mirror of
https://github.com/pese-git/llm-arch-research.git
synced 2026-01-23 21:10:54 +00:00
- Implemented core/gpt2_decoder.py: transformer decoder block with kv cache in GPT2 style - Refactored models/gpt/gpt2.py to use new Gpt2Decoder, improved documentation - Added tests/core/test_gpt2_decoder.py for main features and cache - Temporarily skipped HF proxy integration test for compatibility