コミット履歴

作者 SHA1 メッセージ 日付
  Piotr Wilkin 5f5e30007c Dilution n_seqs -> 1 3 ヶ月 前
  Piotr Wilkin eb0a15fc9b n_tokens -> n_seq_tokens 3 ヶ月 前
  Piotr Wilkin ee52fe36f3 Modify sanity check to handle hybrid models 3 ヶ月 前
  Piotr Wilkin 0dd6110fdc v1.0 3 ヶ月 前
  Piotr Wilkin adcbd9428f Linear layer output convergence 3 ヶ月 前
  Piotr Wilkin 666fc0583d Parity on delta! 3 ヶ月 前
  Piotr Wilkin a2c7b6794e Proper handling for n_tokens > GGML_DELTA_NET_CHUNK 3 ヶ月 前
  Piotr Wilkin c1e46f62fa Achieve pre-chunk-attention parity; remove most of the LLM generated crap 3 ヶ月 前
  Piotr Wilkin c87e8d550c Tensor preparation for delta_net complete 3 ヶ月 前
  Piotr Wilkin 7ec2df64a4 Added: tri, cumsum. Still a mess. 3 ヶ月 前
  Piotr Wilkin 6d0ad37cf4 Fix QKV extraction post-convolution 4 ヶ月 前
  Piotr Wilkin 845a3d7166 Convolution 4 ヶ月 前
  Piotr Wilkin 638057a29b Transpose input for convolution 4 ヶ月 前
  Piotr Wilkin 835d389fc5 Fix BA views as well 4 ヶ月 前
  Piotr Wilkin 594c1f98ef QKV splits done right 4 ヶ月 前
  Piotr Wilkin dbd4d97cf2 Fix cb calls 4 ヶ月 前
  Piotr Wilkin 32dcee47ef Some attempts to get the convolution input right. 4 ヶ月 前
  Piotr Wilkin 397cd9fd67 Fix whitespace / end-of-line newline issues. 4 ヶ月 前
  Piotr Wilkin 5a8463f4a6 Add missing LFM2 code 4 ヶ月 前
  Piotr Wilkin 64de434118 Fixes from main branch 4 ヶ月 前
  Piotr Wilkin 7bedf4c66c Refactor llama-model.cpp 4 ヶ月 前
  Piotr Wilkin 9014feadfa Change RoPE to NeoX 4 ヶ月 前
  Piotr Wilkin f020baa466 Normal attention: apply gate before output 4 ヶ月 前
  Piotr Wilkin 27fa5f335d Correct convolution state dimension calculations 4 ヶ月 前
  Piotr Wilkin e24c9dfa60 Remove OP_DELTA_NET, fix flake8 and editorchecker because why not 4 ヶ月 前
  Piotr Wilkin 6e3abeb6c0 Exclude MTP layers in conversion 4 ヶ月 前
  Piotr Wilkin 43eb7a7757 Now that eval's running move delta net stuff back to llama-model, add cbs 4 ヶ月 前
  Piotr Wilkin 890fa2c1e3 WE HAVE OUTPUT! 4 ヶ月 前
  Piotr Wilkin e590a75905 Cleanup complete, now for the recurrent memory management... 4 ヶ月 前
  Piotr Wilkin 2b0673c315 Cleanup ggml_delta_net 4 ヶ月 前