File size: 15,984 Bytes
becc8f7
 
 
 
 
f7d42c1
becc8f7
 
f7d42c1
 
 
 
becc8f7
f7d42c1
 
 
 
 
 
 
 
 
 
 
be8f70c
f7d42c1
 
 
 
 
 
 
 
 
be8f70c
f7d42c1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
be8f70c
f7d42c1
 
be8f70c
f7d42c1
 
 
 
 
 
 
 
 
be8f70c
f7d42c1
 
 
 
 
 
 
 
 
be8f70c
f7d42c1
 
 
 
 
 
 
be8f70c
f7d42c1
be8f70c
f7d42c1
be8f70c
f7d42c1
becc8f7
 
f7d42c1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
be8f70c
f7d42c1
be8f70c
f7d42c1
be8f70c
f7d42c1
 
 
be8f70c
 
f7d42c1
 
 
 
 
 
becc8f7
 
f7d42c1
 
 
 
 
 
 
 
 
 
 
 
 
 
be8f70c
f7d42c1
be8f70c
f7d42c1
 
becc8f7
 
f7d42c1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
457467f
f7d42c1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
becc8f7
f7d42c1
becc8f7
f7d42c1
becc8f7
f7d42c1
becc8f7
f7d42c1
 
 
2bfc3e4
 
 
be8f70c
2bfc3e4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
be8f70c
 
 
 
 
 
2bfc3e4
 
be8f70c
 
 
 
 
 
becc8f7
f7d42c1
 
2bfc3e4
f7d42c1
 
 
 
 
 
 
2bfc3e4
f7d42c1
 
2bfc3e4
f7d42c1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
be8f70c
f7d42c1
 
 
 
 
 
 
be8f70c
f7d42c1
 
be8f70c
f7d42c1
 
be8f70c
f7d42c1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
becc8f7
 
 
 
 
 
f7d42c1
 
 
 
 
 
 
 
 
 
 
 
 
 
becc8f7
 
 
 
 
 
be8f70c
f7d42c1
becc8f7
be8f70c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
import os
from dotenv import load_dotenv
from operator import itemgetter
from langchain_groq import ChatGroq
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.runnables import RunnableParallel, RunnablePassthrough
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables.history import RunnableWithMessageHistory
from langchain_core.documents import Document
from query_expansion import expand_query_simple
from typing import List, Optional
import time

class GroqAPIKeyManager:
    def __init__(self, api_keys: List[str]):
        self.api_keys = [key for key in api_keys if key and key != "your_groq_api_key_here"]
        if not self.api_keys:
            raise ValueError("No valid API keys provided!")
        
        self.current_index = 0
        self.failed_keys = set()
        self.success_count = {key: 0 for key in self.api_keys}
        self.failure_count = {key: 0 for key in self.api_keys}
        
        print(f"API Key Manager: Loaded {len(self.api_keys)} API keys")
    
    def get_current_key(self) -> str:
        return self.api_keys[self.current_index]
    
    def mark_success(self, api_key: str):
        if api_key in self.success_count:
            self.success_count[api_key] += 1
            if api_key in self.failed_keys:
                self.failed_keys.remove(api_key)
                print(f"API Key #{self.api_keys.index(api_key) + 1} recovered!")
    
    def mark_failure(self, api_key: str):
        if api_key in self.failure_count:
            self.failure_count[api_key] += 1
            self.failed_keys.add(api_key)
    
    def rotate_to_next_key(self) -> bool:
        initial_index = self.current_index
        attempts = 0
        
        while attempts < len(self.api_keys):
            self.current_index = (self.current_index + 1) % len(self.api_keys)
            attempts += 1
            
            current_key = self.api_keys[self.current_index]
            if attempts >= len(self.api_keys):
                print(f"All keys attempted, retrying with key #{self.current_index + 1}")
                return True
            if current_key not in self.failed_keys:
                print(f"Switching to API Key #{self.current_index + 1}")
                return True
        
        return False
    
    def get_statistics(self) -> str:
        stats = []
        for i, key in enumerate(self.api_keys):
            success = self.success_count[key]
            failure = self.failure_count[key]
            status = "FAILED" if key in self.failed_keys else "ACTIVE"
            masked_key = key[:8] + "..." + key[-4:] if len(key) > 12 else "***"
            stats.append(f"   Key #{i+1} ({masked_key}): {success} success, {failure} failures [{status}]")
        return "\n".join(stats)


def load_api_keys_from_hf_secrets() -> List[str]:
    api_keys = []
    secret_names = ['GROQ_API_KEY_1', 'GROQ_API_KEY_2', 'GROQ_API_KEY_3', 'GROQ_API_KEY_4']
    
    print("Loading API keys from Hugging Face Secrets...")
    
    for secret_name in secret_names:
        try:
            api_key = os.getenv(secret_name)
            
            if api_key and api_key.strip() and api_key != "your_groq_api_key_here":
                api_keys.append(api_key.strip())
                print(f" Loaded: {secret_name}")
            else:
                print(f" Not found or empty: {secret_name}")
        except Exception as e:
            print(f" Error loading {secret_name}: {str(e)}")
    return api_keys


def create_llm_with_fallback(
    api_key_manager: GroqAPIKeyManager,
    model_name: str,
    temperature: float,
    max_retries: int = 3
) -> ChatGroq:
    for attempt in range(max_retries):
        current_key = api_key_manager.get_current_key()
        
        try:
            llm = ChatGroq(
                model_name=model_name,
                api_key=current_key,
                temperature=temperature
            )
            test_result = llm.invoke("test")
            api_key_manager.mark_success(current_key)
            return llm
            
        except Exception as e:
            error_msg = str(e).lower()
            api_key_manager.mark_failure(current_key)
            if "rate" in error_msg or "limit" in error_msg:
                print(f"  Rate limit hit on API Key #{api_key_manager.current_index + 1}")
            elif "auth" in error_msg or "api" in error_msg:
                print(f"  Authentication failed on API Key #{api_key_manager.current_index + 1}")
            else:
                print(f"  Error with API Key #{api_key_manager.current_index + 1}: {str(e)[:50]}")
            
            if attempt < max_retries - 1:
                if api_key_manager.rotate_to_next_key():
                    print(f" Retrying with next API key (Attempt {attempt + 2}/{max_retries})...")
                    time.sleep(1)
                else:
                    raise ValueError("All API keys failed!")
            else:
                raise ValueError(f"Failed to initialize LLM after {max_retries} attempts")
    
    raise ValueError("Failed to create LLM with any available API key")


def create_multi_query_retriever(base_retriever, llm, strategy: str = "balanced"):
    def multi_query_retrieve(query: str) -> List[Document]:
        query_variations = expand_query_simple(query, strategy=strategy, llm=llm)
        all_docs = []
        seen_content = set()
        for i, query_var in enumerate(query_variations):
            try:
                docs = base_retriever.invoke(query_var)
                for doc in docs:
                    content_hash = hash(doc.page_content)
                    if content_hash not in seen_content:
                        seen_content.add(content_hash)
                        all_docs.append(doc)
            except Exception as e:
                print(f" Query Expansion Error (Query {i+1}): {str(e)[:50]}")
                continue
        print(f" Query Expansion: Retrieved {len(all_docs)} unique documents.")
        return all_docs
    return multi_query_retrieve


def get_system_prompt(temperature: float) -> str:
    if temperature <= 0.4:
        return """You are CogniChat, an expert document analysis assistant specializing in comprehensive and well-structured answers.

RESPONSE GUIDELINES:

**Structure & Formatting:**
- Start with a direct answer to the question
- Use **bold** for key terms, important concepts, and technical terminology
- Use bullet points (•) for lists, features, or multiple items
- Use numbered lists (1., 2., 3.) for steps, procedures, or sequential information
- Use ### Headers to organize different sections or topics
- Add blank lines between sections for readability

**Source Citation:**
- Always cite information using: [Source: filename, Page: X] and cite it at the end of the entire answer only
- Place citations at the end of your final answer only 
- Do not cite sources within the body of your answer
- Multiple sources: [Source: doc1.pdf, Page: 3; doc2.pdf, Page: 7]

**Completeness:**
- Provide thorough, detailed answers using ALL relevant information from context
- Summarize and properly elaborate each point for increased clarity
- If the question has multiple parts, address each part clearly

**Accuracy:**
- ONLY use information from the provided context documents below
- If information is incomplete, state what IS available and what ISN'T
- If the answer isn't in the context, clearly state: "I cannot find this information in the uploaded documents"
- Never make assumptions or add information not in the context

---

{context}

---

Now answer the following question comprehensively using the context above:"""
    
    elif temperature <= 0.8:
        return """You are CogniChat, an intelligent document analysis assistant that combines accuracy with engaging communication.

RESPONSE GUIDELINES:

**Communication Style:**
- Present information in a clear, engaging manner
- Use **bold** for emphasis on important concepts
- Balance structure with natural flow
- Make complex topics accessible and interesting

**Content Approach:**
- Ground your response firmly in the provided context
- Add helpful explanations and connections between concepts
- Use analogies or examples when they help clarify ideas (but keep them brief)
- Organize information logically with headers (###) and lists where appropriate

**Source Attribution:**
- Cite sources at the end: [Source: filename, Page: X]
- Be transparent about what the documents do and don't contain

**Accuracy:**
- Base your answer on the context documents provided
- If information is partial, explain what's available
- Acknowledge gaps: "The documents don't cover this aspect"

---

{context}

---

Now answer the following question in an engaging yet accurate way:"""
    
    else:  # temperature > 0.8
        # Creative BUT CLEAR prompt - REVISED VERSION
        return """You are CogniChat, a creative document analyst who makes complex information clear, memorable, and engaging.

 YOUR CORE MISSION: **CLARITY FIRST, CREATIVITY SECOND**

Make information easier to understand, not harder. Your creativity should illuminate, not obscure.

**CREATIVE CLARITY PRINCIPLES:**

1. **Simplify, Don't Complicate**
   - Break down complex concepts into simple, digestible parts
   - Use everyday language alongside technical terms
   - Explain jargon immediately in plain English
   - Short sentences for complex ideas, varied length for rhythm

2. **Smart Use of Examples & Analogies** (Use Sparingly!)
   - Only use analogies when they genuinely make something clearer
   - Keep analogies simple and relatable (everyday objects/experiences)
   - Never use metaphors that require explanation themselves
   - If you can explain it directly in simple terms, do that instead

3. **Engaging Structure**
   - Start with the core answer in one clear sentence
   - Use **bold** to highlight key takeaways
   - Break information into logical chunks with ### headers
   - Use bullet points for clarity, not decoration
   - Add brief transition phrases to connect ideas smoothly

4. **Conversational Yet Precise**
   - Write like you're explaining to a smart friend
   - Use "you" and active voice to engage readers
   - Ask occasional rhetorical questions only if they aid understanding
   - Vary sentence length to maintain interest
   - Use emojis sparingly (1-2 max) and only where they add clarity

5. **Visual Clarity**
   - Strategic use of formatting: **bold** for key terms, *italics* for emphasis
   - White space between sections for easy scanning
   - Progressive disclosure: simple concepts first, details after
   - Numbered lists for sequences, bullets for related items

**WHAT TO AVOID:**
-  Flowery or overly descriptive language
-  Complex metaphors that need their own explanation
-  Long narrative storytelling that buries the facts
-  Multiple rhetorical questions in a row
-  Overuse of emojis or exclamation points
-  Making simple things sound complicated

**ACCURACY BOUNDARIES:**
-  Creative explanation and presentation of facts
-  Simple, helpful examples from common knowledge
-  Reorganizing information for better understanding
-  Never invent facts not in the documents
-  Don't contradict source material
-  If info is missing, say so clearly and briefly

**Source Attribution:**
- End with: [Source: filename, Page: X]
- Keep it simple and clear

---

{context}

---

Now, explain the answer clearly and engagingly. Remember: if your grandmother couldn't understand it, simplify more:"""
    


def create_rag_chain(
    retriever,
    get_session_history_func,
    enable_query_expansion=True,
    expansion_strategy="balanced",
    model_name: str = "moonshotai/kimi-k2-instruct",
    temperature: float = 0.2,
    api_keys: Optional[List[str]] = None
):
    if api_keys is None:
        api_keys = load_api_keys_from_hf_secrets()
    
    if not api_keys:
        raise ValueError(
            "No valid API keys found! Please set GROQ_API_KEY or GROQ_API_KEY_1, "
            "GROQ_API_KEY_2, GROQ_API_KEY_3, GROQ_API_KEY_4 in your .env file"
        )
    
    api_key_manager = GroqAPIKeyManager(api_keys)
    
    print(f" RAG: Initializing LLM - Model: {model_name}, Temp: {temperature}")
    
    if temperature <= 0.4:
        creativity_mode = "FACTUAL & STRUCTURED"
    elif temperature <= 0.8:
        creativity_mode = "BALANCED & ENGAGING"
    else:
        creativity_mode = "CREATIVE & STORYTELLING"
    print(f"Creativity Mode: {creativity_mode}")
    
    llm = create_llm_with_fallback(api_key_manager, model_name, temperature)
    print(f"LLM initialized with API Key #{api_key_manager.current_index + 1}")

    if enable_query_expansion:
        print(f"RAG: Query Expansion ENABLED (Strategy: {expansion_strategy})")
        enhanced_retriever = create_multi_query_retriever(
            base_retriever=retriever,
            llm=llm,
            strategy=expansion_strategy
        )
    else:
        enhanced_retriever = retriever

    rewrite_template = """You are an expert at optimizing search queries for document retrieval.

Given the conversation history and a follow-up question, create a comprehensive standalone question that:
1. Incorporates all relevant context from the chat history
2. Expands abbreviations and resolves all pronouns (it, they, this, that, etc.)
3. Includes key technical terms and concepts that would help find relevant documents
4. Maintains the original intent, specificity, and detail level
5. If the question asks for comparison or multiple items, ensure all items are in the query

Chat History:
{chat_history}

Follow-up Question: {question}

Optimized Standalone Question:"""
    rewrite_prompt = ChatPromptTemplate.from_messages([
        ("system", rewrite_template),
        MessagesPlaceholder(variable_name="chat_history"),
        ("human", "{question}")
    ])
    query_rewriter = rewrite_prompt | llm | StrOutputParser()

    def format_docs(docs):
        if not docs:
            return "No relevant documents found in the knowledge base."
        
        formatted_parts = []
        for i, doc in enumerate(docs, 1):
            source = doc.metadata.get('source', 'Unknown Document')
            page = doc.metadata.get('page', 'N/A')
            rerank_score = doc.metadata.get('rerank_score')
            content = doc.page_content.strip()
            
            doc_header = f"{'='*60}\nDOCUMENT {i}\n{'='*60}"
            metadata_line = f"Source: {source} | Page: {page}"
            if rerank_score:
                metadata_line += f" | Relevance: {rerank_score:.3f}"
            
            formatted_parts.append(
                f"{doc_header}\n"
                f"{metadata_line}\n"
                f"{'-'*60}\n"
                f"{content}\n"
            )
        return f"RETRIEVED CONTEXT ({len(docs)} documents):\n\n" + "\n".join(formatted_parts)

    rag_template = get_system_prompt(temperature)
    
    rag_prompt = ChatPromptTemplate.from_messages([
        ("system", rag_template),
        MessagesPlaceholder(variable_name="chat_history"),
        ("human", "{question}"),
    ])

    rewriter_input = RunnableParallel({
        "question": itemgetter("question"),
        "chat_history": itemgetter("chat_history"),
    })

    retrieval_chain = rewriter_input | query_rewriter | enhanced_retriever | format_docs

    conversational_rag_chain = RunnableParallel({
        "context": retrieval_chain,
        "question": itemgetter("question"),
        "chat_history": itemgetter("chat_history"),
    }) | rag_prompt | llm | StrOutputParser()

    chain_with_memory = RunnableWithMessageHistory(
        conversational_rag_chain,
        get_session_history_func,
        input_messages_key="question",
        history_messages_key="chat_history",
    )
    
    print("RAG: Chain created successfully.")
    print("\n" + api_key_manager.get_statistics())
    
    return chain_with_memory, api_key_manager