Update app.py
Browse files
app.py
CHANGED
@@ -1,63 +1,84 @@
|
|
1 |
import os
|
2 |
import gradio as gr
|
3 |
-
from transformers import LlamaTokenizer, AutoModelForCausalLM
|
4 |
import torch
|
5 |
import json
|
|
|
6 |
|
7 |
# Set Hugging Face Token for Authentication (ensure it's set in your environment)
|
8 |
HUGGINGFACE_TOKEN = os.getenv("HUGGINGFACE_TOKEN")
|
9 |
|
10 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
MODEL_NAME = "meta-llama/Llama-3.2-1B-Instruct-QLORA_INT4_EO8"
|
12 |
-
tokenizer =
|
13 |
-
|
14 |
-
|
15 |
-
token=HUGGINGFACE_TOKEN,
|
16 |
-
device_map="cpu"
|
17 |
-
)
|
18 |
-
|
19 |
-
# Load Llama Guard for content moderation on CPU
|
20 |
LLAMA_GUARD_NAME = "meta-llama/Llama-Guard-3-1B-INT4"
|
21 |
-
guard_tokenizer =
|
22 |
-
guard_model = AutoModelForCausalLM.from_pretrained(
|
23 |
-
LLAMA_GUARD_NAME,
|
24 |
-
token=HUGGINGFACE_TOKEN,
|
25 |
-
device_map="cpu"
|
26 |
-
)
|
27 |
|
28 |
# Define Prompt Templates
|
29 |
PROMPTS = {
|
30 |
-
"project_analysis": """Analyze this project description and generate:
|
31 |
1. Project timeline with milestones
|
32 |
2. Required technology stack
|
33 |
3. Potential risks
|
34 |
4. Team composition
|
35 |
5. Cost estimation
|
36 |
|
37 |
-
Project: {project_description}""",
|
38 |
|
39 |
-
"code_generation": """Generate implementation code for this feature:
|
40 |
{feature_description}
|
41 |
|
42 |
Considerations:
|
43 |
- Use {programming_language}
|
44 |
- Follow {coding_standards}
|
45 |
- Include error handling
|
46 |
-
- Add documentation""",
|
47 |
|
48 |
-
"risk_analysis": """Predict potential risks for this project plan:
|
49 |
{project_data}
|
50 |
|
51 |
-
Format output as JSON with risk types, probabilities, and mitigation strategies"""
|
52 |
}
|
53 |
|
54 |
# Function: Content Moderation using Llama Guard
|
55 |
def moderate_input(user_input):
|
56 |
-
|
57 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
58 |
response = guard_tokenizer.decode(outputs[0], skip_special_tokens=True)
|
59 |
|
60 |
-
if "flagged" in response.lower():
|
61 |
return "⚠️ Content flagged by Llama Guard. Please modify your input."
|
62 |
return None # Safe input, proceed normally
|
63 |
|
@@ -69,14 +90,16 @@ def generate_response(prompt_type, **kwargs):
|
|
69 |
if moderation_warning:
|
70 |
return moderation_warning # Stop processing if flagged
|
71 |
|
72 |
-
inputs = tokenizer(prompt, return_tensors="pt",
|
73 |
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
|
|
|
|
80 |
|
81 |
return tokenizer.decode(outputs[0], skip_special_tokens=True)
|
82 |
|
@@ -92,7 +115,12 @@ def generate_code(feature_desc, lang="Python", standards="PEP8"):
|
|
92 |
def predict_risks(project_data):
|
93 |
risks = generate_response("risk_analysis", project_data=project_data)
|
94 |
try:
|
95 |
-
|
|
|
|
|
|
|
|
|
|
|
96 |
except json.JSONDecodeError:
|
97 |
return {"error": "Invalid JSON response. Please refine your input."}
|
98 |
|
@@ -104,7 +132,7 @@ def create_gradio_interface():
|
|
104 |
# Project Analysis Tab
|
105 |
with gr.Tab("Project Setup"):
|
106 |
project_input = gr.Textbox(label="Project Description", lines=5, placeholder="Describe your project...")
|
107 |
-
project_output = gr.
|
108 |
analyze_btn = gr.Button("Analyze Project")
|
109 |
analyze_btn.click(analyze_project, inputs=project_input, outputs=project_output)
|
110 |
|
@@ -137,14 +165,27 @@ def create_gradio_interface():
|
|
137 |
chat_history.append((message, moderation_warning))
|
138 |
return "", chat_history
|
139 |
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
145 |
|
146 |
-
inputs = tokenizer(prompt, return_tensors="pt")
|
147 |
-
outputs = model.generate(inputs.input_ids, max_length=1024)
|
148 |
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
149 |
chat_history.append((message, response))
|
150 |
return "", chat_history
|
@@ -157,4 +198,4 @@ def create_gradio_interface():
|
|
157 |
# Run Gradio App
|
158 |
if __name__ == "__main__":
|
159 |
interface = create_gradio_interface()
|
160 |
-
interface.launch(share=True)
|
|
|
1 |
import os
|
2 |
import gradio as gr
|
|
|
3 |
import torch
|
4 |
import json
|
5 |
+
from transformers import AutoTokenizer
|
6 |
|
7 |
# Set Hugging Face Token for Authentication (ensure it's set in your environment)
|
8 |
HUGGINGFACE_TOKEN = os.getenv("HUGGINGFACE_TOKEN")
|
9 |
|
10 |
+
# Function to load Llama model
|
11 |
+
def load_llama_model(model_name):
|
12 |
+
from transformers import LlamaForCausalLM, LlamaTokenizer
|
13 |
+
|
14 |
+
# Use AutoTokenizer which will handle various tokenizer types
|
15 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name, token=HUGGINGFACE_TOKEN, use_fast=False)
|
16 |
+
|
17 |
+
# Use the LlamaForCausalLM class which can properly load the consolidated.00.pth format
|
18 |
+
model = LlamaForCausalLM.from_pretrained(
|
19 |
+
model_name,
|
20 |
+
token=HUGGINGFACE_TOKEN,
|
21 |
+
torch_dtype=torch.float16, # Use float16 to reduce memory usage on CPU
|
22 |
+
low_cpu_mem_usage=True, # Optimize for low memory usage
|
23 |
+
device_map="cpu"
|
24 |
+
)
|
25 |
+
|
26 |
+
return tokenizer, model
|
27 |
+
|
28 |
+
# Load Llama 3.2 model
|
29 |
MODEL_NAME = "meta-llama/Llama-3.2-1B-Instruct-QLORA_INT4_EO8"
|
30 |
+
tokenizer, model = load_llama_model(MODEL_NAME)
|
31 |
+
|
32 |
+
# Load Llama Guard for content moderation
|
|
|
|
|
|
|
|
|
|
|
33 |
LLAMA_GUARD_NAME = "meta-llama/Llama-Guard-3-1B-INT4"
|
34 |
+
guard_tokenizer, guard_model = load_llama_model(LLAMA_GUARD_NAME)
|
|
|
|
|
|
|
|
|
|
|
35 |
|
36 |
# Define Prompt Templates
|
37 |
PROMPTS = {
|
38 |
+
"project_analysis": """<|begin_of_text|><|prompt|>Analyze this project description and generate:
|
39 |
1. Project timeline with milestones
|
40 |
2. Required technology stack
|
41 |
3. Potential risks
|
42 |
4. Team composition
|
43 |
5. Cost estimation
|
44 |
|
45 |
+
Project: {project_description}<|completion|>""",
|
46 |
|
47 |
+
"code_generation": """<|begin_of_text|><|prompt|>Generate implementation code for this feature:
|
48 |
{feature_description}
|
49 |
|
50 |
Considerations:
|
51 |
- Use {programming_language}
|
52 |
- Follow {coding_standards}
|
53 |
- Include error handling
|
54 |
+
- Add documentation<|completion|>""",
|
55 |
|
56 |
+
"risk_analysis": """<|begin_of_text|><|prompt|>Predict potential risks for this project plan:
|
57 |
{project_data}
|
58 |
|
59 |
+
Format output as JSON with risk types, probabilities, and mitigation strategies<|completion|>"""
|
60 |
}
|
61 |
|
62 |
# Function: Content Moderation using Llama Guard
|
63 |
def moderate_input(user_input):
|
64 |
+
# Llama Guard specific prompt format
|
65 |
+
prompt = f"""<|begin_of_text|><|user|>
|
66 |
+
Input: {user_input}
|
67 |
+
Please verify that this input doesn't violate any content policies.
|
68 |
+
<|assistant|>"""
|
69 |
+
|
70 |
+
inputs = guard_tokenizer(prompt, return_tensors="pt", truncation=True)
|
71 |
+
|
72 |
+
with torch.no_grad(): # Disable gradient calculation for inference
|
73 |
+
outputs = guard_model.generate(
|
74 |
+
inputs.input_ids,
|
75 |
+
max_length=256,
|
76 |
+
temperature=0.1
|
77 |
+
)
|
78 |
+
|
79 |
response = guard_tokenizer.decode(outputs[0], skip_special_tokens=True)
|
80 |
|
81 |
+
if "flagged" in response.lower() or "violated" in response.lower() or "policy violation" in response.lower():
|
82 |
return "⚠️ Content flagged by Llama Guard. Please modify your input."
|
83 |
return None # Safe input, proceed normally
|
84 |
|
|
|
90 |
if moderation_warning:
|
91 |
return moderation_warning # Stop processing if flagged
|
92 |
|
93 |
+
inputs = tokenizer(prompt, return_tensors="pt", truncation=True)
|
94 |
|
95 |
+
with torch.no_grad(): # Disable gradient calculation for inference
|
96 |
+
outputs = model.generate(
|
97 |
+
inputs.input_ids,
|
98 |
+
max_length=1024,
|
99 |
+
temperature=0.7 if prompt_type == "project_analysis" else 0.5,
|
100 |
+
top_p=0.9,
|
101 |
+
do_sample=True
|
102 |
+
)
|
103 |
|
104 |
return tokenizer.decode(outputs[0], skip_special_tokens=True)
|
105 |
|
|
|
115 |
def predict_risks(project_data):
|
116 |
risks = generate_response("risk_analysis", project_data=project_data)
|
117 |
try:
|
118 |
+
# Try to extract JSON part from the response
|
119 |
+
import re
|
120 |
+
json_match = re.search(r'\{.*\}', risks, re.DOTALL)
|
121 |
+
if json_match:
|
122 |
+
return json.loads(json_match.group(0))
|
123 |
+
return {"error": "Could not parse JSON response"}
|
124 |
except json.JSONDecodeError:
|
125 |
return {"error": "Invalid JSON response. Please refine your input."}
|
126 |
|
|
|
132 |
# Project Analysis Tab
|
133 |
with gr.Tab("Project Setup"):
|
134 |
project_input = gr.Textbox(label="Project Description", lines=5, placeholder="Describe your project...")
|
135 |
+
project_output = gr.Textbox(label="Project Analysis", lines=15) # Changed from JSON to Textbox for better formatting
|
136 |
analyze_btn = gr.Button("Analyze Project")
|
137 |
analyze_btn.click(analyze_project, inputs=project_input, outputs=project_output)
|
138 |
|
|
|
165 |
chat_history.append((message, moderation_warning))
|
166 |
return "", chat_history
|
167 |
|
168 |
+
# Format chat history for context
|
169 |
+
history_text = ""
|
170 |
+
for i, (usr, ai) in enumerate(chat_history[-3:]): # Use last 3 messages for context
|
171 |
+
history_text += f"User: {usr}\nAI: {ai}\n"
|
172 |
+
|
173 |
+
prompt = f"""<|begin_of_text|><|prompt|>Project Management Chat:
|
174 |
+
Context: {message}
|
175 |
+
Chat History: {history_text}
|
176 |
+
User: {message}<|completion|>"""
|
177 |
+
|
178 |
+
inputs = tokenizer(prompt, return_tensors="pt", truncation=True)
|
179 |
+
|
180 |
+
with torch.no_grad():
|
181 |
+
outputs = model.generate(
|
182 |
+
inputs.input_ids,
|
183 |
+
max_length=1024,
|
184 |
+
temperature=0.7,
|
185 |
+
top_p=0.9,
|
186 |
+
do_sample=True
|
187 |
+
)
|
188 |
|
|
|
|
|
189 |
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
190 |
chat_history.append((message, response))
|
191 |
return "", chat_history
|
|
|
198 |
# Run Gradio App
|
199 |
if __name__ == "__main__":
|
200 |
interface = create_gradio_interface()
|
201 |
+
interface.launch(share=True)
|