try HG model w sample params

README.md

@@ -6,12 +6,13 @@ colorTo: blue
 sdk: static
 pinned: false
 models:
+- HuggingFaceH4/zephyr-7b-beta
 # - openai-community/gpt2
 # - meta-llama/Meta-Llama-3-70B-Instruct
 # - Xenova/detr-resnet-50
 # - Xenova/gpt2
 # - Xenova/bloom-560m
-- Xenova/distilgpt2
+# - Xenova/distilgpt2
 # - Xenova/gpt-3.5-turbo
 # - Xenova/llama-68m
 # - Xenova/LaMini-Flan-T5-783M

sketch.js

@@ -161,26 +161,34 @@ async function runModel(PREPROMPT, PROMPT){
     let inputText = PREPROMPT + PROMPT
 
     // pipeline/transformers version
-    let pipe = await pipeline('text-generation');
+    // let generator = await pipeline('text-generation', 'Xenova/distilgpt2');
     // seems to work with default model distilgpt2 ugh
+    let generator = pipeline(model="HuggingFaceH4/zephyr-7b-beta")
+
+    // IMPORTANT: different models have different input/output structures for their API so look to the samples and references
 
     // 'meta-llama/Meta-Llama-3-70B-Instruct'
     // 'openai-community/gpt2'
     // 'Xenova/gpt-3.5-turbo'
     // , 'Xenova/distilgpt2'
 
-    let out = await pipe(inputText, {
-        max_tokens: 250
-        // return_full_text: false
-        // repetition_penalty: 1.5,
-        // num_return_sequences: 1 //must be 1 for greedy search
-    })
+    // let out = await generator(inputText, {
+    //     max_tokens: 250,
+    //     return_full_text: false
+    //     repetition_penalty: 1.5,
+    //     num_return_sequences: 1 //must be 1 for greedy search
+    // })
 
-    // let out = await pipe(inputText)
+    let out = await generator({
+        "role": "user",
+        "content": inputText,
+        "max_tokens": 250
+    })
     
     console.log(out)
 
-    var modelResult = await out[0].generated_text
+    // var modelResult = await out[0].generated_text
+    var modelResult = await out[0].generated_text[0].content
     console.log(modelResult)
 
     return modelResult