fix: handle serverless 404 by falling back to DialoGPT/distilgpt2; default serverless model to DialoGPT

feat(endpoint): support HF_ENDPOINT_URL (OpenAI-style chat completions) using HUGGINGFAC_API_KEY_2; keep serverless fallback

fix(chat): stop dumping company overview; use training answer at >=0.85 similarity; only show overview for generic queries

Serverless-safe defaults: fallback from LLaMA/TinyLlama to DialoGPT; 404 auto-fallback to distilgpt2; robust prompt/cleanup for both families

Default to TinyLlama; treat TinyLlama like Llama; add 404 fallback to TinyLlama; update available models

Guard against DEFAULT_MODEL=gpt2: force Llama-3.2-1B; support standard API key envs; add Llama prompt/cleanup; refresh models list

Fix: switch default model from gpt2 to Llama-3.2-1B; add Llama prompt/cleanup; update models list; add test scripts

Add company overview synthesis when query mentions 'textilindo' without 95% match

Increase minimum threshold in find_best_match to 85%

Increase similarity threshold to 95% and add debug logging

Fix all indentation errors in app.py

Simplify generate_response logic to only use training data for 90%+ similarity matches

Disable auto-training for debugging weather question issue

Add explicit weather word detection to force general question classification

Fix syntax errors in app.py

Simplify general question detection logic

Add debug logging for training data matching

Add debug logging for general question detection

Add more general question indicators including 'right now'

Move general question detection before training data check

Add general question detection to skip training data for weather/time questions

Improve similarity matching by filtering common words to prevent false positives

Prioritize high similarity training data matches (90%+) over AI model

Fix question classification: prevent general questions from using training data

Remove HTML interface, keep API only

Add better error logging and contextual fallback responses

Switch to GPT-2 model for better API reliability

Fix AI response priority: Use AI model for general questions, training data only for business questions

Fix 404 error: Switch from Llama 3.1 8B to DialoGPT-medium (available on HF API)

Add auto-training functionality - continuous background training with trained response cache

Configure to use HUGGINGFAC_API_KEY_2 and Meta Llama 3.1 8B as default model

Add Llama models to AI training: Include Llama 3.1 8B, 3.2 1B, and 3.2 3B as training options

Add AI training functionality: Integrate training scripts with web interface and API endpoints

Update documentation and add training data integration

Integrate training data: Use actual training data instead of mock responses for intelligent AI responses

Fix indentation error in get_gpu_info function

Enhance mock responses with better context awareness and general knowledge

Switch to DialoGPT-small with proper formatting

Force GPT-2 model selection

Switch to GPT-2 model with proper formatting

Simplify to use DialoGPT-medium with proper formatting

Switch to DialoGPT-large and add better logging

Switch to Llama-2-7b-chat-hf model with proper chat formatting

Switch to DialoGPT-small and add direct AI response for non-Textilindo questions

Add better logging and direct AI test endpoint

Improve keyword matching for better responses

Fix model compatibility and improve chat interface

Fix short_description length in README.md

Fix README.md YAML frontmatter configuration for HF Spaces

Simplify app.py and requirements.txt for HF Spaces compatibility