Create app.py

app.py

@@ -0,0 +1,77 @@
+import gradio as gr
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+# Hugging Face repo + subfolder där den mergade modellen ligger
+MODEL_ID = "Jeppcode/ScalableLab2"
+SUBFOLDER = "merged-model-fp16"
+
+print(f"Loading model {MODEL_ID}/{SUBFOLDER} ...")
+
+# Ladda tokenizer och modell från subfoldern
+tokenizer = AutoTokenizer.from_pretrained(
+    MODEL_ID,
+    subfolder=SUBFOLDER,
+)
+
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_ID,
+    subfolder=SUBFOLDER,
+    device_map="auto",   # på HF CPU-space hamnar den på cpu
+)
+
+def build_prompt(message, history):
+    """
+    Bygger upp en lista av chat-meddelanden i samma format
+    som användes vid träning, och använder sedan chat_template.
+    """
+    messages = []
+    for user_msg, bot_msg in history:
+        if user_msg is not None:
+            messages.append({"role": "user", "content": user_msg})
+        if bot_msg is not None:
+            messages.append({"role": "assistant", "content": bot_msg})
+
+    messages.append({"role": "user", "content": message})
+
+    # Din tokenizer har en chat_template.jinja, så apply_chat_template ska fungera
+    prompt = tokenizer.apply_chat_template(
+        messages,
+        tokenize=False,
+        add_generation_prompt=True,
+    )
+    return prompt
+
+def chat_fn(message, history):
+    prompt = build_prompt(message, history)
+
+    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=256,
+            do_sample=True,
+            temperature=0.7,
+            top_p=0.9,
+            pad_token_id=tokenizer.eos_token_id,
+        )
+
+    generated = tokenizer.decode(
+        outputs[0][inputs["input_ids"].shape[1]:],
+        skip_special_tokens=True,
+    ).strip()
+
+    return generated
+
+demo = gr.ChatInterface(
+    fn=chat_fn,
+    title="Lab 2 – Fine-tuned merged model (fp16)",
+    description=(
+        "Chat with our fine-tuned Llama-based model, merged to fp16 and "
+        "loaded from Jeppcode/ScalableLab2/merged-model-fp16."
+    ),
+)
+
+if __name__ == "__main__":
+    demo.launch()