Spaces:
Running
Running
Commit
Β·
ab6a69a
1
Parent(s):
92ae1b2
Modified the prompt and confidence score
Browse files- README.md +1 -3
- app.py +96 -45
- data_filters.py +5 -5
README.md
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
---
|
2 |
-
title:
|
3 |
emoji: π
|
4 |
colorFrom: green
|
5 |
colorTo: indigo
|
@@ -10,5 +10,3 @@ pinned: false
|
|
10 |
license: unknown
|
11 |
short_description: SLM with RAG for Financial Reports
|
12 |
---
|
13 |
-
|
14 |
-
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
|
|
1 |
---
|
2 |
+
title: SLM Financial RAG
|
3 |
emoji: π
|
4 |
colorFrom: green
|
5 |
colorTo: indigo
|
|
|
10 |
license: unknown
|
11 |
short_description: SLM with RAG for Financial Reports
|
12 |
---
|
|
|
|
app.py
CHANGED
@@ -37,7 +37,8 @@ os.makedirs("data", exist_ok=True)
|
|
37 |
# SLM: Microsoft PHI-2 model is loaded
|
38 |
# It does have higher memory and compute requirements compared to TinyLlama and Falcon
|
39 |
# But it gives the best results among the three
|
40 |
-
DEVICE = "cpu" # or cuda
|
|
|
41 |
# MODEL_NAME = "TinyLlama/TinyLlama_v1.1"
|
42 |
# MODEL_NAME = "tiiuae/falcon-rw-1b"
|
43 |
MODEL_NAME = "microsoft/phi-2"
|
@@ -54,7 +55,7 @@ if tokenizer.pad_token is None:
|
|
54 |
# Since the model is to be hosted on a cpu instance, we use float32
|
55 |
# For GPU, we can use float16 or bfloat16
|
56 |
model = AutoModelForCausalLM.from_pretrained(
|
57 |
-
MODEL_NAME, torch_dtype=torch.
|
58 |
).to(DEVICE)
|
59 |
model.eval()
|
60 |
# model = torch.quantization.quantize_dynamic(model, {torch.nn.Linear}, dtype=torch.qint8)
|
@@ -233,6 +234,13 @@ def process_files(files, chunk_size=512):
|
|
233 |
pickle.dump(bm25_data, f)
|
234 |
return "Files processed successfully! You can now query."
|
235 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
236 |
|
237 |
# Input guardrail implementation
|
238 |
# Regex is used to filter queries related to sensitive topics
|
@@ -242,7 +250,7 @@ def process_files(files, chunk_size=512):
|
|
242 |
def is_query_allowed(query):
|
243 |
"""Checks if the query violates security or confidentiality rules"""
|
244 |
for pattern in restricted_patterns:
|
245 |
-
if re.search(pattern, query, re.IGNORECASE):
|
246 |
return False, "This query requests sensitive or confidential information."
|
247 |
doc = nlp(query)
|
248 |
for ent in doc.ents:
|
@@ -309,45 +317,65 @@ def hybrid_retrieve(query, chunk_texts, index, bm25, top_k=5, lambda_faiss=0.7):
|
|
309 |
return final_results
|
310 |
|
311 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
312 |
# A confidence score is computed using FAISS and BM25 ranking
|
313 |
-
# FAISS: The similarity score between the
|
314 |
-
# BM25: The BM25 scores for the query is normalized
|
315 |
-
#
|
316 |
-
|
317 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
318 |
if not retrieved_chunks:
|
319 |
-
return 0
|
320 |
-
|
321 |
-
|
322 |
" ".join(retrieved_chunks), normalize_embeddings=True
|
323 |
)
|
324 |
-
|
325 |
-
faiss_score = np.dot(
|
|
|
326 |
normalized_faiss = (faiss_score + 1) / 2
|
327 |
-
# BM25
|
328 |
-
|
329 |
-
bm25_scores = bm25.get_scores(
|
|
|
330 |
if bm25_scores.size > 0:
|
331 |
-
|
332 |
-
|
333 |
-
|
334 |
-
|
335 |
-
|
336 |
-
)
|
337 |
-
bm25_score = (
|
338 |
-
np.mean([bm25_scores[idx] for idx in range(len(retrieved_chunks))])
|
339 |
-
if len(retrieved_chunks) > 0
|
340 |
else 0
|
341 |
)
|
342 |
-
normalized_bm25 = (bm25_score - min_bm25) / (max_bm25 - min_bm25)
|
343 |
normalized_bm25 = max(0, min(1, normalized_bm25))
|
344 |
else:
|
345 |
-
normalized_bm25 = 0
|
346 |
-
|
347 |
-
|
348 |
-
|
|
|
|
|
|
|
|
|
|
|
349 |
)
|
350 |
-
return confidence_score
|
351 |
|
352 |
|
353 |
# UI handle for query model button
|
@@ -405,40 +433,62 @@ def query_model(
|
|
405 |
else:
|
406 |
break
|
407 |
prompt = (
|
408 |
-
|
409 |
-
"
|
410 |
-
"
|
411 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
412 |
)
|
413 |
inputs = tokenizer(prompt, return_tensors="pt", padding=True).to(DEVICE)
|
414 |
inputs.pop("token_type_ids", None)
|
415 |
logger.info("Generating output")
|
416 |
input_len = inputs["input_ids"].shape[-1]
|
|
|
417 |
with torch.inference_mode():
|
418 |
output = model.generate(
|
419 |
**inputs,
|
420 |
max_new_tokens=max_new_tokens,
|
421 |
num_return_sequences=1,
|
422 |
repetition_penalty=repetition_penalty,
|
|
|
|
|
423 |
pad_token_id=tokenizer.eos_token_id,
|
424 |
)
|
425 |
-
|
426 |
-
if use_extraction:
|
427 |
-
start_len = input_len
|
428 |
-
output = output[0][start_len:]
|
429 |
execution_time = time.perf_counter() - start_time
|
430 |
logger.info(f"Query processed in {execution_time:.2f} seconds.")
|
431 |
-
|
432 |
-
|
433 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
434 |
)
|
435 |
-
logger.info(f"Confidence: {confidence_score
|
436 |
if confidence_score <= 0.3:
|
437 |
logger.error(f"The system is unsure about this response.")
|
438 |
response += "\nThe system is unsure about this response."
|
|
|
|
|
|
|
|
|
439 |
return (
|
440 |
response,
|
441 |
-
f"Confidence: {confidence_score
|
442 |
)
|
443 |
|
444 |
|
@@ -464,7 +514,7 @@ with gr.Blocks(title="Financial Statement RAG with LLM") as ui:
|
|
464 |
top_k_input = gr.Number(value=15, label="Top K (Default: 15)")
|
465 |
lambda_faiss_input = gr.Slider(0, 1, value=0.5, label="Lambda FAISS (0-1)")
|
466 |
repetition_penalty = gr.Slider(
|
467 |
-
1, 2, value=1.
|
468 |
)
|
469 |
max_tokens_input = gr.Number(value=100, label="Max New Tokens")
|
470 |
use_extraction = gr.Checkbox(label="Retrieve only the answer", value=False)
|
@@ -485,6 +535,7 @@ with gr.Blocks(title="Financial Statement RAG with LLM") as ui:
|
|
485 |
],
|
486 |
outputs=[query_output, time_output],
|
487 |
)
|
|
|
488 |
# Application entry point
|
489 |
if __name__ == "__main__":
|
490 |
logger.info("Starting Gradio server...")
|
|
|
37 |
# SLM: Microsoft PHI-2 model is loaded
|
38 |
# It does have higher memory and compute requirements compared to TinyLlama and Falcon
|
39 |
# But it gives the best results among the three
|
40 |
+
# DEVICE = "cpu" # or cuda
|
41 |
+
DEVICE = "cuda" # or cuda
|
42 |
# MODEL_NAME = "TinyLlama/TinyLlama_v1.1"
|
43 |
# MODEL_NAME = "tiiuae/falcon-rw-1b"
|
44 |
MODEL_NAME = "microsoft/phi-2"
|
|
|
55 |
# Since the model is to be hosted on a cpu instance, we use float32
|
56 |
# For GPU, we can use float16 or bfloat16
|
57 |
model = AutoModelForCausalLM.from_pretrained(
|
58 |
+
MODEL_NAME, torch_dtype=torch.bfloat16, trust_remote_code=True
|
59 |
).to(DEVICE)
|
60 |
model.eval()
|
61 |
# model = torch.quantization.quantize_dynamic(model, {torch.nn.Linear}, dtype=torch.qint8)
|
|
|
234 |
pickle.dump(bm25_data, f)
|
235 |
return "Files processed successfully! You can now query."
|
236 |
|
237 |
+
def contains_financial_entities(query):
|
238 |
+
"""Check if the query has financial entities"""
|
239 |
+
doc = nlp(query)
|
240 |
+
for ent in doc.ents:
|
241 |
+
if ent.label_ in FINANCIAL_ENTITY_LABELS:
|
242 |
+
return True
|
243 |
+
return False
|
244 |
|
245 |
# Input guardrail implementation
|
246 |
# Regex is used to filter queries related to sensitive topics
|
|
|
250 |
def is_query_allowed(query):
|
251 |
"""Checks if the query violates security or confidentiality rules"""
|
252 |
for pattern in restricted_patterns:
|
253 |
+
if re.search(pattern, query.lower(), re.IGNORECASE):
|
254 |
return False, "This query requests sensitive or confidential information."
|
255 |
doc = nlp(query)
|
256 |
for ent in doc.ents:
|
|
|
317 |
return final_results
|
318 |
|
319 |
|
320 |
+
def compute_entropy(logits):
|
321 |
+
"""Compute entropy from logits."""
|
322 |
+
probs = torch.softmax(logits, dim=-1)
|
323 |
+
log_probs = torch.log(probs + 1e-9)
|
324 |
+
entropy = -(probs * log_probs).sum(dim=-1)
|
325 |
+
return entropy.mean().item()
|
326 |
+
|
327 |
+
|
328 |
# A confidence score is computed using FAISS and BM25 ranking
|
329 |
+
# FAISS: The similarity score between the response and the retrieved chunks are normalized
|
330 |
+
# BM25: The BM25 scores for the query and response combined tokens is normalized
|
331 |
+
# The mean of top token probability mean and 1-entropy score is the model_conf_signal
|
332 |
+
# FAISS, BM25 and the model_conf_signal are combined using a weighted sum
|
333 |
+
def compute_response_confidence(
|
334 |
+
query,
|
335 |
+
response,
|
336 |
+
retrieved_chunks,
|
337 |
+
bm25,
|
338 |
+
model_conf_signal=0.5,
|
339 |
+
lambda_faiss=0.4,
|
340 |
+
lambda_conf=0.4,
|
341 |
+
lambda_bm25=1.8,
|
342 |
+
):
|
343 |
+
"""Calculates a confidence score using FAISS, BM25, top token probabilites and entropy score"""
|
344 |
if not retrieved_chunks:
|
345 |
+
return 0.0
|
346 |
+
# Compute FAISS similarity
|
347 |
+
retrieved_embedding = embed_model.encode(
|
348 |
" ".join(retrieved_chunks), normalize_embeddings=True
|
349 |
)
|
350 |
+
response_embedding = embed_model.encode(response, normalize_embeddings=True)
|
351 |
+
faiss_score = np.dot(retrieved_embedding, response_embedding)
|
352 |
+
# Normalize the FAISS score
|
353 |
normalized_faiss = (faiss_score + 1) / 2
|
354 |
+
# Compute BM25 for combined query + response
|
355 |
+
tokenized_combined = (query + " " + response).lower().split()
|
356 |
+
bm25_scores = bm25.get_scores(tokenized_combined)
|
357 |
+
# Normalize the BM25 score
|
358 |
if bm25_scores.size > 0:
|
359 |
+
bm25_score = np.mean(bm25_scores)
|
360 |
+
min_bm25, max_bm25 = np.min(bm25_scores), np.max(bm25_scores)
|
361 |
+
normalized_bm25 = (
|
362 |
+
(bm25_score - min_bm25) / (max_bm25 - min_bm25 + 1e-6)
|
363 |
+
if min_bm25 != max_bm25
|
|
|
|
|
|
|
|
|
364 |
else 0
|
365 |
)
|
|
|
366 |
normalized_bm25 = max(0, min(1, normalized_bm25))
|
367 |
else:
|
368 |
+
normalized_bm25 = 0.0
|
369 |
+
logger.info(
|
370 |
+
f"Faiss score: {normalized_faiss}, bm25: {normalized_bm25}, "
|
371 |
+
f"Mean Top Token + Entropy Avg: {model_conf_signal}"
|
372 |
+
)
|
373 |
+
confidence_score = (
|
374 |
+
lambda_faiss * normalized_faiss
|
375 |
+
+ model_conf_signal * lambda_conf
|
376 |
+
+ lambda_bm25 * normalized_bm25
|
377 |
)
|
378 |
+
return round(min(100, max(0, confidence_score.item() * 100)), 2)
|
379 |
|
380 |
|
381 |
# UI handle for query model button
|
|
|
433 |
else:
|
434 |
break
|
435 |
prompt = (
|
436 |
+
"You are a financial analyst. Answer financial queries concisely using only the numerical data "
|
437 |
+
"explicitly present in the provided financial context:\n\n"
|
438 |
+
f"{context}\n\n"
|
439 |
+
"Strictly use only the given financial data. Do not assume, infer, or generate missing data."
|
440 |
+
" Retain the original format of financial figures exactly as given."
|
441 |
+
" Do not attempt to convert the currency into any other format."
|
442 |
+
" If the requested information is not available in the provided context, respond with "
|
443 |
+
"'No relevant financial data available.'"
|
444 |
+
" Provide exactly one answer in a single sentence."
|
445 |
+
" Do not generate explanations, additional text, or answer multiple queries."
|
446 |
+
f"\nQuery: {query}"
|
447 |
)
|
448 |
inputs = tokenizer(prompt, return_tensors="pt", padding=True).to(DEVICE)
|
449 |
inputs.pop("token_type_ids", None)
|
450 |
logger.info("Generating output")
|
451 |
input_len = inputs["input_ids"].shape[-1]
|
452 |
+
logger.info(f"Input len: {input_len}")
|
453 |
with torch.inference_mode():
|
454 |
output = model.generate(
|
455 |
**inputs,
|
456 |
max_new_tokens=max_new_tokens,
|
457 |
num_return_sequences=1,
|
458 |
repetition_penalty=repetition_penalty,
|
459 |
+
output_scores=True,
|
460 |
+
return_dict_in_generate=True,
|
461 |
pad_token_id=tokenizer.eos_token_id,
|
462 |
)
|
463 |
+
sequences = output["sequences"][0][input_len:]
|
|
|
|
|
|
|
464 |
execution_time = time.perf_counter() - start_time
|
465 |
logger.info(f"Query processed in {execution_time:.2f} seconds.")
|
466 |
+
log_probs = output["scores"] # List of logits per generated token
|
467 |
+
token_probs = [torch.softmax(lp, dim=-1) for lp in log_probs]
|
468 |
+
# Extract top token probabilities for each step
|
469 |
+
token_confidences = [tp.max().item() for tp in token_probs]
|
470 |
+
# Compute final confidence score
|
471 |
+
top_token_conf = sum(token_confidences) / len(token_confidences)
|
472 |
+
print(f"Token Token Probability Mean: {top_token_conf:.4f}")
|
473 |
+
entropy_score = sum(compute_entropy(lp) for lp in log_probs) / len(log_probs)
|
474 |
+
entropy_conf = 1 - (entropy_score / torch.log(torch.tensor(tokenizer.vocab_size)))
|
475 |
+
print(f"Entropy-based Confidence: {entropy_conf:.4f}")
|
476 |
+
model_conf_signal = (top_token_conf + (1 - entropy_conf)) / 2
|
477 |
+
response = tokenizer.decode(sequences, skip_special_tokens=True)
|
478 |
+
confidence_score = compute_response_confidence(
|
479 |
+
query, response, retrieved_chunks, bm25, model_conf_signal
|
480 |
)
|
481 |
+
logger.info(f"Confidence: {confidence_score}%")
|
482 |
if confidence_score <= 0.3:
|
483 |
logger.error(f"The system is unsure about this response.")
|
484 |
response += "\nThe system is unsure about this response."
|
485 |
+
final_out = ""
|
486 |
+
if not use_extraction:
|
487 |
+
final_out += f"Context: {context}\nQuery: {query}\n"
|
488 |
+
final_out += f"Response: {response}"
|
489 |
return (
|
490 |
response,
|
491 |
+
f"Confidence: {confidence_score}%\nTime taken: {execution_time:.2f} seconds",
|
492 |
)
|
493 |
|
494 |
|
|
|
514 |
top_k_input = gr.Number(value=15, label="Top K (Default: 15)")
|
515 |
lambda_faiss_input = gr.Slider(0, 1, value=0.5, label="Lambda FAISS (0-1)")
|
516 |
repetition_penalty = gr.Slider(
|
517 |
+
1, 2, value=1.2, label="Repetition Penality (1-2)"
|
518 |
)
|
519 |
max_tokens_input = gr.Number(value=100, label="Max New Tokens")
|
520 |
use_extraction = gr.Checkbox(label="Retrieve only the answer", value=False)
|
|
|
535 |
],
|
536 |
outputs=[query_output, time_output],
|
537 |
)
|
538 |
+
|
539 |
# Application entry point
|
540 |
if __name__ == "__main__":
|
541 |
logger.info("Starting Gradio server...")
|
data_filters.py
CHANGED
@@ -1,16 +1,16 @@
|
|
1 |
"""Sensitive data filters"""
|
2 |
|
3 |
restricted_patterns = [
|
4 |
-
r"\b(?:
|
5 |
-
r"\b(?:salary|compensation|bonus|pay|income)\b.*\b(?:
|
6 |
r"\b(?:acquisition|merger|buyout)\b.*\b(?:before|pre-announcement|leak|inside information)\b",
|
7 |
r"\b(?:before|pre-announcement|leak|inside information)\b.*\b(?:acquisition|merger|buyout)\b",
|
8 |
r"\b(?:stock price|share price|insider trading|buying shares)\b",
|
9 |
r"\b(?:internal policy|data breach|security protocol|confidential|classified)\b",
|
10 |
r"\b(?:password|access credentials|encryption key|secure key)\b",
|
11 |
-
r"\b(?:social security number|
|
12 |
-
r"\b(?:employee records|payroll|medical records|
|
13 |
-
r"\b(?:
|
14 |
]
|
15 |
|
16 |
restricted_topics = {
|
|
|
1 |
"""Sensitive data filters"""
|
2 |
|
3 |
restricted_patterns = [
|
4 |
+
r"\b(?:cfo|ceo|cto|executive|director|manager|employee|staff|worker)\b.*\b(?:salary|compensation|bonus|pay|income)\b",
|
5 |
+
r"\b(?:salary|compensation|bonus|pay|income)\b.*\b(?:cfo|ceo|cto|executive|director|manager|employee|staff|worker)\b",
|
6 |
r"\b(?:acquisition|merger|buyout)\b.*\b(?:before|pre-announcement|leak|inside information)\b",
|
7 |
r"\b(?:before|pre-announcement|leak|inside information)\b.*\b(?:acquisition|merger|buyout)\b",
|
8 |
r"\b(?:stock price|share price|insider trading|buying shares)\b",
|
9 |
r"\b(?:internal policy|data breach|security protocol|confidential|classified)\b",
|
10 |
r"\b(?:password|access credentials|encryption key|secure key)\b",
|
11 |
+
r"\b(?:social security number|ssn|passport number|credit card|bank account|tax id|tin|personal details)\b",
|
12 |
+
r"\b(?:employee records|payroll|medical records|hr data|salary data|pii|personally identifiable information)\b",
|
13 |
+
r"\b(?:cfo|ceo|cto|executive|director|manager|employee|staff|worker)\b.*\b(?:address|work location|home location|residence|personal contact|phone number|email|office location)\b",
|
14 |
]
|
15 |
|
16 |
restricted_topics = {
|