Spaces:
Build error
Build error
File size: 5,732 Bytes
0f06270 7b71f79 83aa28d 0f06270 83aa28d 0f06270 83aa28d 0f06270 83aa28d 0f06270 7b71f79 0f06270 83aa28d 0f06270 7b71f79 0f06270 7b71f79 0f06270 7b71f79 0f06270 7b71f79 0f06270 7b71f79 0f06270 7b71f79 0f06270 7b71f79 0f06270 7b71f79 0f06270 7b71f79 0f06270 7b71f79 0f06270 640a737 7b71f79 0f06270 7b71f79 0f06270 7b71f79 640a737 |
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 |
import os
import sys
import evaluate
import gradio as gr
from huggingface_hub import InferenceClient, login
from dotenv import find_dotenv, load_dotenv
from huggingface_hub import login
found_dotenv = find_dotenv(".env")
if len(found_dotenv) == 0:
found_dotenv = find_dotenv(".env.example")
print(f"loading env vars from: {found_dotenv}")
load_dotenv(found_dotenv, override=False)
path = os.path.dirname(found_dotenv)
print(f"Adding {path} to sys.path")
sys.path.append(path)
from llm_toolkit.llm_utils import *
from llm_toolkit.translation_utils import *
from eval_modules.calc_repetitions_v2e import detect_repetitions
model_name = os.getenv("MODEL_NAME") or "microsoft/Phi-3.5-mini-instruct"
num_shots = int(os.getenv("NUM_SHOTS", 10))
data_path = os.getenv("DATA_PATH")
hf_token = os.getenv("HF_TOKEN")
login(token=hf_token, add_to_git_credential=True)
comet = evaluate.load("comet", config_name="Unbabel/wmt22-cometkiwi-da", gpus=1)
meteor = evaluate.load("meteor")
bleu = evaluate.load("bleu")
rouge = evaluate.load("rouge")
def calc_perf_scores(prediction, source, reference, debug=False):
if debug:
print("prediction:", prediction)
print("source:", source)
print("reference:", reference)
if reference:
bleu_scores = bleu.compute(
predictions=[prediction], references=[reference], max_order=1
)
rouge_scores = rouge.compute(predictions=[prediction], references=[reference])
rouge_scores = rouge.compute(predictions=[prediction], references=[reference])
meteor_scores = meteor.compute(predictions=[prediction], references=[reference])
comet_metric = comet.compute(
predictions=[prediction], sources=[source], references=[reference]
)
result = {"bleu_scores": bleu_scores, "rouge_scores": rouge_scores, "meteor_scores":meteor_scores, "comet_scores": comet_metric}
if debug:
print("result:", result)
return result
"""
For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
"""
# client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
client = InferenceClient(model_name, token=hf_token)
datasets = load_translation_dataset(data_path)
print_row_details(datasets["test"].to_pandas())
translation_prompt = get_few_shot_prompt(datasets["train"], num_shots)
examples = [[row["chinese"]] for row in datasets["test"]][:5]
def respond(
message,
history: list[tuple[str, str]],
system_message,
max_tokens,
temperature,
top_p,
):
messages = [{"role": "system", "content": system_message}]
source = message
for val in history:
if val[0]:
messages.append({"role": "user", "content": val[0]})
if val[1]:
messages.append({"role": "assistant", "content": val[1]})
messages.append({"role": "user", "content": translation_prompt.format(input=message)})
partial_text = ""
finish_reason = None
for message in client.chat_completion(
messages,
max_tokens=max_tokens,
stream=True,
temperature=temperature,
frequency_penalty=None, # frequency_penalty,
presence_penalty=None, # presence_penalty,
top_p=top_p,
seed=42,
):
finish_reason = message.choices[0].finish_reason
# print("finish_reason:", finish_reason)
if finish_reason is None:
new_text = message.choices[0].delta.content
partial_text += new_text
yield partial_text
else:
break
answer = partial_text
(whitespace_score, repetition_score, total_repetitions) = detect_repetitions(answer, debug=True)
partial_text += "\n\nRepetition Metrics:\n"
partial_text += f"1. Whitespace Score: {whitespace_score:.3f}\n"
partial_text += f"1. Repetition Score: {repetition_score:.3f}\n"
partial_text += f"1. Total Repetitions: {total_repetitions:.3f}\n"
partial_text += (
f"1. Non-Repetitive Ratio: {1 - total_repetitions / len(answer):.3f}\n"
)
partial_text += "\n\n Performance Metrics:\n"
if [source] in examples:
idx = examples.index([source])
reference = datasets["test"]["english"][idx]
else:
reference = ""
scores = calc_perf_scores(answer, source, reference, debug=True)
partial_text += f'1. COMET: {scores["comet_scores"]["mean_score"]:.3f}\n'
if reference:
partial_text += f'1. METEOR: {scores["meteor_scores"]["meteor"]:.3f}\n'
partial_text += f'1. BLEU-1: {scores["bleu_scores"]["bleu"]:.3f}\n'
partial_text += f'1. RougeL: {scores["rouge_scores"]["rougeL"]:.3f}\n'
partial_text += f"\n\nGround truth: {reference}\n"
partial_text += f"\n\nThe text generation has ended because: {finish_reason}\n"
yield partial_text
"""
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
"""
demo = gr.ChatInterface(
respond,
examples=examples,
cache_examples=False,
textbox=gr.Textbox(placeholder="Enter your Chinese sentence for translation"),
additional_inputs=[
gr.Textbox(value="You are a helpful assistant that translates Chinese to English.", label="System message"),
gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
gr.Slider(
minimum=0.1,
maximum=1.0,
value=0.95,
step=0.05,
label="Top-p (nucleus sampling)",
),
],
)
if __name__ == "__main__":
demo.launch()
|