
     const script = p === '/tokenize'
-      ? `import sys,json;from transformers import AutoTokenizer;t=AutoTokenizer.from_pretrained('tiiuae/falcon-mamba-7b-instruct');d=json.loads(sys.stdin.read());text=d['text'];msgs=[{'role':'user','content':text}];templated=t.apply_chat_template(msgs,tokenize=False,add_generation_prompt=True);print(json.dumps(t.encode(templated)))`
+      ? `import sys,json;from transformers import AutoTokenizer;t=AutoTokenizer.from_pretrained('tiiuae/falcon-mamba-7b-instruct');d=json.loads(sys.stdin.read());text=d['text'];system=d.get('system','');msgs=[];
+if system:msgs.append({'role':'system','content':system})
+msgs.append({'role':'user','content':text});templated=t.apply_chat_template(msgs,tokenize=False,add_generation_prompt=True);print(json.dumps(t.encode(templated)))`
       : `import sys,json;from transformers import AutoTokenizer;t=AutoTokenizer.from_pretrained('tiiuae/falcon-mamba-7b-instruct');print(json.dumps(t.decode(json.loads(sys.stdin.read())['tokens'])))`;
     const child = execFile('python', ['-c', script], { maxBuffer: 10*1024*1024 }, (err, stdout) => {
       res.writeHead(200, { 'Content-Type': 'application/json', 'Access-Control-Allow-Origin': '*' });