switched back to NPC_brain with different prompt insertion

app.py

@@ -2,7 +2,7 @@ import gradio as gr
 import random
 import spaces
 import torch
-from transformers import AutoTokenizer, AutoModelForMultimodalLM
+from transformers import pipeline
 import csv
 import os
 import uuid
@@ -13,27 +13,7 @@ os.makedirs(DATA_DIR, exist_ok=True)
 csv_path = None
 
 
-tokenizer = AutoTokenizer.from_pretrained("openbmb/MiniCPM5-1B")
-model = AutoModelForMultimodalLM.from_pretrained("openbmb/MiniCPM5-1B")
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-model.to(device)
-
-
-def generate_from_messages(messages, max_new_tokens=256, enable_thinking=False):
-    generation_messages = list(messages)
-    if enable_thinking:
-        generation_messages.append(
-            {"role": "user", "content": "Please think through the answer before responding."}
-        )
-    inputs = tokenizer.apply_chat_template(
-        generation_messages,
-        add_generation_prompt=True,
-        tokenize=True,
-        return_dict=True,
-        return_tensors="pt",
-    ).to(device)
-    output_ids = model.generate(**inputs, max_new_tokens=max_new_tokens)
-    return tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0].strip()
+pipe = pipeline("text-generation", model="mecoffey/NPC_brain")
 
 DESC_SYSTEM_PROMPT = (
     "Write a description of a fantasy character based on the user prompt"
@@ -265,37 +245,29 @@ def append_to_csv(csv_path, prompt, description, backstory, strength, dexterity,
         ])
 
 @spaces.GPU
-def char_description(prompt, enable_thinking=False):
+def char_description(prompt):
 
-    prompt1 = [
-        {"role": "system", "content": DESC_SYSTEM_PROMPT},
-        {"role": "user", "content": prompt},
-    ]
-    output1_text = generate_from_messages(prompt1, max_new_tokens=256, enable_thinking=enable_thinking)
+    prompt1 = [{"role": "user", "content": prompt}]
+    output1 = pipe(prompt1, return_full_text=False, max_new_tokens=256)
+    output1_text = output1[0]["generated_text"].strip()
+    combine = str(prompt) + "\n" + str(output1_text) + "\nBackstory:"
     prompt2 = [
-        {"role": "system", "content": BACK_SYSTEM_PROMPT},
-        {"role": "user", "content": prompt},
-        {"role": "user", "content": output1_text},
-        {"role": "user", "content": "Backstory:"},
+        {"role": "user", "content": combine},
     ]
-    output2_text = generate_from_messages(prompt2, max_new_tokens=256, enable_thinking=enable_thinking)
-    return output1_text, output2_text
+    output2 = pipe(prompt2, return_full_text=False, max_new_tokens=256)
+    return output1_text, output2[0]["generated_text"].strip()
 
 @spaces.GPU
-def rand_char_description(prompt, enable_thinking=False):
-    prompt1 = [
-        {"role": "system", "content": DESC_SYSTEM_PROMPT},
-        {"role": "user", "content": prompt},
-    ]
-    output1_text = generate_from_messages(prompt1, max_new_tokens=256, enable_thinking=enable_thinking)
+def rand_char_description(prompt):
+    prompt1 = [{"role": "user", "content": prompt}]
+    output1 = pipe(prompt1, return_full_text=False, max_new_tokens=256)
+    output1_text = output1[0]["generated_text"].strip()
+    combine = str(prompt) + "\n" + str(output1_text) + "\nBackstory:"
     prompt2 = [
-        {"role": "system", "content": BACK_SYSTEM_PROMPT},
-        {"role": "user", "content": prompt},
-        {"role": "user", "content": output1_text},
-        {"role": "user", "content": "Backstory:"},
+        {"role": "user", "content": combine},
     ]
-    output2_text = generate_from_messages(prompt2, max_new_tokens=256, enable_thinking=enable_thinking)
-    return output1_text, output2_text
+    output2 = pipe(prompt2, return_full_text=False, max_new_tokens=256)
+    return output1_text, output2[0]["generated_text"].strip()
 
 
 def char_description_csv(prompt, state):