File size: 9,900 Bytes
ad23950 2626977 ad23950 b76ffc6 ad23950 b76ffc6 ad23950 b76ffc6 ad23950 b76ffc6 ad23950 b76ffc6 ad23950 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 |
import os
import sys
import math
import torch
import argparse
import textwrap
import transformers
from peft import PeftModel
from transformers import GenerationConfig, TextStreamer
#from llama_attn_replace import replace_llama_attn
from transformers import RobertaTokenizerFast, RobertaForSequenceClassification
import re
import gradio as gr
PROMPT_DICT = {
"prompt_no_input": (
"Below is an instruction that describes a task. "
"Write a response that appropriately completes the request.\n\n"
"### Instruction:\n{instruction}\n\n### Response:"
),
"prompt_no_input_llama2": (
"<s>[INST] <<SYS>>\n"
"You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.\n\n"
"If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.\n"
"<</SYS>> \n\n {instruction} [/INST]"
),
"prompt_input_llama2": (
"<s>[INST] <<SYS>>\n"
"You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, #while being safe. Your answers should not include any harmful, unethical, racist, sexist, #toxic, dangerous, or illegal content. Please ensure that your responses are socially #unbiased and positive in nature.\n\n"
"If a question does not make any sense, or is not factually coherent, explain why instead of #answering something not correct. If you don't know the answer to a question, please don't #share false information.\n"
"<</SYS>> \n\n {instruction} [/INST]"
)
}
class Args:
def __init__(self):
self.bert_rounter = "waleyWang/CO2RRChat/bert_router_model"
self.bert_model = "waleyWang/CO2RRChat/product_predict"
self.com_plan_model = "waleyWang/CO2RRChat/computational_plan"
self.code_model = "waleyWang/CO2RRChat/code_generate"
self.context_size = 32768
self.max_gen_len = 30000
self.cache_dir = "./cache"
self.temperature = 0.6
self.top_p = 0.9
def get_label_map():
return {
0: "The main product is CH3CH2OH",
1: "The main product is C2H4",
2: "The main product is HCOOH/HCOO-",
3: "The main product is C2+",
4: "The main product is CH3OH",
5: "The main product is CH4",
6: "The main product is CO"}
def load_BERT_Rounter(model_path, query):
tokenizer = transformers.BertTokenizer.from_pretrained(model_path)
model = transformers.BertForSequenceClassification.from_pretrained(model_path, num_labels=3)
model.eval()
label_list = [f'Label_{i}' for i in range(model.config.num_labels)]
inputs = tokenizer(query, return_tensors="pt",
padding=True,
truncation=True,
max_length=512)
with torch.no_grad():
outputs = model(**inputs)
probabilities = torch.nn.functional.softmax(outputs.logits, dim=-1)
predicted_label_id = torch.argmax(probabilities, dim=-1).item()
predicted_label = label_list[predicted_label_id]
confidence = probabilities[0][predicted_label_id].item()
return predicted_label
def load_bert(model_name):
tokenizer = RobertaTokenizerFast.from_pretrained(model_name)
model = RobertaForSequenceClassification.from_pretrained(model_name)
return model, tokenizer
def run_bert(text, model, tokenizer):
label_map = get_label_map()
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True)
outputs = model(**inputs)
logits = outputs.logits
predictions = torch.argmax(logits, dim=-1)
product_label = label_map[predictions.item()]
return product_label
def build_generator(model, tokenizer, temperature=0.6, top_p=0.9, max_gen_len=4096, use_cache=True):
def response(prompt):
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
output = model.generate(
**inputs,
max_new_tokens=max_gen_len,
temperature=temperature,
top_p=top_p,
use_cache=use_cache
)
out = tokenizer.decode(output[0], skip_special_tokens=True)
if "[/INST]" in out:
out = out.split("[/INST]")[1].strip()
return out
return response
def load_llama_model(model_path, device, context_size, cache_dir):
config = transformers.AutoConfig.from_pretrained(
model_path,
cache_dir=cache_dir,
)
orig_ctx_len = getattr(config, "max_position_embeddings", None)
if orig_ctx_len and context_size > orig_ctx_len:
scaling_factor = float(math.ceil(context_size / orig_ctx_len))
config.rope_scaling = {"type": "linear", "factor": scaling_factor}
model = transformers.AutoModelForCausalLM.from_pretrained(
model_path,
config=config,
cache_dir=cache_dir,
torch_dtype=torch.float16,
device_map="auto",
)
model.to(device)
model.resize_token_embeddings(32001)
tokenizer = transformers.AutoTokenizer.from_pretrained(
model_path,
cache_dir=cache_dir,
model_max_length=context_size if context_size > orig_ctx_len else orig_ctx_len,
padding_side="right",
use_fast=False,
)
model.eval()
if torch.__version__ >= "2" and sys.platform != "win32":
model = torch.compile(model)
model.eval()
return model, tokenizer
def generate_llama_output(model, tokenizer, temperature=0.6, top_p=0.9, max_gen_len=4096, use_cache=True):
def response(prompt):
print("Original prompt:", prompt)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
print("Tokenized prompt:", inputs)
streamer = TextStreamer(tokenizer)
print("Generation parameters:")
print("Max new tokens:", max_gen_len)
print("Temperature:", temperature)
print("Top p:", top_p)
output = model.generate(
**inputs,
max_new_tokens=max_gen_len,
temperature=temperature,
top_p=top_p,
use_cache=use_cache,
streamer=streamer,
)
print("Raw model output:", output)
out = tokenizer.decode(output[0], skip_special_tokens= False)
print("Decoded output:", out)
#out = out.split(prompt.lstrip("<s>"))[1].strip()
return out
return response
def main(args):
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
predicted_label = load_BERT_Rounter(args.bert_rounter, args.question)
try:
if predicted_label == 'Label_0':
bert_model, bert_tokenizer = load_bert(args.bert_model)
product_label = run_bert(args.question, bert_model, bert_tokenizer)
response = f"Product Prediction Result:\n{product_label}"
print("Generating response:", response)
return response
elif predicted_label == 'Label_1':
plan_model, plan_tokenizer = load_llama_model(args.com_plan_model, device, args.context_size, args.cache_dir)
plan_response = build_generator(plan_model, plan_tokenizer, temperature=args.temperature, top_p=args.top_p, max_gen_len=args.max_gen_len, use_cache=True)
prompt_no_input = PROMPT_DICT["prompt_no_input_llama2"]
prompt = prompt_no_input.format_map({"instruction": args.question})
output = plan_response(prompt=prompt)
response = f"Computational Planning Result:\n{output}"
print("Generating response:", response)
return response
else:
code_model, code_tokenizer = load_llama_model(args.code_model, device, args.context_size, args.cache_dir)
code_response = build_generator(code_model, code_tokenizer, temperature=args.temperature, top_p=args.top_p, max_gen_len=args.max_gen_len, use_cache=True)
prompt_no_input = PROMPT_DICT["prompt_no_input_llama2"]
prompt = prompt_no_input.format_map({"instruction": args.question})
output = code_response(prompt=prompt)
response = f"Code Generation Result:\n{output}"
print("Generating response:", response)
return response
except Exception as e:
error_msg = f"Error occurred: {str(e)}"
print("Error:", error_msg)
return error_msg
def process_question(question):
args = Args()
args.question = question
try:
result = main(args)
return result
except Exception as e:
return f"Error occurred: {str(e)}"
def create_demo():
iface = gr.Interface(
fn=process_question,
inputs=gr.Textbox(
lines=3,
placeholder="Enter your CO2RR related question here...",
label="Question"
),
outputs=gr.Textbox(
lines=10,
label="Response"
),
title="CO2RR Assistant",
description="Ask questions about CO2 reduction reaction (CO2RR), including product prediction, computational planning, and code generation.",
examples=[
["What is the main product of CO2RR on Cu(100) surface at -0.9V vs. RHE?"],
["Simulate the CO2 reduction reaction (CO2RR) to produce CH4 (methane) on a Cs-Lu alloy (111) surface."],
["Generate VASP input files for CO2RR simulation on Cu(111) surface"]
]
)
return iface
if __name__ == "__main__":
demo = create_demo()
demo.launch(
share=True,
server_name="0.0.0.0"
) |