Upload 5 files

CHANGELOG.md

@@ -10,7 +10,7 @@ All notable changes to ConversAI will be documented in this file.
   - **No API endpoint issues** - everything runs on your Space
   - **Faster after first load** - models cached in memory
   - **100% private** - all processing happens locally
-  - Default model: **google/flan-t5-base** (250MB, very fast)
+  - Default model: **google/flan-t5-large** (1.2GB, good quality)
   - Supports all Flan-T5 variants (base, large, xl, xxl)
 
 ### Added
@@ -43,10 +43,16 @@ All notable changes to ConversAI will be documented in this file.
   - Added model caching to keep models in memory
   - Auto-detects CUDA/CPU and optimizes accordingly
 
-- **Default model**: `google/flan-t5-base` (line 83)
+- **Default model**: `google/flan-t5-large` (line 84)
   - Changed from API-based to local transformers
-  - Smaller model for faster loading
-  - User can upgrade to larger models via LLM_MODEL env var
+  - 1.2GB model provides good balance of quality and speed
+  - Better at JSON generation than smaller models
+  - User can upgrade to xl/xxl or downgrade to base via LLM_MODEL env var
+
+- **Improved prompts** in `survey_generator.py`:
+  - Simplified prompts for better T5 model compatibility
+  - Added fallback survey generation if JSON parsing fails
+  - More direct instructions with concrete examples
 
 - **New dependencies added** to requirements.txt:
   - transformers>=4.36.0

README.md

@@ -57,9 +57,10 @@ Battle the blank page, reach global audiences, and uncover insights with AI assi
 
 **✨ Zero configuration needed!** ConversAI works out-of-the-box on HuggingFace Spaces using local model loading.
 
-**Default Model:** google/flan-t5-base
+**Default Model:** google/flan-t5-large
 - ✅ **100% Free** - No API keys, no costs, ever
-- ✅ **Fast** - Models load locally, typically 2-5 seconds per request after loading
+- ✅ **Good quality** - 1.2GB model, excellent at following instructions
+- ✅ **Fast after loading** - Typically 3-8 seconds per request after initial load
 - ✅ **No API dependencies** - Runs entirely on your Space's compute
 - ✅ **Private** - All processing happens locally, nothing sent to external APIs
 - ✅ **Reliable** - Google's instruction-tuned model, battle-tested
@@ -77,9 +78,9 @@ You can try different free models by setting the `LLM_MODEL` environment variabl
 
 | Model | Best For | Speed | Quality | Model Size |
 |-------|----------|-------|---------|------------|
-| **google/flan-t5-base** (default) | Balanced - fast & small | ⚡⚡⚡ Very Fast | ⭐⭐ Good | 250MB |
-| **google/flan-t5-large** | Better quality | ⚡⚡ Fast | ⭐⭐⭐ Better | 1.2GB |
-| **google/flan-t5-xl** | Best quality | ⚡ Medium | ⭐⭐⭐⭐ Excellent | 3GB |
+| **google/flan-t5-base** | Testing - fastest | ⚡⚡⚡ Very Fast | ⭐⭐ Basic | 250MB |
+| **google/flan-t5-large** (default) | **Recommended** - balanced | ⚡⚡ Fast | ⭐⭐⭐ Good | 1.2GB |
+| **google/flan-t5-xl** | Better quality | ⚡ Medium | ⭐⭐⭐⭐ Excellent | 3GB |
 | **google/flan-t5-xxl** | Maximum quality | ⚡ Slower | ⭐⭐⭐⭐⭐ Best | 11GB |
 
 **Note:** Flan-T5 models are Google's instruction-tuned models, specifically designed for following instructions. They run locally with transformers library.
@@ -102,12 +103,12 @@ LLM_MODEL=google/flan-t5-xl
 
 ### Tips for Best Performance with Local Models
 
-1. **Start with flan-t5-base** - Fast loading and good results
-2. **First load takes time** - Model downloads and loads (~1-2 minutes for base)
-3. **Subsequent requests are fast** - Model stays in memory (2-5 seconds)
-4. **Upgrade model size for quality** - flan-t5-large or xl for better results
-5. **Keep prompts concise** - Shorter outlines = faster generation
-6. **Monitor memory** - Larger models (XL, XXL) need more RAM
+1. **Default model (flan-t5-large) is recommended** - Good balance of quality and speed
+2. **First load takes time** - Model downloads and loads (~2-3 minutes for large)
+3. **Subsequent requests are fast** - Model stays in memory (3-8 seconds)
+4. **For simple testing** - Use flan-t5-base (faster loading)
+5. **For best quality** - Use flan-t5-xl or xxl (requires more memory)
+6. **Keep prompts clear** - Simpler outlines work better with smaller models
 
 ## 📦 Installation
 

llm_backend.py

@@ -78,9 +78,10 @@ class LLMBackend:
         defaults = {
             LLMProvider.OPENAI: "gpt-4o-mini",
             LLMProvider.ANTHROPIC: "claude-3-5-sonnet-20241022",
-            # Using Flan-T5-Base - small, fast, works locally with transformers
-            # For larger models, try: google/flan-t5-large or google/flan-t5-xl
-            LLMProvider.HUGGINGFACE: "google/flan-t5-base",
+            # Using Flan-T5-Large - good balance of size (1.2GB) and quality
+            # For smaller/faster: google/flan-t5-base (250MB)
+            # For better quality: google/flan-t5-xl (3GB) or google/flan-t5-xxl (11GB)
+            LLMProvider.HUGGINGFACE: "google/flan-t5-large",
             LLMProvider.LM_STUDIO: "google/gemma-3-27b"
         }
         return os.getenv("LLM_MODEL", defaults[self.provider])

survey_generator.py

@@ -58,64 +58,22 @@ class SurveyGenerator:
 
     def _get_system_prompt(self) -> str:
         """System prompt for survey generation"""
-        return """You are an expert survey designer and qualitative researcher with deep knowledge of:
-- Industry best practices for survey design
-- Question formulation techniques (open-ended, closed-ended, Likert scales)
-- Avoiding bias and leading questions
-- Survey flow and respondent experience
-- Research methodologies (interviews, focus groups, ethnographic studies)
-
-Your task is to generate professional, well-structured surveys that will yield high-quality research data.
-Follow these principles:
-1. Use clear, unambiguous language
-2. Avoid double-barreled questions
-3. Include a logical flow from general to specific
-4. Balance open-ended and structured questions appropriately
-5. Consider the respondent's cognitive load
-6. Include screening questions when relevant
-7. Add instructions and context where helpful
-
-Always respond with valid JSON containing the survey structure."""
+        return """You are a professional survey designer. Create surveys in valid JSON format only."""
 
     def _build_generation_prompt(self, outline, survey_type, num_questions, target_audience) -> str:
         """Build the user prompt for survey generation"""
-        return f"""Generate a professional {survey_type} survey based on the following outline:
-
-OUTLINE:
-{outline}
-
-REQUIREMENTS:
-- Target number of questions: {num_questions}
-- Target audience: {target_audience}
-- Survey type: {survey_type}
-
-Please generate a complete survey with:
-1. A clear title
-2. An introduction/welcome message
-3. Well-crafted questions following best practices
-4. Appropriate question types for the research goals
-5. A thank you/closing message
-
-Respond with a JSON object in this exact format:
-{{
-  "title": "Survey Title",
-  "introduction": "Welcome message and instructions",
-  "questions": [
-    {{
-      "id": 1,
-      "question_text": "The question to ask",
-      "question_type": "open_ended|multiple_choice|likert_scale|yes_no|rating",
-      "options": ["option1", "option2"],
-      "required": true|false,
-      "help_text": "Optional clarification"
-    }}
-  ],
-  "closing": "Thank you message"
-}}
-
-For open-ended questions, omit the "options" field.
-For multiple choice and Likert questions, include appropriate options.
-Ensure questions follow best practices and are unbiased."""
+        # For T5 models, we need a very simple, direct instruction
+        return f"""Task: Generate a JSON survey.
+
+Topic: {outline}
+Questions needed: {num_questions}
+Audience: {target_audience}
+Type: {survey_type}
+
+Required JSON format:
+{{"title": "Survey Title Here", "introduction": "Welcome message here", "questions": [{{"id": 1, "question_text": "Your first question?", "question_type": "open_ended", "required": true}}, {{"id": 2, "question_text": "Your second question?", "question_type": "open_ended", "required": true}}], "closing": "Thank you message here"}}
+
+Generate the complete survey JSON now:"""
 
     def _parse_survey_response(self, response: str) -> Dict:
         """Parse LLM response into survey structure"""
@@ -132,6 +90,12 @@ Ensure questions follow best practices and are unbiased."""
             end = response.find("```", start)
             response = response[start:end].strip()
 
+        # Try to find JSON object in response
+        if "{" in response and "}" in response:
+            start = response.find("{")
+            end = response.rfind("}") + 1
+            response = response[start:end]
+
         try:
             survey_data = json.loads(response)
 
@@ -147,8 +111,49 @@ Ensure questions follow best practices and are unbiased."""
 
             return survey_data
 
-        except json.JSONDecodeError as e:
-            raise Exception(f"Failed to parse survey JSON: {str(e)}\nResponse: {response}")
+        except (json.JSONDecodeError, ValueError) as e:
+            # Fallback: Try to create a simple survey from the response
+            print(f"Warning: JSON parsing failed, attempting fallback. Error: {e}")
+            return self._create_fallback_survey(response)
+
+    def _create_fallback_survey(self, response: str) -> Dict:
+        """Create a basic survey structure from non-JSON response"""
+        # Extract potential questions from numbered list
+        lines = [line.strip() for line in response.split('\n') if line.strip()]
+
+        # Look for numbered items or lines with question marks
+        questions = []
+        question_id = 1
+
+        for line in lines:
+            # Remove leading numbers, bullets, etc.
+            clean_line = line.lstrip('0123456789.-) ')
+
+            # Check if it looks like a question
+            if len(clean_line) > 10 and (clean_line.endswith('?') or
+                                          any(word in clean_line.lower() for word in ['what', 'how', 'why', 'when', 'where', 'which', 'would', 'could', 'should', 'do you'])):
+                questions.append({
+                    "id": question_id,
+                    "question_text": clean_line,
+                    "question_type": "open_ended",
+                    "required": True
+                })
+                question_id += 1
+
+        # If we didn't find enough questions, create generic ones
+        if len(questions) < 3:
+            questions = [
+                {"id": 1, "question_text": "What are your thoughts on this topic?", "question_type": "open_ended", "required": True},
+                {"id": 2, "question_text": "Can you describe your experience?", "question_type": "open_ended", "required": True},
+                {"id": 3, "question_text": "What suggestions do you have for improvement?", "question_type": "open_ended", "required": True}
+            ]
+
+        return {
+            "title": "Survey",
+            "introduction": "Thank you for participating in this survey. Please answer the following questions.",
+            "questions": questions[:10],  # Limit to 10 questions
+            "closing": "Thank you for your time and feedback!"
+        }
 
     def refine_question(self, question: str, improvement_type: str = "clarity") -> str:
         """