try new zephyr model with lots of params

sketch.js

@@ -158,14 +158,27 @@ new p5(function (p5) {
 async function runModel(PREPROMPT, PROMPT){
     // // Chat completion API
 
-    let inputText = PREPROMPT + PROMPT
+    // let inputText = PREPROMPT + PROMPT
+
+    // for zephyr customizing
+    let inputs = [
+        {
+            "role": "system",
+            "content": PREPROMPT
+        },{
+            "role": "user",
+            "content": PROMPT,
+            "max_tokens": 250
+        }
+    ]
+
 
     // pipeline/transformers version
     // let generator = await pipeline('text-generation', 'Xenova/distilgpt2');
     // seems to work with default model distilgpt2 ugh
-    let generator = pipeline(model="HuggingFaceH4/zephyr-7b-beta")
+    let generator = pipeline('text-generation', "HuggingFaceH4/zephyr-7b-beta")
 
-    // IMPORTANT: different models have different input/output structures for their API so look to the samples and references
+    // IMPORTANT: different models have different input/output structures for their API so look to the samples and references on the specific model page for help :)
 
     // 'meta-llama/Meta-Llama-3-70B-Instruct'
     // 'openai-community/gpt2'
@@ -179,11 +192,20 @@ async function runModel(PREPROMPT, PROMPT){
     //     num_return_sequences: 1 //must be 1 for greedy search
     // })
 
-    let out = await generator({
-        "role": "user",
-        "content": inputText,
-        "max_tokens": 250
+    // for zephyr customizing
+    let prompt = pipe.tokenizer.apply_chat_template(inputs, {
+        tokenize: false,
+        add_generation_prompt: true
     })
+
+    let out = await generator(prompt, {
+        max_new_tokens: 256,
+        do_sample: true,
+        temperature: 0.7,
+        top_k: 50,
+        top_p: 0.95
+    })
+        
     
     console.log(out)