Spaces:
Sleeping
Sleeping
Commit History
-
Phase 2 Quick Fix: Switch to clean Qwen2.5-0.5B-Instruct base model
Verified signatureCRITICAL FIX: Replaced corrupted anktechsol/anki-2.5 model with clean Qwen/Qwen2.5-0.5B-Instruct base model. Issue: Previous model (checkpoint-1000) was producing gibberish outputs with hallucinations and token artifacts. Solution: Switching to official Qwen2.5-0.5B-Instruct model which is properly instruction-tuned and should provide coherent responses. This is a temporary fix. Next step: Proper LoRA fine-tuning on quality chat data from scratch.e89521f -
Phase 1 Quick Wins: Optimize inference for 3-5x speed boost
Verified signatureImplemented Phase 1 optimizations: - Changed torch.float32 to torch.float16 (50% memory reduction, 2x faster) - Added tokenizer.pad_token configuration - Replaced token-by-token loop with batch generation using TextIteratorStreamer - Added no_repeat_ngram_size=3 to prevent repetitive text - Added repetition_penalty=1.1 for better output quality - Imported TextIteratorStreamer and Thread for efficient streaming Expected results: 3-5x faster inference (60-100 tokens/sec vs 20-30 tokens/sec)9c878f1
-
Remove OAuth config and update README for direct model loading
Verified signature76dd32e -
Fix app.py: Replace InferenceClient with direct model loading using transformers
Verified signatured5f4c16 -
Add requirements.txt with transformers dependencies
Verified signature1624583
-
Update app.py
Verified signatured22c6b3
-
Remove qwen references and update to Anki 2.5
Verified signaturedf3e993 -
initial commit
Verified signaturef8d7cbd -
Duplicate from gradio-templates/chatbot
Verified signatureCo-authored-by: Pete <pngwn@users.noreply.huggingface.co>8f136d4
