Spaces:
Runtime error
Runtime error
Commit
·
c12c1d4
1
Parent(s):
465c5b4
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,36 +1,61 @@
|
|
| 1 |
-
|
|
|
|
| 2 |
import gradio as gr
|
| 3 |
import torch
|
|
|
|
| 4 |
|
| 5 |
-
|
|
|
|
|
|
|
| 6 |
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
description = """# <h1 style="text-align: center; color: white;"><span style='color: #F26207;'> Code Completion with falcoder-7b </h1>
|
| 10 |
-
<span style="color: white; text-align: center;"> falcoder-7b You can click the button to generate your code.</span>"""
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
token = os.environ["HUB_TOKEN"]
|
| 14 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 15 |
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
|
|
|
|
| 34 |
custom_css = """
|
| 35 |
.gradio-container {
|
| 36 |
background-color: #0D1525;
|
|
@@ -46,18 +71,52 @@ custom_css = """
|
|
| 46 |
"""
|
| 47 |
|
| 48 |
def post_processing(prompt, completion):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 49 |
return prompt + completion
|
| 50 |
-
# completion = "<span style='color: #499cd5;'>" + completion + "</span>"
|
| 51 |
-
# prompt = "<span style='color: black;'>" + prompt + "</span>"
|
| 52 |
-
# code_html = f"<hr><br><pre style='font-size: 14px'><code>{prompt}{completion}</code></pre><br><hr>"
|
| 53 |
-
# return code_html
|
| 54 |
|
| 55 |
-
|
| 56 |
def code_generation(prompt, max_new_tokens, temperature=0.2, seed=42, top_p=0.9, top_k=None, use_cache=True, repetition_penalty=1.0):
|
| 57 |
-
|
| 58 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 59 |
x = tokenizer.encode(prompt, return_tensors="pt", max_length=MAX_INPUT_TOKENS, truncation=True).to(device)
|
| 60 |
-
|
|
|
|
|
|
|
| 61 |
set_seed(seed)
|
| 62 |
y = model.generate(x,
|
| 63 |
max_new_tokens=max_new_tokens,
|
|
@@ -71,75 +130,33 @@ def code_generation(prompt, max_new_tokens, temperature=0.2, seed=42, top_p=0.9,
|
|
| 71 |
)
|
| 72 |
completion = tokenizer.decode(y[0], skip_special_tokens=True, clean_up_tokenization_spaces=False)
|
| 73 |
completion = completion[len(prompt):]
|
|
|
|
| 74 |
return post_processing(prompt, completion)
|
| 75 |
|
|
|
|
|
|
|
| 76 |
|
| 77 |
-
|
| 78 |
-
css=custom_css
|
| 79 |
-
)
|
| 80 |
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
code = gr.Code(lines=28,label='Input', value="def sieve_eratosthenes(n):")
|
| 87 |
-
with settings_col:
|
| 88 |
-
with gr.Accordion("Generation Settings", open=True):
|
| 89 |
-
max_new_tokens= gr.Slider(
|
| 90 |
-
minimum=8,
|
| 91 |
-
maximum=128,
|
| 92 |
-
step=1,
|
| 93 |
-
value=48,
|
| 94 |
-
label="Max Tokens",
|
| 95 |
-
)
|
| 96 |
-
temperature = gr.Slider(
|
| 97 |
-
minimum=0.1,
|
| 98 |
-
maximum=2.5,
|
| 99 |
-
step=0.1,
|
| 100 |
-
value=0.2,
|
| 101 |
-
label="Temperature",
|
| 102 |
-
)
|
| 103 |
-
repetition_penalty = gr.Slider(
|
| 104 |
-
minimum=1.0,
|
| 105 |
-
maximum=1.9,
|
| 106 |
-
step=0.1,
|
| 107 |
-
value=1.0,
|
| 108 |
-
label="Repetition Penalty. 1.0 means no penalty.",
|
| 109 |
-
)
|
| 110 |
-
seed = gr.Slider(
|
| 111 |
-
minimum=0,
|
| 112 |
-
maximum=1000,
|
| 113 |
-
step=1,
|
| 114 |
-
label="Random Seed"
|
| 115 |
-
)
|
| 116 |
-
top_p = gr.Slider(
|
| 117 |
-
minimum=0.1,
|
| 118 |
-
maximum=1.0,
|
| 119 |
-
step=0.1,
|
| 120 |
-
value=0.9,
|
| 121 |
-
label="Top P",
|
| 122 |
-
)
|
| 123 |
-
top_k = gr.Slider(
|
| 124 |
-
minimum=1,
|
| 125 |
-
maximum=64,
|
| 126 |
-
step=1,
|
| 127 |
-
value=4,
|
| 128 |
-
label="Top K",
|
| 129 |
-
)
|
| 130 |
-
use_cache = gr.Checkbox(
|
| 131 |
-
label="Use Cache",
|
| 132 |
-
value=True
|
| 133 |
-
)
|
| 134 |
-
|
| 135 |
-
with gr.Row():
|
| 136 |
-
run = gr.Button(elem_id="orange-button", value="Generate")
|
| 137 |
|
| 138 |
-
|
| 139 |
-
|
| 140 |
-
# # with middle_col_row_2:
|
| 141 |
-
# output = gr.HTML(label="Generated Code")
|
| 142 |
|
| 143 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 144 |
|
| 145 |
-
|
|
|
|
|
|
| 1 |
+
# Import necessary libraries
|
| 2 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 3 |
import gradio as gr
|
| 4 |
import torch
|
| 5 |
+
import logging
|
| 6 |
|
| 7 |
+
# Set up logging
|
| 8 |
+
logging.basicConfig(level=logging.INFO)
|
| 9 |
+
logger = logging.getLogger(__name__)
|
| 10 |
|
| 11 |
+
# Set device to GPU if available, otherwise CPU
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 13 |
|
| 14 |
+
# Load tokenizer and model
|
| 15 |
+
tokenizer = AutoTokenizer.from_pretrained("mrm8488/falcoder-7b")
|
| 16 |
+
model = AutoModelForCausalLM.from_pretrained("mrm8488/falcoder-7b")
|
| 17 |
+
|
| 18 |
+
def generate_text(prompt, max_length, do_sample, temperature, top_k, top_p):
|
| 19 |
+
"""
|
| 20 |
+
Generates text completion given a prompt and specified parameters.
|
| 21 |
+
|
| 22 |
+
:param prompt: Input prompt for text generation.
|
| 23 |
+
:type prompt: str
|
| 24 |
+
:param max_length: Maximum length of generated text.
|
| 25 |
+
:type max_length: int
|
| 26 |
+
:param do_sample: Whether to use sampling for text generation.
|
| 27 |
+
:type do_sample: bool
|
| 28 |
+
:param temperature: Sampling temperature for text generation.
|
| 29 |
+
:type temperature: float
|
| 30 |
+
:param top_k: Value for top-k sampling.
|
| 31 |
+
:type top_k: int
|
| 32 |
+
:param top_p: Value for top-p sampling.
|
| 33 |
+
:type top_p: float
|
| 34 |
+
:return: Generated text completion.
|
| 35 |
+
:rtype: str
|
| 36 |
+
"""
|
| 37 |
+
|
| 38 |
+
# Format prompt
|
| 39 |
+
formatted_prompt = "\n" + prompt
|
| 40 |
+
if not ',' in prompt:
|
| 41 |
+
formatted_prompt += ','
|
| 42 |
+
|
| 43 |
+
# Tokenize prompt and move to device
|
| 44 |
+
prompt = tokenizer(formatted_prompt, return_tensors='pt')
|
| 45 |
+
prompt = {key: value.to(device) for key, value in prompt.items()}
|
| 46 |
+
|
| 47 |
+
# Generate text completion using model and specified parameters
|
| 48 |
+
out = model.generate(**prompt, max_length=max_length, do_sample=do_sample, temperature=temperature,
|
| 49 |
+
no_repeat_ngram_size=3, top_k=top_k, top_p=top_p)
|
| 50 |
+
output = tokenizer.decode(out[0])
|
| 51 |
+
clean_output = output.replace('\n', '\n')
|
| 52 |
+
|
| 53 |
+
# Log generated text completion
|
| 54 |
+
logger.info("Text generated: %s", clean_output)
|
| 55 |
+
|
| 56 |
+
return clean_output
|
| 57 |
|
| 58 |
+
# Define Gradio interface
|
| 59 |
custom_css = """
|
| 60 |
.gradio-container {
|
| 61 |
background-color: #0D1525;
|
|
|
|
| 71 |
"""
|
| 72 |
|
| 73 |
def post_processing(prompt, completion):
|
| 74 |
+
"""
|
| 75 |
+
Formats generated text completion for display.
|
| 76 |
+
|
| 77 |
+
:param prompt: Input prompt for text generation.
|
| 78 |
+
:type prompt: str
|
| 79 |
+
:param completion: Generated text completion.
|
| 80 |
+
:type completion: str
|
| 81 |
+
:return: Formatted text completion.
|
| 82 |
+
:rtype: str
|
| 83 |
+
"""
|
| 84 |
return prompt + completion
|
|
|
|
|
|
|
|
|
|
|
|
|
| 85 |
|
|
|
|
| 86 |
def code_generation(prompt, max_new_tokens, temperature=0.2, seed=42, top_p=0.9, top_k=None, use_cache=True, repetition_penalty=1.0):
|
| 87 |
+
"""
|
| 88 |
+
Generates code completion given a prompt and specified parameters.
|
| 89 |
+
|
| 90 |
+
:param prompt: Input prompt for code generation.
|
| 91 |
+
:type prompt: str
|
| 92 |
+
:param max_new_tokens: Maximum number of tokens to generate.
|
| 93 |
+
:type max_new_tokens: int
|
| 94 |
+
:param temperature: Sampling temperature for code generation.
|
| 95 |
+
:type temperature: float
|
| 96 |
+
:param seed: Random seed for code generation.
|
| 97 |
+
:type seed: int
|
| 98 |
+
:param top_p: Value for top-p sampling.
|
| 99 |
+
:type top_p: float
|
| 100 |
+
:param top_k: Value for top-k sampling.
|
| 101 |
+
:type top_k: int
|
| 102 |
+
:param use_cache: Whether to use cache for code generation.
|
| 103 |
+
:type use_cache: bool
|
| 104 |
+
:param repetition_penalty: Value for repetition penalty.
|
| 105 |
+
:type repetition_penalty: float
|
| 106 |
+
:return: Generated code completion.
|
| 107 |
+
:rtype: str
|
| 108 |
+
"""
|
| 109 |
+
|
| 110 |
+
# Truncate prompt if too long
|
| 111 |
+
MAX_INPUT_TOKENS = 2048
|
| 112 |
+
if len(prompt) > MAX_INPUT_TOKENS:
|
| 113 |
+
prompt = prompt[-MAX_INPUT_TOKENS:]
|
| 114 |
+
|
| 115 |
+
# Tokenize prompt and move to device
|
| 116 |
x = tokenizer.encode(prompt, return_tensors="pt", max_length=MAX_INPUT_TOKENS, truncation=True).to(device)
|
| 117 |
+
logger.info("Prompt shape: %s", x.shape)
|
| 118 |
+
|
| 119 |
+
# Generate code completion using model and specified parameters
|
| 120 |
set_seed(seed)
|
| 121 |
y = model.generate(x,
|
| 122 |
max_new_tokens=max_new_tokens,
|
|
|
|
| 130 |
)
|
| 131 |
completion = tokenizer.decode(y[0], skip_special_tokens=True, clean_up_tokenization_spaces=False)
|
| 132 |
completion = completion[len(prompt):]
|
| 133 |
+
|
| 134 |
return post_processing(prompt, completion)
|
| 135 |
|
| 136 |
+
description = """
|
| 137 |
+
### Falcoder
|
| 138 |
|
| 139 |
+
Falcoder is a GPT-2 model fine-tuned on Python code. It can be used for generating code completions given a prompt.
|
|
|
|
|
|
|
| 140 |
|
| 141 |
+
### Text Generation
|
| 142 |
+
|
| 143 |
+
Use the text generation section to generate text completions given a prompt. You can adjust the maximum length of the generated text, whether to use sampling, the sampling temperature, and the top-k and top-p values for sampling.
|
| 144 |
+
|
| 145 |
+
### Code Generation
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 146 |
|
| 147 |
+
Use the code generation section to generate code completions given a prompt. You can adjust the maximum number of tokens to generate, the sampling temperature, the random seed, the top-p and top-k values for sampling, whether to use cache, and the repetition penalty.
|
| 148 |
+
"""
|
|
|
|
|
|
|
| 149 |
|
| 150 |
+
demo = gr.Interface(
|
| 151 |
+
[generate_text, code_generation],
|
| 152 |
+
["textbox", "textbox"],
|
| 153 |
+
["textbox", "textbox"],
|
| 154 |
+
title="Falcoder",
|
| 155 |
+
description=description,
|
| 156 |
+
theme="compact",
|
| 157 |
+
layout="vertical",
|
| 158 |
+
css=custom_css
|
| 159 |
+
)
|
| 160 |
|
| 161 |
+
# Launch Gradio interface
|
| 162 |
+
demo.launch()
|