Spaces:
Running
Running
File size: 8,737 Bytes
ce78ec6 b3feaa3 51294aa b3feaa3 3287a07 b3feaa3 3287a07 b3feaa3 f0edf49 b3feaa3 3287a07 ce78ec6 bdaa284 ce78ec6 b3feaa3 ce78ec6 278e21a b3feaa3 3287a07 a072676 b3feaa3 3287a07 b3feaa3 3287a07 b3feaa3 3287a07 ce78ec6 b3feaa3 90c9489 ff079ce b3feaa3 ff079ce b3feaa3 ff079ce 90c9489 b3feaa3 90c9489 b3feaa3 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 |
import torch
import sys
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
import json
tokenizer = AutoTokenizer.from_pretrained('google/gemma-2-2b-it')
# Configure 4-bit quantization using BitsAndBytesConfig
quantization_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_compute_dtype=torch.bfloat16,
bnb_4bit_quant_type="nf4",
)
# Load the model with the quantization configuration
model = AutoModelForCausalLM.from_pretrained(
'google/gemma-2-2b-it',
device_map="auto",
quantization_config=quantization_config,
)
# Definir el prompt para generar un JSON con eventos anidados
prompt = (
"Genera un JSON que describa una serie de eventos consecutivos en un formato similar al siguiente:\n\n"
"{\n"
" \"events\": {\n"
" \"event\": {\n"
" \"event_number\": 1,\n"
" \"name\": \"conflict_start\",\n"
" \"description\": \"Tensions escalate between Iran and Israel\",\n"
" \"probability\": 70,\n"
" \"duration_days\": 30,\n"
" \"subevents\": {\n"
" \"event\": {\n"
" \"event_number\": 2,\n"
" \"name\": \"diplomatic_failure\",\n"
" \"description\": \"Diplomatic negotiations fail\",\n"
" \"probability\": 60,\n"
" \"duration_days\": 15,\n"
" \"subevents\": {\n"
" \"event\": {\n"
" \"event_number\": 3,\n"
" \"name\": \"military_clash\",\n"
" \"description\": \"Initial military clash at the border\",\n"
" \"probability\": 50,\n"
" \"duration_days\": 10,\n"
" \"subevents\": {\n"
" \"event\": [\n"
" {\n"
" \"event_number\": 4,\n"
" \"name\": \"escalation\",\n"
" \"description\": \"Conflict escalates into full-scale war\",\n"
" \"probability\": 40,\n"
" \"duration_days\": 180,\n"
" \"subevents\": {\n"
" \"event\": [\n"
" {\n"
" \"event_number\": 5,\n"
" \"name\": \"regional_involvement\",\n"
" \"description\": \"Other Middle Eastern countries get involved\",\n"
" \"probability\": 30,\n"
" \"duration_days\": 365,\n"
" \"subevents\": {\n"
" \"event\": [\n"
" {\n"
" \"event_number\": 6,\n"
" \"name\": \"ceasefire\",\n"
" \"description\": \"International powers broker a ceasefire\",\n"
" \"probability\": 20,\n"
" \"duration_days\": 30\n"
" },\n"
" {\n"
" \"event_number\": 7,\n"
" \"name\": \"prolonged_conflict\",\n"
" \"description\": \"Conflict continues for over a year\",\n"
" \"probability\": 50,\n"
" \"duration_days\": 365\n"
" }\n"
" ]\n"
" }\n"
" },\n"
" {\n"
" \"event_number\": 8,\n"
" \"name\": \"international_intervention\",\n"
" \"description\": \"UN or other international organizations intervene\",\n"
" \"probability\": 25,\n"
" \"duration_days\": 60\n"
" }\n"
" ]\n"
" }\n"
" },\n"
" {\n"
" \"event_number\": 9,\n"
" \"name\": \"containment\",\n"
" \"description\": \"Conflict is contained and doesn't escalate\",\n"
" \"probability\": 30,\n"
" \"duration_days\": 90\n"
" }\n"
" ]\n"
" }\n"
" },\n"
" \"event\": {\n"
" \"event_number\": 10,\n"
" \"name\": \"sanctions\",\n"
" \"description\": \"Increased sanctions on Iran\",\n"
" \"probability\": 70,\n"
" \"duration_days\": 180,\n"
" \"subevents\": {\n"
" \"event\": [\n"
" {\n"
" \"event_number\": 11,\n"
" \"name\": \"iran_retaliates\",\n"
" \"description\": \"Iran retaliates with cyberattacks\",\n"
" \"probability\": 40,\n"
" \"duration_days\": 60\n"
" },\n"
" {\n"
" \"event_number\": 12,\n"
" \"name\": \"israel_response\",\n"
" \"description\": \"Israel responds with targeted airstrikes\",\n"
" \"probability\": 50,\n"
" \"duration_days\": 60\n"
" }\n"
" ]\n"
" }\n"
" }\n"
" }\n"
" },\n"
" \"event\": {\n"
" \"event_number\": 13,\n"
" \"name\": \"diplomatic_success\",\n"
" \"description\": \"Successful diplomatic negotiations\",\n"
" \"probability\": 40,\n"
" \"duration_days\": 30,\n"
" \"subevents\": {\n"
" \"event\": [\n"
" {\n"
" \"event_number\": 14,\n"
" \"name\": \"peace_agreement\",\n"
" \"description\": \"Iran and Israel sign a peace agreement\",\n"
" \"probability\": 20,\n"
" \"duration_days\": 60\n"
" },\n"
" {\n"
" \"event_number\": 15,\n"
" \"name\": \"temporary_truce\",\n"
" \"description\": \"A temporary truce is established\",\n"
" \"probability\": 30,\n"
" \"duration_days\": 30\n"
" }\n"
" ]\n"
" }\n"
" }\n"
" }\n"
" }\n"
" }\n"
"}\n\n"
"Ahora, genera un JSON similar con eventos anidados, pero cambia los detalles y números para hacer que sea con el input que viene a continuacion, respondiendo solo el JSON empezando con <json>:"
)
def generate(event):
combined_input = f"{prompt} {event}" # Combine prompt and event
prompt_msg = [{'role': 'user', 'content': combined_input}]
inputs = tokenizer.apply_chat_template(
prompt_msg,
add_generation_prompt=True,
return_tensors='pt'
)
tokens = model.generate(
inputs.to(model.device),
max_new_tokens=1024,
temperature=0.5,
do_sample=True
)
output_text = tokenizer.decode(tokens[0], skip_special_tokens=False)
print(output_text)
user_prompt_length = len(f"<bos><start_of_turn>user\n{prompt}\n{event}<end_of_turn>\n<start_of_turn>model\n")
json_start_index = output_text.find("<json>")
json_end_index = output_text.find("</json>")
if json_start_index != -1 and json_end_index != -1:
json_string = output_text[max(json_start_index + 6, user_prompt_length):json_end_index].strip()
# Remove any leading/trailing non-JSON characters (if present)
if not json_string.startswith("{"):
first_brace_index = json_string.find("{")
if first_brace_index != -1:
json_string = json_string[first_brace_index:]
if not json_string.endswith("}"):
last_brace_index = json_string.rfind("}")
if last_brace_index != -1:
json_string = json_string[:last_brace_index + 1]
# Load JSON without validation
try:
data = json.loads(json_string)
return data # Return the parsed JSON data
except json.JSONDecodeError as e:
return f"Error: Invalid JSON - {e}"
else:
return "Error: <json> or </json> not found in generated output"
|