fix typos and add 8B scripts

app.py

@@ -79,7 +79,7 @@ def respond(
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
-        gr.Textbox(value="You are Magpie, a helpful AI assistant. For simple qeuries, try to answer them directly; for complex questions, try to think step-by-step before providing an answer.", label="System message"),
+        gr.Textbox(value="You are Magpie, a helpful AI assistant. For simple queries, try to answer them directly; for complex questions, try to think step-by-step before providing an answer.", label="System message"),
         gr.Slider(minimum=128, maximum=2048, value=512, step=64, label="Max new tokens"),
         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
         gr.Slider(
@@ -89,7 +89,7 @@ demo = gr.ChatInterface(
             step=0.1,
             label="Top-p (nucleus sampling)",
         ),
-        gr.Slider(minimum=0.5, maximum=1.5, value=1.0, step=0.1, label="Repetation Penalty"),
+        gr.Slider(minimum=0.5, maximum=1.5, value=1.0, step=0.1, label="Repetition Penalty"),
     ],
     description=header,  # Add the header as the description
     title="MagpieLM-4B Chat (v0.1)",

app_8B.py

@@ -0,0 +1,101 @@
+import gradio as gr
+from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
+import spaces
+from threading import Thread
+from typing import Iterator
+
+# Add markdown header
+header = """
+# 🐦‍⬛ MagpieLMs: Open LLMs with Fully Transparent Alignment Recipes
+
+💬 We've aligned Llama-3.1-8B and a 4B version (distilled by NVIDIA) using purely synthetic data generated by our [Magpie](https://arxiv.org/abs/2406.08464) method. Our open-source post-training recipe includes: SFT and DPO data, all training configs + logs. This allows everyone to reproduce the alignment process for their own research. Note that our data does not contain any GPT-generated data, and has a much friendly license for both commercial and academic use.
+
+- **Magpie Collection**: [Magpie on Hugging Face](https://lnkd.in/g_pgX5Y2)
+- **Magpie Paper**: [Read the research paper](https://arxiv.org/abs/2406.08464) 
+
+Contact: [Zhangchen Xu](https://zhangchenxu.com) and [Bill Yuchen Lin](https://yuchenlin.xyz).
+
+---
+"""
+
+# Load model and tokenizer
+model_name = "Magpie-Align/MagpieLM-8B-Chat-v0.1"
+
+device = "cuda" # the device to load the model onto
+tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
+model = AutoModelForCausalLM.from_pretrained(
+    model_name,
+    torch_dtype="auto",
+    ignore_mismatched_sizes=True
+)
+model.to(device)
+
+MAX_INPUT_TOKEN_LENGTH = 4096  # You may need to adjust this value
+
+@spaces.GPU
+def respond(
+    message: str,
+    chat_history: list[tuple[str, str]],
+    system_prompt: str,
+    max_new_tokens: int = 1024,
+    temperature: float = 0.6,
+    top_p: float = 0.9,
+    top_k: int = 50,
+    repetition_penalty: float = 1.2,
+) -> Iterator[str]:
+    conversation = []
+    if system_prompt:
+        conversation.append({"role": "system", "content": system_prompt})
+    for user, assistant in chat_history:
+        conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
+    conversation.append({"role": "user", "content": message})
+
+    input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt")
+    if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
+        input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
+        gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
+    input_ids = input_ids.to(model.device)
+
+    streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
+    generate_kwargs = dict(
+        input_ids=input_ids,
+        streamer=streamer,
+        max_new_tokens=max_new_tokens,
+        do_sample=True,
+        top_p=top_p,
+        top_k=top_k,
+        temperature=temperature,
+        num_beams=1,
+        repetition_penalty=repetition_penalty,
+    )
+    t = Thread(target=model.generate, kwargs=generate_kwargs)
+    t.start()
+
+    outputs = []
+    for text in streamer:
+        outputs.append(text)
+        yield "".join(outputs)
+
+demo = gr.ChatInterface(
+    respond,
+    additional_inputs=[
+        gr.Textbox(value="You are Magpie, a helpful AI assistant. For simple queries, try to answer them directly; for complex questions, try to think step-by-step before providing an answer.", label="System message"),
+        gr.Slider(minimum=128, maximum=2048, value=512, step=64, label="Max new tokens"),
+        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
+        gr.Slider(
+            minimum=0.1,
+            maximum=1.0,
+            value=0.9,
+            step=0.1,
+            label="Top-p (nucleus sampling)",
+        ),
+        gr.Slider(minimum=0.5, maximum=1.5, value=1.0, step=0.1, label="Repetition Penalty"),
+    ],
+    description=header,  # Add the header as the description
+    title="MagpieLM-8B Chat (v0.1)",
+    theme=gr.themes.Soft()
+)
+
+if __name__ == "__main__":
+    demo.queue()
+    demo.launch(share=True)