Update codebase and configure automatic environment switcher (excluding MLX adapters)

.gitignore

@@ -1,2 +1,6 @@
 data/mintoak/chroma_db/
-data/mintoak/chroma_db/
+adapters/
+mlx-env/
+.DS_Store
+__pycache__/
+*.pyc

app.py

@@ -14,7 +14,15 @@ from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStream
 BASE_DIR = os.path.dirname(os.path.abspath(__file__))
 CHROMA_DB_PATH = os.path.join(BASE_DIR, "data/mintoak/chroma_db")
 COLLECTION_NAME = "mintoak_content"
-MODEL_PATH = "Qwen/Qwen2.5-1.5B-Instruct" # Using the 1.5B parameters model
+# Automatically choose model size based on environment
+# Hugging Face Spaces sets "HF_SPACE" or "SPACE_ID" automatically
+IS_HF_SPACE = "HF_SPACE" in os.environ or "SPACE_ID" in os.environ
+if IS_HF_SPACE:
+    MODEL_PATH = "Qwen/Qwen2.5-7B-Instruct"  # 7B model on Hugging Face
+    print("Running on Hugging Face Spaces. Using Qwen 2.5 7B model.")
+else:
+    MODEL_PATH = "Qwen/Qwen2.5-1.5B-Instruct"  # 1.5B model locally
+    print("Running locally. Using Qwen 2.5 1.5B model.")
 PORT = 7860 # Hugging Face Spaces requires port 7860
 
 # Init Flask
@@ -48,12 +56,12 @@ if _collection.count() == 0 and os.path.exists(CHUNKS_JSON_PATH):
         )
     print(f"Database populated with {_collection.count()} chunks!")
 
-# Load Model on GPU (CUDA)
+# Load Model on GPU (CUDA) if available
 device = "cuda" if torch.cuda.is_available() else "cpu"
-print(f"Loading Qwen 2.5 1.5B on device: {device}...")
+print(f"Loading base model {MODEL_PATH} on device: {device}...")
 _tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
 
-# Using float16 for fast GPU inference
+# Using float16 for fast GPU inference if CUDA is available
 if device == "cuda":
     _model = AutoModelForCausalLM.from_pretrained(
         MODEL_PATH, 
@@ -375,8 +383,18 @@ def chat():
             thread = Thread(target=_model.generate, kwargs=generation_kwargs)
             thread.start()
             
+            citation_started = False
             for token_text in streamer:
                 full_response += token_text
+                
+                if citation_started:
+                    continue
+                
+                last_snippet = full_response[-40:].lower()
+                if "👉" in token_text or "for more details" in last_snippet or "details, visit" in last_snippet:
+                    citation_started = True
+                    continue
+                    
                 yield f"data: {json.dumps({'event': 'token', 'token': token_text})}\n\n"
 
         latency = round(time.time() - t0, 2)

data/mintoak/leads.jsonl

@@ -0,0 +1 @@
+{"timestamp": "2026-05-30 11:16:59", "name": "Rutvij", "email": "rutvijdoshi07@gmail.com", "query": "I want to know the demo about mintoak products for my business", "conversation_id": "conv_1780119934617_475w"}

scripts/mintoak/chat_server.py

@@ -26,7 +26,7 @@ BASE_DIR       = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath
 CHROMA_DB_PATH = os.path.join(BASE_DIR, "data/mintoak/chroma_db")
 COLLECTION_NAME = "mintoak_content"
 CHUNKS_JSON    = os.path.join(BASE_DIR, "data/mintoak/mintoak_chunks.json")
-MODEL_PATH     = "mlx-community/Qwen2.5-3B-Instruct-4bit"
+MODEL_PATH     = "mlx-community/Qwen2.5-7B-Instruct-4bit"
 ADAPTER_PATH   = os.path.join(BASE_DIR, "adapters/mintoak")
 PORT           = 5001
 
@@ -119,7 +119,7 @@ QUERY_ENHANCEMENT_RULES = [
     },
     {
         "keywords": ["document", "documents", "checklist", "paperwork", "certificates"],
-        "expansion": "What documents are needed for merchant onboarding? DigiOnboard KYC KYB",
+        "expansion": "Mintoak DigiOnboard merchant onboarding documents checklist GST PAN registration certificates cancelled cheques board resolutions",
         "prompt_note": "Ensure you list the specific documents needed: registration certificates, GST certificates, cancelled cheques, and board resolutions.",
         "top_k": 3
     },

scripts/mintoak/evaluate_rag.py

@@ -10,7 +10,7 @@ from mlx_lm.sample_utils import make_sampler
 # Configurations matching rag_assistant.py
 CHROMA_DB_PATH = "data/mintoak/chroma_db"
 COLLECTION_NAME = "mintoak_content"
-MODEL_PATH = "mlx-community/Qwen2.5-1.5B-Instruct-4bit"
+MODEL_PATH = "mlx-community/Qwen2.5-7B-Instruct-4bit"
 ADAPTER_PATH = "adapters/mintoak"
 CHUNKS_JSON_PATH = "data/mintoak/mintoak_chunks.json"
 
@@ -231,7 +231,7 @@ def main():
     import argparse
     import sys
     
-    parser = argparse.ArgumentParser(description="RR RAG Evaluator")
+    parser = argparse.ArgumentParser(description="Mintoak RAG Evaluator")
     parser.add_argument("--batch_size", type=int, default=50, help="Batch size for progress updates & saving")
     parser.add_argument("--max_cases", type=int, default=None, help="Maximum number of test cases to run")
     parser.add_argument("--progress_file", type=str, default="data/mintoak/eval_progress.json", help="Path to save intermediate evaluation progress")

scripts/mintoak/fine_tune.sh

@@ -28,7 +28,7 @@ echo "Creating folder adapters/mintoak..."
 mkdir -p adapters/mintoak
 
 echo "Starting MLX LoRA Fine-Tuning..."
-echo "Using base model: mlx-community/Qwen2.5-1.5B-Instruct-4bit"
+echo "Using base model: mlx-community/Qwen2.5-7B-Instruct-4bit"
 echo "Data directory: ./data/mintoak"
 echo "Adapters output path: ./adapters/mintoak"
 
@@ -44,7 +44,7 @@ echo "Adapters output path: ./adapters/mintoak"
 #   --num-layers: Number of layers we are fine-tuning (out of the base model's layers)
 #   --adapter-path: The output folder where the adapters (.safetensors) and config will be saved
 python -m mlx_lm lora \
-  --model mlx-community/Qwen2.5-1.5B-Instruct-4bit \
+  --model mlx-community/Qwen2.5-7B-Instruct-4bit \
   --train \
   --data ./data/mintoak \
   --iters 50 \

scripts/mintoak/generate_html_report.py

@@ -136,8 +136,8 @@ def main():
 <head>
     <meta charset="UTF-8">
     <meta name="viewport" content="width=device-width, initial-scale=1.0">
-    <title>RR RAG Evaluation Dashboard</title>
-    <meta name="description" content="Interactive evaluation report for RR RAG Assistant — {total_cases} test cases, {pass_rate:.1f}% pass rate.">
+    <title>Mintoak RAG Evaluation Dashboard</title>
+    <meta name="description" content="Interactive evaluation report for Mintoak RAG Assistant — {total_cases} test cases, {pass_rate:.1f}% pass rate.">
     <link href="https://fonts.googleapis.com/css2?family=Plus+Jakarta+Sans:wght@300;400;500;600;700;800&display=swap" rel="stylesheet">
     <script src="https://cdn.jsdelivr.net/npm/chart.js@4.4.2/dist/chart.umd.min.js"></script>
     <style>
@@ -416,7 +416,7 @@ def main():
     <div class="topbar-brand">
         <div class="topbar-logo">M</div>
         <div>
-            <div class="topbar-title">RR RAG · Evaluation Dashboard</div>
+            <div class="topbar-title">Mintoak RAG · Evaluation Dashboard</div>
             <div class="topbar-sub">{total_cases} test cases · generated {__import__('datetime').datetime.now().strftime('%d %b %Y, %H:%M')}</div>
         </div>
     </div>

scripts/mintoak/rag_assistant.py

@@ -22,7 +22,7 @@ COLLECTION_NAME = "mintoak_content"
 CHUNKS_JSON_PATH = "data/mintoak/mintoak_chunks.json"
 
 # The base language model we are loading from Hugging Face / local cache
-MODEL_PATH = "mlx-community/Qwen2.5-1.5B-Instruct-4bit"
+MODEL_PATH = "mlx-community/Qwen2.5-7B-Instruct-4bit"
 
 # Path to the directory containing our trained fine-tuning LoRA adapters
 ADAPTER_PATH = "adapters/mintoak"

scripts/mintoak/templates/index.html

@@ -1793,7 +1793,7 @@
               botBubbleEl.innerHTML = marked.parse(displayText);
               scrollBottom();
             } else if (data.event === 'done') {
-              let finalContent = accumulatedText;
+              let finalContent = data.text || accumulatedText;
               let leadCapture = false;
               if (finalContent.includes('[CAPTURE_LEAD]')) {
                 finalContent = finalContent.replace('[CAPTURE_LEAD]', '').trim();