Update app.py
Browse files
app.py
CHANGED
|
@@ -1,12 +1,10 @@
|
|
| 1 |
-
import os
|
| 2 |
-
import math
|
| 3 |
import torch
|
| 4 |
import torch.nn as nn
|
| 5 |
from torch.nn import functional as F
|
| 6 |
-
import gradio as gr
|
| 7 |
import tiktoken
|
|
|
|
| 8 |
|
| 9 |
-
#
|
| 10 |
class GPTConfig:
|
| 11 |
def __init__(self):
|
| 12 |
self.block_size = 1024
|
|
@@ -100,7 +98,7 @@ class GPT(nn.Module):
|
|
| 100 |
|
| 101 |
return logits, loss
|
| 102 |
|
| 103 |
-
#
|
| 104 |
def load_model(model_path):
|
| 105 |
config = GPTConfig()
|
| 106 |
model = GPT(config)
|
|
@@ -110,46 +108,63 @@ def load_model(model_path):
|
|
| 110 |
print("Checkpoint keys:", checkpoint.keys()) # Debug print
|
| 111 |
|
| 112 |
if 'model_state_dict' in checkpoint:
|
| 113 |
-
# If the checkpoint contains a 'model_state_dict' key, use that
|
| 114 |
model.load_state_dict(checkpoint['model_state_dict'])
|
| 115 |
else:
|
| 116 |
-
# Otherwise, try to load the state dict directly
|
| 117 |
model.load_state_dict(checkpoint)
|
| 118 |
|
| 119 |
model.eval()
|
| 120 |
return model
|
| 121 |
|
| 122 |
-
# Load the
|
| 123 |
model = load_model('gpt_5000.pt') # Replace with the actual path to your .pt file
|
| 124 |
enc = tiktoken.get_encoding('gpt2')
|
| 125 |
|
| 126 |
-
|
|
|
|
| 127 |
input_ids = torch.tensor(enc.encode(prompt)).unsqueeze(0)
|
|
|
|
| 128 |
|
| 129 |
with torch.no_grad():
|
| 130 |
for _ in range(max_length):
|
| 131 |
outputs, _ = model(input_ids)
|
| 132 |
-
next_token_logits = outputs[:, -1, :]
|
| 133 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 134 |
input_ids = torch.cat([input_ids, next_token], dim=-1)
|
|
|
|
| 135 |
|
| 136 |
-
if
|
|
|
|
| 137 |
break
|
| 138 |
|
| 139 |
-
generated_text = enc.decode(
|
| 140 |
-
return generated_text
|
| 141 |
|
| 142 |
# Gradio interface
|
|
|
|
|
|
|
|
|
|
| 143 |
iface = gr.Interface(
|
| 144 |
-
fn=
|
| 145 |
inputs=[
|
| 146 |
gr.Textbox(label="Prompt", placeholder="Enter your prompt here..."),
|
| 147 |
-
gr.Slider(minimum=
|
| 148 |
-
gr.Slider(minimum=0.1, maximum=1.0, value=0.7, step=0.1, label="Temperature")
|
|
|
|
| 149 |
],
|
| 150 |
outputs=gr.Textbox(label="Generated Text"),
|
| 151 |
-
title="GPT
|
| 152 |
-
description="Enter a prompt and generate text using a fine-tuned GPT
|
| 153 |
)
|
| 154 |
|
| 155 |
# Launch the app
|
|
|
|
|
|
|
|
|
|
| 1 |
import torch
|
| 2 |
import torch.nn as nn
|
| 3 |
from torch.nn import functional as F
|
|
|
|
| 4 |
import tiktoken
|
| 5 |
+
import gradio as gr
|
| 6 |
|
| 7 |
+
# Define the model architecture
|
| 8 |
class GPTConfig:
|
| 9 |
def __init__(self):
|
| 10 |
self.block_size = 1024
|
|
|
|
| 98 |
|
| 99 |
return logits, loss
|
| 100 |
|
| 101 |
+
# Load the model
|
| 102 |
def load_model(model_path):
|
| 103 |
config = GPTConfig()
|
| 104 |
model = GPT(config)
|
|
|
|
| 108 |
print("Checkpoint keys:", checkpoint.keys()) # Debug print
|
| 109 |
|
| 110 |
if 'model_state_dict' in checkpoint:
|
|
|
|
| 111 |
model.load_state_dict(checkpoint['model_state_dict'])
|
| 112 |
else:
|
|
|
|
| 113 |
model.load_state_dict(checkpoint)
|
| 114 |
|
| 115 |
model.eval()
|
| 116 |
return model
|
| 117 |
|
| 118 |
+
# Load the model
|
| 119 |
model = load_model('gpt_5000.pt') # Replace with the actual path to your .pt file
|
| 120 |
enc = tiktoken.get_encoding('gpt2')
|
| 121 |
|
| 122 |
+
# Improved text generation function
|
| 123 |
+
def generate_text(prompt, max_length=100, temperature=0.7, top_k=50):
|
| 124 |
input_ids = torch.tensor(enc.encode(prompt)).unsqueeze(0)
|
| 125 |
+
generated = []
|
| 126 |
|
| 127 |
with torch.no_grad():
|
| 128 |
for _ in range(max_length):
|
| 129 |
outputs, _ = model(input_ids)
|
| 130 |
+
next_token_logits = outputs[:, -1, :]
|
| 131 |
+
|
| 132 |
+
# Apply temperature
|
| 133 |
+
next_token_logits = next_token_logits / temperature
|
| 134 |
+
|
| 135 |
+
# Apply top-k filtering
|
| 136 |
+
top_k_logits, top_k_indices = torch.topk(next_token_logits, top_k, dim=-1)
|
| 137 |
+
next_token_probs = F.softmax(top_k_logits, dim=-1)
|
| 138 |
+
|
| 139 |
+
# Sample from the filtered distribution
|
| 140 |
+
next_token_index = torch.multinomial(next_token_probs, num_samples=1)
|
| 141 |
+
next_token = top_k_indices.gather(-1, next_token_index)
|
| 142 |
+
|
| 143 |
input_ids = torch.cat([input_ids, next_token], dim=-1)
|
| 144 |
+
generated.append(next_token.item())
|
| 145 |
|
| 146 |
+
# Stop if we generate a newline, but only after generating at least 20 tokens
|
| 147 |
+
if next_token.item() == enc.encode('\n')[0] and len(generated) > 20:
|
| 148 |
break
|
| 149 |
|
| 150 |
+
generated_text = enc.decode(generated)
|
| 151 |
+
return prompt + generated_text
|
| 152 |
|
| 153 |
# Gradio interface
|
| 154 |
+
def gradio_generate(prompt, max_length, temperature, top_k):
|
| 155 |
+
return generate_text(prompt, max_length, temperature, top_k)
|
| 156 |
+
|
| 157 |
iface = gr.Interface(
|
| 158 |
+
fn=gradio_generate,
|
| 159 |
inputs=[
|
| 160 |
gr.Textbox(label="Prompt", placeholder="Enter your prompt here..."),
|
| 161 |
+
gr.Slider(minimum=20, maximum=500, value=100, step=1, label="Max Length"),
|
| 162 |
+
gr.Slider(minimum=0.1, maximum=1.0, value=0.7, step=0.1, label="Temperature"),
|
| 163 |
+
gr.Slider(minimum=1, maximum=100, value=50, step=1, label="Top-k")
|
| 164 |
],
|
| 165 |
outputs=gr.Textbox(label="Generated Text"),
|
| 166 |
+
title="GPT Text Generator",
|
| 167 |
+
description="Enter a prompt and adjust parameters to generate text using a fine-tuned GPT model."
|
| 168 |
)
|
| 169 |
|
| 170 |
# Launch the app
|