Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,6 +1,5 @@
|
|
| 1 |
-
import time
|
| 2 |
-
import torch
|
| 3 |
import gradio as gr
|
|
|
|
| 4 |
from transformers import pipeline
|
| 5 |
from langchain_community.vectorstores import FAISS
|
| 6 |
from langchain_huggingface import HuggingFaceEmbeddings
|
|
@@ -13,7 +12,6 @@ print("Loading Lightweight Embedding Engine...")
|
|
| 13 |
embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
|
| 14 |
|
| 15 |
print("Initializing High-Performance Qwen Core on CPU...")
|
| 16 |
-
# PyTorch CPU Inference Engine Execution
|
| 17 |
chat_pipeline = pipeline(
|
| 18 |
"text-generation",
|
| 19 |
model="Qwen/Qwen2.5-1.5B-Instruct",
|
|
@@ -32,54 +30,44 @@ vector_store = FAISS.from_documents(documents, embedding_model)
|
|
| 32 |
|
| 33 |
|
| 34 |
# ==========================================
|
| 35 |
-
# 2.
|
| 36 |
# ==========================================
|
| 37 |
def execution_engine(workspace_type, user_input, max_tokens, temp, top_p):
|
| 38 |
if not user_input.strip():
|
| 39 |
return "β οΈ **System Input Error:** Workspace field cannot be processed empty."
|
| 40 |
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
# 1. System Prompt Engineering & Schema Enforcement
|
| 44 |
if workspace_type == "debate":
|
| 45 |
retriever = vector_store.as_retriever(search_kwargs={"k": 1})
|
| 46 |
matched_docs = retriever.invoke(user_input)
|
| 47 |
context = matched_docs[0].page_content if matched_docs else "No specific context available."
|
| 48 |
|
| 49 |
system_instruction = (
|
| 50 |
-
"You are
|
| 51 |
-
|
| 52 |
-
"
|
| 53 |
-
"###
|
| 54 |
-
"
|
| 55 |
-
"
|
| 56 |
-
"**Argument Logic Score:** [X/10]\n\n"
|
| 57 |
-
"### π·οΈ Categorized Insights\n"
|
| 58 |
-
"- **Factual Alignment:** [Brief note on data backing]\n"
|
| 59 |
-
"- **Fallacy/Risk Alert:** [Key vulnerability in the stance]"
|
| 60 |
)
|
| 61 |
-
else: # Generalize
|
| 62 |
system_instruction = (
|
| 63 |
-
"You are an
|
| 64 |
-
"
|
| 65 |
-
"###
|
| 66 |
-
"*
|
| 67 |
-
"*
|
| 68 |
-
"### π·οΈ Categorized Insights\n"
|
| 69 |
-
"- **Core Takeaway:** [One-sentence main point]\n"
|
| 70 |
-
"- **Strategic Impact:** [Market or operational implications]\n"
|
| 71 |
-
"- **Risk & Mitigation:** [Identified risk factor]"
|
| 72 |
)
|
| 73 |
|
| 74 |
-
# 2.
|
| 75 |
formatted_prompt = (
|
| 76 |
f"<|im_start|>system\n{system_instruction}\n"
|
| 77 |
-
"CRITICAL:
|
| 78 |
f"<|im_start|>user\n{user_input}<|im_end|>\n"
|
| 79 |
"<|im_start|>assistant\n"
|
| 80 |
)
|
| 81 |
|
| 82 |
-
# 3. CPU
|
| 83 |
try:
|
| 84 |
raw_output = chat_pipeline(
|
| 85 |
formatted_prompt,
|
|
@@ -93,78 +81,52 @@ def execution_engine(workspace_type, user_input, max_tokens, temp, top_p):
|
|
| 93 |
|
| 94 |
processed_text = raw_output[0]["generated_text"].split("<|im_start|>assistant\n")[-1].strip()
|
| 95 |
|
| 96 |
-
#
|
| 97 |
if processed_text and processed_text[-1] not in [".", "!", "?", '"', "*"]:
|
| 98 |
last_punctuation = max(processed_text.rfind("."), processed_text.rfind("!"), processed_text.rfind("?"))
|
| 99 |
if last_punctuation != -1:
|
| 100 |
processed_text = processed_text[:last_punctuation + 1]
|
| 101 |
-
|
| 102 |
-
|
| 103 |
-
latency = round(time.time() - start_time, 2)
|
| 104 |
-
prompt_tokens = len(formatted_prompt.split())
|
| 105 |
-
output_tokens = len(processed_text.split())
|
| 106 |
-
total_tokens = prompt_tokens + output_tokens
|
| 107 |
-
|
| 108 |
-
# Enterprise System Analytics Metadata Block
|
| 109 |
-
metadata_block = (
|
| 110 |
-
"\n\n---\n"
|
| 111 |
-
"### β‘ Runtime Analytics & Inference Metadata\n"
|
| 112 |
-
f"- **Inference Engine:** PyTorch CPU Direct Pipeline (`Qwen2.5-1.5B-Instruct`)\n"
|
| 113 |
-
f"- **Context Window Capacity:** 32,768 Tokens *(Used: {total_tokens} tokens)*\n"
|
| 114 |
-
f"- **Prompt-Level KV Caching:** Enabled *(Zero-copy Key-Value state reuse)*\n"
|
| 115 |
-
f"- **Structured Output Schema:** Enforced ChatML Schema Validation\n"
|
| 116 |
-
f"- **Execution Latency:** `{latency}s` | **Generation Speed:** ~{round(output_tokens/max(latency, 0.01), 1)} tokens/sec"
|
| 117 |
-
)
|
| 118 |
-
|
| 119 |
-
return processed_text + metadata_block
|
| 120 |
|
| 121 |
except Exception as error:
|
| 122 |
return f"β **Pipeline Fault:** Execution error on CPU node. Details: {str(error)}"
|
| 123 |
|
| 124 |
|
| 125 |
# ==========================================
|
| 126 |
-
# 3. MODULAR UI DASHBOARD (
|
| 127 |
# ==========================================
|
| 128 |
-
with gr.Blocks(
|
| 129 |
-
|
| 130 |
-
# Hero Title Header
|
| 131 |
-
gr.Markdown("# π€ Enterprise AI News Debate & Summary Agent Suite")
|
| 132 |
-
gr.Markdown(
|
| 133 |
-
"A low-latency, modular NLP workstation engineered for adversarial debate logic and automated "
|
| 134 |
-
"complex-news generalization. Optimized for high-throughput CPU inference with zero truncation errors."
|
| 135 |
-
)
|
| 136 |
|
| 137 |
-
#
|
| 138 |
with gr.Row():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 139 |
with gr.Column(scale=1):
|
| 140 |
-
gr.
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
|
| 147 |
|
| 148 |
gr.HTML("<hr style='border: 0.5px solid #374151; margin: 15px 0;'/>")
|
| 149 |
|
| 150 |
-
#
|
| 151 |
-
gr.Markdown("### π Platform Enterprise Capabilities")
|
| 152 |
-
with gr.Row():
|
| 153 |
-
with gr.Column(scale=1):
|
| 154 |
-
gr.Markdown("#### π Context Verification\nExtracts factual context from vector databases to verify user claims.")
|
| 155 |
-
with gr.Column(scale=1):
|
| 156 |
-
gr.Markdown("#### π Logic Evaluation\nScores debate arguments out of 10 based on structure and factual accuracy.")
|
| 157 |
-
with gr.Column(scale=1):
|
| 158 |
-
gr.Markdown("#### π·οΈ Categorized Insights\nBreaks outputs into strategic takeaways, risk alerts, and operational impacts.")
|
| 159 |
-
with gr.Column(scale=1):
|
| 160 |
-
gr.Markdown("#### β‘ Optimized Inference\nLeverages prompt-level KV caching and low-memory micro-models on CPU.")
|
| 161 |
-
|
| 162 |
-
gr.HTML("<hr style='border: 0.5px solid #374151; margin: 20px 0;'/>")
|
| 163 |
-
|
| 164 |
-
# Main Workspace Area
|
| 165 |
with gr.Row():
|
| 166 |
|
| 167 |
-
# Left Panel
|
| 168 |
with gr.Column(scale=1):
|
| 169 |
gr.Markdown("### π οΈ Agent Workspaces")
|
| 170 |
|
|
@@ -174,8 +136,9 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
| 174 |
with gr.Tab(label="ποΈ News Debate Agent", id=0):
|
| 175 |
gr.Markdown("*State a topic and your stance. The agent will pull factual context, debate you, and score your logic.*")
|
| 176 |
|
|
|
|
| 177 |
debate_example_picker = gr.Dropdown(
|
| 178 |
-
label="π‘ Preset Debate
|
| 179 |
choices=[
|
| 180 |
"LLMs are indispensable for application development.",
|
| 181 |
"Electric vehicles are the only way to save the environment.",
|
|
@@ -185,18 +148,21 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
| 185 |
|
| 186 |
debate_text_input = gr.Textbox(
|
| 187 |
label="Your Stance",
|
| 188 |
-
placeholder="Select or
|
| 189 |
)
|
| 190 |
|
|
|
|
| 191 |
debate_example_picker.change(fn=lambda x: x, inputs=debate_example_picker, outputs=debate_text_input)
|
|
|
|
| 192 |
run_debate_action = gr.Button("Deploy Counter-Argument & Score", variant="primary")
|
| 193 |
|
| 194 |
# Tab (ii): Text Simplification Module
|
| 195 |
with gr.Tab(label="π° Snippet Generalizer", id=1):
|
| 196 |
-
gr.Markdown("*Paste complex, jargon-heavy news columns here to output plain-English executive summaries.*")
|
| 197 |
|
|
|
|
| 198 |
generalizer_example_picker = gr.Dropdown(
|
| 199 |
-
label="π‘ Preset
|
| 200 |
choices=[
|
| 201 |
"Despite aggressive macroeconomic headwinds and structural quantitative tightening by central banking systems, retail index trends demonstrated non-linear elasticity during fiscal evaluations.",
|
| 202 |
"The ecosystem validation vectors indicate substantial degradation in sub-surface lithospheric layers owing to chemical manufacturing runoffs, sparking compliance friction."
|
|
@@ -211,22 +177,25 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
| 211 |
placeholder="Paste text block or official reports here..."
|
| 212 |
)
|
| 213 |
|
|
|
|
| 214 |
generalizer_example_picker.change(fn=lambda x: x, inputs=generalizer_example_picker, outputs=simplify_text_input)
|
| 215 |
-
|
|
|
|
| 216 |
|
| 217 |
gr.HTML("<div style='margin-bottom: 20px;'></div>")
|
|
|
|
| 218 |
|
| 219 |
-
# Param Configuration
|
| 220 |
with gr.Accordion("βοΈ Runtime Parameter Tuning", open=False):
|
| 221 |
-
max_tokens_val = gr.Slider(minimum=64, maximum=
|
| 222 |
temp_val = gr.Slider(minimum=0.1, maximum=1.0, value=0.25, step=0.05, label="Temperature (Low = Deterministic)")
|
| 223 |
top_p_val = gr.Slider(minimum=0.50, maximum=1.00, value=0.85, step=0.05, label="Top-p")
|
| 224 |
|
| 225 |
-
# Right Panel Output
|
| 226 |
with gr.Column(scale=1):
|
| 227 |
gr.Markdown("### π₯οΈ Active Output Stream")
|
| 228 |
active_output_panel = gr.Markdown(
|
| 229 |
-
value="*System ready. Select a workspace on the left panel and click deploy to review
|
| 230 |
)
|
| 231 |
|
| 232 |
# ==========================================
|
|
@@ -245,4 +214,4 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
| 245 |
)
|
| 246 |
|
| 247 |
if __name__ == "__main__":
|
| 248 |
-
demo.launch(server_name="0.0.0.0", server_port=7860)
|
|
|
|
|
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
+
import torch
|
| 3 |
from transformers import pipeline
|
| 4 |
from langchain_community.vectorstores import FAISS
|
| 5 |
from langchain_huggingface import HuggingFaceEmbeddings
|
|
|
|
| 12 |
embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
|
| 13 |
|
| 14 |
print("Initializing High-Performance Qwen Core on CPU...")
|
|
|
|
| 15 |
chat_pipeline = pipeline(
|
| 16 |
"text-generation",
|
| 17 |
model="Qwen/Qwen2.5-1.5B-Instruct",
|
|
|
|
| 30 |
|
| 31 |
|
| 32 |
# ==========================================
|
| 33 |
+
# 2. FEW-SHOT PROMPT COMPILER
|
| 34 |
# ==========================================
|
| 35 |
def execution_engine(workspace_type, user_input, max_tokens, temp, top_p):
|
| 36 |
if not user_input.strip():
|
| 37 |
return "β οΈ **System Input Error:** Workspace field cannot be processed empty."
|
| 38 |
|
| 39 |
+
# 1. Structural Instructions + Few-Shot Examples (The Quality Anchors)
|
|
|
|
|
|
|
| 40 |
if workspace_type == "debate":
|
| 41 |
retriever = vector_store.as_retriever(search_kwargs={"k": 1})
|
| 42 |
matched_docs = retriever.invoke(user_input)
|
| 43 |
context = matched_docs[0].page_content if matched_docs else "No specific context available."
|
| 44 |
|
| 45 |
system_instruction = (
|
| 46 |
+
"You are a sharp, polite adversarial debate assistant. Disagree with the user's stance. "
|
| 47 |
+
"Provide exactly one paragraph of logical counter-argument, then provide a score out of 10.\n\n"
|
| 48 |
+
f"VERIFIED CONTEXT TO USE: {context}\n\n"
|
| 49 |
+
"### GOOD EXAMPLE RESPONSE FORMAT:\n"
|
| 50 |
+
"While your point highlights consumer convenience, verified data shows that global emissions from battery manufacturing offset initial green gains by up to 40%. Therefore, an immediate full transition remains structurally flawed.\n\n"
|
| 51 |
+
"**Argument Logic Score: 6/10**"
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
)
|
| 53 |
+
else: # Generalize mode
|
| 54 |
system_instruction = (
|
| 55 |
+
"You are an expert news editor. Translate complex, jargon-heavy statements into plain, simple English. "
|
| 56 |
+
"Output your answer as clear, high-level summary bullet points under 50 words total.\n\n"
|
| 57 |
+
"### GOOD EXAMPLE RESPONSE FORMAT:\n"
|
| 58 |
+
"* Global markets fell sharply due to sudden central bank interest rate changes.\n"
|
| 59 |
+
"* Local manufacturing activity slowed down to an all-time low."
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
)
|
| 61 |
|
| 62 |
+
# 2. Construct Prompt using Official Qwen ChatML Template format
|
| 63 |
formatted_prompt = (
|
| 64 |
f"<|im_start|>system\n{system_instruction}\n"
|
| 65 |
+
"CRITICAL: Keep your response concise. Do not ramble. Complete your final sentence perfectly.<|im_end|>\n"
|
| 66 |
f"<|im_start|>user\n{user_input}<|im_end|>\n"
|
| 67 |
"<|im_start|>assistant\n"
|
| 68 |
)
|
| 69 |
|
| 70 |
+
# 3. CPU Execution Pipeline
|
| 71 |
try:
|
| 72 |
raw_output = chat_pipeline(
|
| 73 |
formatted_prompt,
|
|
|
|
| 81 |
|
| 82 |
processed_text = raw_output[0]["generated_text"].split("<|im_start|>assistant\n")[-1].strip()
|
| 83 |
|
| 84 |
+
# Programmatic Clean Up: Ensure no trailing cut-off sentences are displayed
|
| 85 |
if processed_text and processed_text[-1] not in [".", "!", "?", '"', "*"]:
|
| 86 |
last_punctuation = max(processed_text.rfind("."), processed_text.rfind("!"), processed_text.rfind("?"))
|
| 87 |
if last_punctuation != -1:
|
| 88 |
processed_text = processed_text[:last_punctuation + 1]
|
| 89 |
+
|
| 90 |
+
return processed_text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 91 |
|
| 92 |
except Exception as error:
|
| 93 |
return f"β **Pipeline Fault:** Execution error on CPU node. Details: {str(error)}"
|
| 94 |
|
| 95 |
|
| 96 |
# ==========================================
|
| 97 |
+
# 3. MODULAR UI DASHBOARD (GRADIO 6.0)
|
| 98 |
# ==========================================
|
| 99 |
+
with gr.Blocks() as demo:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 100 |
|
| 101 |
+
# Header Module
|
| 102 |
with gr.Row():
|
| 103 |
+
with gr.Column(scale=3):
|
| 104 |
+
gr.Markdown("# π€ AI News Debate & Summary Agent Suite")
|
| 105 |
+
gr.Markdown(
|
| 106 |
+
"**About This Agent:** This framework utilizes localized vector embeddings alongside "
|
| 107 |
+
"few-shot prompting to anchor small model reasoning. It delivers deterministic, "
|
| 108 |
+
"concise outputs optimized for zero-error execution on base CPU clusters. "
|
| 109 |
+
"This framework processes user statements and news artifacts to provide dual NLP functions: "
|
| 110 |
+
"*Adversarial Debate Environment* using real-time vectorized context verification to engage with users "
|
| 111 |
+
"and *Automated Generalization Engine* for complex-news summarization and analysis."
|
| 112 |
+
)
|
| 113 |
+
|
| 114 |
+
# Status Module Panel
|
| 115 |
with gr.Column(scale=1):
|
| 116 |
+
gr.HighlightedText(value=[("ONLINE", "π’ NLP Match Engine Status")], show_label=False)
|
| 117 |
+
gr.Markdown(
|
| 118 |
+
"πͺͺ **Agent Profile:** Few-Shot Context Assistant \n"
|
| 119 |
+
"π¦ **Model Matrix:** Qwen2.5-1.5B-Instruct \n"
|
| 120 |
+
"𧬠**Vectorization Engine:** Text FAISS Engine \n"
|
| 121 |
+
"π» **Processing Hardware:** Base CPU Tier"
|
| 122 |
+
)
|
| 123 |
|
| 124 |
gr.HTML("<hr style='border: 0.5px solid #374151; margin: 15px 0;'/>")
|
| 125 |
|
| 126 |
+
# Lower Workspace Splitting Layout
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 127 |
with gr.Row():
|
| 128 |
|
| 129 |
+
# Left Panel Workspace
|
| 130 |
with gr.Column(scale=1):
|
| 131 |
gr.Markdown("### π οΈ Agent Workspaces")
|
| 132 |
|
|
|
|
| 136 |
with gr.Tab(label="ποΈ News Debate Agent", id=0):
|
| 137 |
gr.Markdown("*State a topic and your stance. The agent will pull factual context, debate you, and score your logic.*")
|
| 138 |
|
| 139 |
+
# Robust Example System: Swapped to an immune dropdown component to bypass the hidden tab rendering bug
|
| 140 |
debate_example_picker = gr.Dropdown(
|
| 141 |
+
label="π‘ Suggestions for Preset Debate Topic",
|
| 142 |
choices=[
|
| 143 |
"LLMs are indispensable for application development.",
|
| 144 |
"Electric vehicles are the only way to save the environment.",
|
|
|
|
| 148 |
|
| 149 |
debate_text_input = gr.Textbox(
|
| 150 |
label="Your Stance",
|
| 151 |
+
placeholder="Select or paste your claim here..."
|
| 152 |
)
|
| 153 |
|
| 154 |
+
# Dynamically push selected dropdown choice into the text input instantly
|
| 155 |
debate_example_picker.change(fn=lambda x: x, inputs=debate_example_picker, outputs=debate_text_input)
|
| 156 |
+
|
| 157 |
run_debate_action = gr.Button("Deploy Counter-Argument & Score", variant="primary")
|
| 158 |
|
| 159 |
# Tab (ii): Text Simplification Module
|
| 160 |
with gr.Tab(label="π° Snippet Generalizer", id=1):
|
| 161 |
+
gr.Markdown("*Paste complex, jargon-heavy news columns here to output plain-English, executive summaries.*")
|
| 162 |
|
| 163 |
+
# Robust Example System: Swapped to an immune dropdown component to bypass the hidden tab rendering bug
|
| 164 |
generalizer_example_picker = gr.Dropdown(
|
| 165 |
+
label="π‘ Examples for Preset Complex News",
|
| 166 |
choices=[
|
| 167 |
"Despite aggressive macroeconomic headwinds and structural quantitative tightening by central banking systems, retail index trends demonstrated non-linear elasticity during fiscal evaluations.",
|
| 168 |
"The ecosystem validation vectors indicate substantial degradation in sub-surface lithospheric layers owing to chemical manufacturing runoffs, sparking compliance friction."
|
|
|
|
| 177 |
placeholder="Paste text block or official reports here..."
|
| 178 |
)
|
| 179 |
|
| 180 |
+
# Dynamically push selected dropdown choice into the text input instantly
|
| 181 |
generalizer_example_picker.change(fn=lambda x: x, inputs=generalizer_example_picker, outputs=simplify_text_input)
|
| 182 |
+
|
| 183 |
+
run_simplifier_action = gr.Button("Deploy for generalization system", variant="primary")
|
| 184 |
|
| 185 |
gr.HTML("<div style='margin-bottom: 20px;'></div>")
|
| 186 |
+
|
| 187 |
|
| 188 |
+
# Param Configuration Module (Hidden Tuning to keep CPU fast and reliable)
|
| 189 |
with gr.Accordion("βοΈ Runtime Parameter Tuning", open=False):
|
| 190 |
+
max_tokens_val = gr.Slider(minimum=64, maximum=256, value=140, step=16, label="Max tokens")
|
| 191 |
temp_val = gr.Slider(minimum=0.1, maximum=1.0, value=0.25, step=0.05, label="Temperature (Low = Deterministic)")
|
| 192 |
top_p_val = gr.Slider(minimum=0.50, maximum=1.00, value=0.85, step=0.05, label="Top-p")
|
| 193 |
|
| 194 |
+
# Right Panel Output Monitor
|
| 195 |
with gr.Column(scale=1):
|
| 196 |
gr.Markdown("### π₯οΈ Active Output Stream")
|
| 197 |
active_output_panel = gr.Markdown(
|
| 198 |
+
value="*System ready. Select a workspace on the left panel and click deploy to review parsed generation stream.*"
|
| 199 |
)
|
| 200 |
|
| 201 |
# ==========================================
|
|
|
|
| 214 |
)
|
| 215 |
|
| 216 |
if __name__ == "__main__":
|
| 217 |
+
demo.launch(server_name="0.0.0.0", server_port=7860, theme=gr.themes.Soft())
|