File size: 13,975 Bytes
766ea9e |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 |
import argparse
import jsonlines
import json
# from deepeval.scorer import Scorer
from deepeval.models import OllamaModel
from deepeval.metrics import (
ContextualRelevancyMetric,
ContextualRecallMetric,
ContextualPrecisionMetric,
AnswerRelevancyMetric,
FaithfulnessMetric
)
# import docx
from deepeval.test_case import LLMTestCase
from deepeval.dataset import EvaluationDataset, Golden
from deepeval import evaluate
from deepeval.models import OllamaModel
import transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
from Llemma_Finetuned import Llemma_Finetuned
import ollama
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from deepeval.models import DeepEvalBaseLLM
class CustomLlemma(DeepEvalBaseLLM):
def __init__(self):
self.torch_device = "cuda" if torch.cuda.is_available() else "cpu"
# finetuned
model = AutoModelForCausalLM.from_pretrained("./merged_models/llemma_lora_merged").to(self.torch_device)
tokenizer = AutoTokenizer.from_pretrained("./merged_models/llemma_lora_merged")
self.model = model
self.tokenizer = tokenizer
def load_model(self):
return self.model
def generate(self, prompt: str) -> str:
model = self.load_model()
pipeline = transformers.pipeline(
"text-generation",
model=model,
tokenizer=self.tokenizer,
framework="pt",
device=0,
max_length=4096,
eos_token_id=self.tokenizer.eos_token_id,
pad_token_id=self.tokenizer.eos_token_id,
)
return pipeline(prompt)
# inputs = self.tokenizer(prompt, return_tensors='pt').to(self.torch_device)
# output = self.model.generate(**inputs)
# a_output = self.tokenizer.decode(output[0])
# return json.dumps(a_output)
async def a_generate(self, prompt: str) -> str:
return self.generate(prompt)
def get_model_name(self):
return "Llemma Fine-tuned"
#ollama run Hudson/llemma:7b
#deepeval set-ollama Hudson/llemma:7b
def is_invalid_length(text, length=4096):
if len(text) <= length:
return False
else:
return True
if __name__=="__main__":
# Initialize parser
parser = argparse.ArgumentParser()
# Adding optional argument
parser.add_argument("-t", "--test", help = "Test to run (ar, cp, crec, crel, f)")
parser.add_argument("-d", "--dataset", help = "Path to test case dataset")
# Read arguments from command line
args = parser.parse_args()
test_type = str(args.test)
test_data = str(args.dataset)
dataset = EvaluationDataset()
# Add as test cases
dataset.add_test_cases_from_json_file(
# file_path is the absolute path to you .json file
file_path=test_data,
input_key_name="input",
actual_output_key_name="actual_output",
expected_output_key_name="expected_output",
context_key_name="context",
retrieval_context_key_name="retrieval_context",
)
# orig
# model = ollama.pull(model="Hudson/llemma:7b")
#OllamaModel(model="Hudson/llemma:7b")
custom_llm = CustomLlemma()
# finetuned
# llemma_model = AutoModelForCausalLM.from_pretrained("./train_llemma/merged_models/llemma_lora_merged")
# tokenizer = AutoTokenizer.from_pretrained("./train_llemma/merged_models/llemma_lora_merged")
# model = Llemma_Finetuned(model=llemma_model, tokenizer=tokenizer)
# sorted_rows = []
# with open('dataset_row_stl.txt', 'r') as file:
# sorted_rows = file.readlines()
# # print(sorted_rows)
# sorted_rows = sorted_rows[0:num_shot]
# sorted_rows = [int(x) for x in sorted_rows]
# print("Read in sorted rows.")
# examples = "Here are " + str(num_shot) + " examples of math questions (Q) with given answers (A).\n"
# with jsonlines.open("mse_text_img_QA_ds_test.jsonl", mode='r') as fp:
# #with open("mse_text_img_QA_ds_test.jsonl", mode='r') as fp:
# n = 0
# for j, data in enumerate(fp):
# if j + 1 in sorted_rows:
# print("Num shot row " + str(j + 1))
# # data = json.loads(line)
# examples += "Q: " + data["body"] + "\n\n"
# is_accepted = False
# best_score = float('-inf')
# output_text = ""
# for i in range(len(data["answers"])):
# if bool(data["answers"][i]["accepted"]) == True:
# if is_accepted == False:
# is_accepted = True
# best_score = int(data["answers"][i]["score"])
# output_text = data["answers"][i]["body"]
# elif int(data["answers"][i]["score"]) > best_score:
# best_score = int(data["answers"][i]["score"])
# output_text = data["answers"][i]["body"]
# elif int(data["answers"][i]["score"]) > best_score:
# best_score = int(data["answers"][i]["score"])
# output_text = data["answers"][i]["body"]
# examples += "A: " + output_text + "\n\n"
# if n == (num_shot - 1):
# examples += "Provide an answer (A) to the following math question (Q) in a similar manner to the previous example(s) given.\n\nQ: "
# # 26th line
# n += 1
# elif n >= num_shot:
# break
# else:
# continue
# print("Generated examples for", str(num_shot), "shot.")
# mse_dataset = []
# with jsonlines.open("mse_text_img_QA_ds_test.jsonl", mode='r') as reader:
# count = 0
# curr_row = 0
# for row in reader.iter(type=dict, skip_invalid=True):
# curr_row += 1
# if curr_row <= skip_to:
# continue
# elif curr_row == 33 or curr_row == 36 or curr_row == 69 \
# or curr_row == 24 or curr_row == 76 \
# or curr_row == 66 or curr_row == 9 \
# or curr_row == 26 or curr_row == 27 \
# or curr_row == 37 or curr_row == 55 \
# or curr_row == 54 or curr_row == 138 \
# or curr_row == 77 or curr_row == 84 or curr_row == 87 \
# or curr_row == 80 or curr_row == 81 or curr_row == 97 \
# or curr_row == 115 or curr_row == 106:
# print("Skipped row " + str(curr_row))
# continue
# elif curr_row in sorted_rows:
# print("Skipped row " + str(curr_row) + " because it is a shorter example")
# continue
# # question_path = "output/" + row["id"]
# # if count ual<= 0:
# # print(obj)
# if count >= test_case_num:
# break
# else:
# input_text = row["body"]
# # response = ollama.generate(model='Hudson/llemma:7b', prompt=input_text)
# # actual_response = response['response']
# is_accepted = False
# best_score = float('-inf')
# output_text = ""
# # context = []
# next_best_answer = ""
# for i in range(len(row["answers"])):
# if bool(row["answers"][i]["accepted"]) == True:
# if is_accepted == False:
# is_accepted = True
# next_best_answer = output_text
# best_score = int(row["answers"][i]["score"])
# output_text = row["answers"][i]["body"]
# elif int(row["answers"][i]["score"]) > best_score:
# next_best_answer = output_text
# best_score = int(row["answers"][i]["score"])
# output_text = row["answers"][i]["body"]
# # else:
# # context.append(row["answers"][i]["body"])
# elif int(row["answers"][i]["score"]) > best_score:
# next_best_answer = output_text
# best_score = int(row["answers"][i]["score"])
# output_text = row["answers"][i]["body"]
# # else:
# # context.append(row["answers"][i]["body"])
# if next_best_answer == "" or next_best_answer is None:
# next_best_answer = row["title"]
# # test_case_dataset.append(LLMTestCase(input=input_text, actual_output=actual_response, expected_output=output_text, retrieval_context=None))
# # test_case_dataset.append(LLMTestCase(input=input_text, actual_output=model.generate(input_text), expected_output=output_text, retrieval_context=context))
# if num_shot == 0:
# i_text = json.dumps(input_text)
# e_output = json.dumps(output_text)
# r_context = json.dumps(next_best_answer)
# gen_answer = ollama.generate(model="Hudson/llemma:7b", prompt=i_text)
# a_output = json.dumps(gen_answer.response)
# # print("i_text = ", i_text)
# # print("a_output = ", a_output)
# # print("e_output = ", e_output)
# # print("r_context = ", r_context)
# # r_context = gen_answer.context
# # if is_invalid_length(i_text) or is_invalid_length(e_output) or is_invalid_length(r_context):
# # continue
# mse_dataset.append(LLMTestCase(input=i_text, actual_output=a_output, expected_output=e_output, retrieval_context=[r_context]))
# else:
# i_text = json.dumps(examples + input_text)
# e_output = json.dumps(output_text)
# r_context = json.dumps(next_best_answer)
# gen_answer = ollama.generate(model="Hudson/llemma:7b", prompt=i_text)
# a_output = json.dumps(gen_answer.response)
# # r_context = gen_answer.context
# # print("i_text = ", i_text)
# # print("a_output = ", a_output)
# # print("e_output = ", e_output)
# # print("r_context = ", r_context)
# # if is_invalid_length(i_text) or is_invalid_length(e_output) or is_invalid_length(r_context):
# # continue
# mse_dataset.append(LLMTestCase(input=i_text, actual_output=a_output, expected_output=e_output, retrieval_context=[r_context]))
# count = count + 1
# # if curr_row % 1 == 0:
# print("At", str(count), "out of", str(test_case_num), " current row =", str(curr_row))
# first_test_case = LLMTestCase(input="...", actual_output="...", context=["..."])
# second_test_case = LLMTestCase(input="...", actual_output="...", context=["..."])
# dataset = EvaluationDataset(test_cases=mse_dataset)
pass_threshold = 0.7
# eval_output = ""
if test_type == "ar":
# answer_relevancy = AnswerRelevancyMetric(model=model, threshold=pass_threshold, async_mode=False)
answer_relevancy = AnswerRelevancyMetric(model=custom_llm, threshold=pass_threshold)
# evaluate(dataset, metrics=[answer_relevancy], out_file=out_path, run_async=True)
# with open(out_path, "a") as f:
# # f.write(dataset.evaluate([answer_relevancy]))
# eval_output = dataset.evaluate([answer_relevancy])
# evaluate(goldens=dataset.goldens, metrics=[answer_relevancy])
evaluate(dataset, metrics=[answer_relevancy])
elif test_type == "cp":
contextual_precision = ContextualPrecisionMetric(model=custom_llm, threshold=pass_threshold)
# evaluate(dataset, metrics=[contextual_precision], out_file=out_path, run_async=True)
# evaluate(dataset, metrics=[contextual_precision])
# eval_output = dataset.evaluate([contextual_precision])
# evaluate(goldens=dataset.goldens, metrics=[contextual_precision])
evaluate(dataset, metrics=[contextual_precision])
elif test_type == "crec":
contextual_recall = ContextualRecallMetric(model=custom_llm, threshold=pass_threshold)
# evaluate(dataset, metrics=[contextual_recall], out_file=out_path, run_async=True)
# evaluate(dataset, metrics=[contextual_recall])
# eval_output = dataset.evaluate([contextual_recall])
# evaluate(goldens=dataset.goldens, metrics=[contextual_recall])
evaluate(dataset, metrics=[contextual_recall])
else:
print("Test case (" + test_type + ") not covered")
# with open(out_path, "a") as f:
# f.write(str(eval_output))
# Create a document
# doc = docx.Document()
# # Add a paragraph to the document
# p = doc.add_paragraph()
# # Add some formatting to the paragraph
# p.paragraph_format.line_spacing = 1
# p.paragraph_format.space_after = 0
# # Add a run to the paragraph
# run = p.add_run(eval_output)
# # Add some formatting to the run
# run.bold = False
# run.italic = False
# run.font.name = 'Arial'
# run.font.size = docx.shared.Pt(12)
# # Save the document
# doc.save(out_path) |