david167/question-generation-api

Enable concurrency_limit=10 for better parallel processing

Upgrade to Llama 3.1 8B-Instruct for better long-form content

Fix build error: Remove flash-attn dependency

Speed optimizations: Switch to Mistral-7B + optimize generation params

COMPLETE API REBUILD: ZERO TRUNCATION PRINCIPLE - Intelligent extraction, generous tokens, never cut content

ULTRA CONSERVATIVE EXTRACTION: Find JSON array boundaries properly, extensive logging, no aggressive cutting

FIX TRUNCATION: Improved response extraction logic, conservative cutting, detailed logging - NO MORE TRUNCATION && git push

FIX TUPLE ISSUE: Return single string output instead of tuple - eliminates ('content', '') wrapper

DIRECT CONTENT API: Return just the generated content, no API wrappers or tuples - perfect for client parsing

ULTRA SIMPLE FIX: Remove ALL JSON components, use only Textbox inputs/outputs, no state anywhere

BULLETPROOF API: Remove ALL State components, use JSON inputs instead, proper input/output matching, ZERO GRADIO ERRORS

SIMPLE WORKING API: Fix Gradio interface issues, use simple Interface instead of Blocks, proper API structure

Fix Gradio interface: Remove chatbot format issues, add proper API endpoint structure

ELEGANT API REWRITE: Clean architecture, smart token allocation, proper JSON extraction - eliminate placeholder generation

MAXIMUM TOKEN SETTINGS: Use 131k context, 16k max_new_tokens, 2k min_tokens for CoT - eliminate all truncation

Aggressive fix for CoT truncation: increase min_new_tokens to 1500, suppress EOS token for CoT requests, cap max_new_tokens

Fix CoT truncation: increase min_new_tokens to 1000, add generation logging, improve truncated JSON handling

Improve Chain of Thinking support: increase min_new_tokens to 500 for CoT requests, improve JSON bracket tracking for nested objects

Fix response truncation - improve extraction logic to find actual content start

Simplify API - remove all templates, just prompt-in response-out

Add 'list' template for better summarization with specific content extraction

Fix JSON templates - use instructional format instead of literal examples

Fix response extraction - prevent truncation at beginning of JSON responses

Fix JSON array generation - add explicit array requirements and improve JSON parsing

Increase max_new_tokens to 8192 for unlimited length responses

Improve model generation parameters and add logging - fix response truncation issues

Fix JSON response templates for better prompt generation - simplified templates for more reliable JSON parsing

Fix invisible text with comprehensive CSS targeting

Fix font visibility: Add dark text colors for better contrast

Fix UI: Reduce dialog size and prevent input focus layout shifts

Fix requirements.txt - Add missing transformers and ML dependencies

Major update: Add NFL training data generation and improve model handling

Fix layout jumping when focusing input field

Add complete JSON functionality to Gradio interface

Add JSON imports for structured response functionality

Update UI for full-width display on big screens with responsive design

Fix: Add missing generated_text variable definition

DEBUG: Show complete raw model output and prompt to identify clipping source

COMPLETE REWRITE: Clean ChatGPT-style interface with proper response handling

TEMPORARY: Show full model response for debugging clipping issue

Fix response truncation: disable early stopping, increase token limits to 4096, add debugging logs

Fix response clipping: use robust assistant header detection instead of prompt length

Replace with simplified raw chat interface for prompt testing

Fix syntax errors: correct comma placement and indentation

Fix remaining device_map auto in gradio_app.py

Force all CUDA operations to cuda:0 and use device_map to prevent multi-GPU distribution

Fix multi-GPU device placement error: disable device_map auto and ensure tensors on same device

Fix duplicate import in app.py

Switch to Llama-3.1-8B-Instruct: update model loading, prompts, and generation parameters

Switch back to Llama-3.1-8B-Instruct model: update prompts, generation params, and UI descriptions