 Piotr Wilkin
|
666fc0583d
Parity on delta!
|
3 months ago |
|
Piotr Wilkin
|
a2c7b6794e
Proper handling for n_tokens > GGML_DELTA_NET_CHUNK
|
3 months ago |
|
Piotr Wilkin
|
c1e46f62fa
Achieve pre-chunk-attention parity; remove most of the LLM generated crap
|
3 months ago |
|
Piotr Wilkin
|
c87e8d550c
Tensor preparation for delta_net complete
|
3 months ago |
|
Piotr Wilkin
|
7ec2df64a4
Added: tri, cumsum. Still a mess.
|
3 months ago |
|
Piotr Wilkin
|
6d0ad37cf4
Fix QKV extraction post-convolution
|
3 months ago |
|
Piotr Wilkin
|
845a3d7166
Convolution
|
3 months ago |
|
Piotr Wilkin
|
638057a29b
Transpose input for convolution
|
3 months ago |
|
Piotr Wilkin
|
835d389fc5
Fix BA views as well
|
3 months ago |
|
Piotr Wilkin
|
594c1f98ef
QKV splits done right
|
3 months ago |
|
Piotr Wilkin
|
dbd4d97cf2
Fix cb calls
|
3 months ago |
|
Piotr Wilkin
|
32dcee47ef
Some attempts to get the convolution input right.
|
3 months ago |
|
Piotr Wilkin
|
397cd9fd67
Fix whitespace / end-of-line newline issues.
|
3 months ago |
|
Piotr Wilkin
|
5a8463f4a6
Add missing LFM2 code
|
3 months ago |
|
Piotr Wilkin
|
64de434118
Fixes from main branch
|
3 months ago |
|
Piotr Wilkin
|
7bedf4c66c
Refactor llama-model.cpp
|
3 months ago |
|
Piotr Wilkin
|
9014feadfa
Change RoPE to NeoX
|
3 months ago |
|
Piotr Wilkin
|
f020baa466
Normal attention: apply gate before output
|
3 months ago |
|
Piotr Wilkin
|
27fa5f335d
Correct convolution state dimension calculations
|
3 months ago |
|
Piotr Wilkin
|
e24c9dfa60
Remove OP_DELTA_NET, fix flake8 and editorchecker because why not
|
3 months ago |
|
Piotr Wilkin
|
6e3abeb6c0
Exclude MTP layers in conversion
|
3 months ago |
|
Piotr Wilkin
|
43eb7a7757
Now that eval's running move delta net stuff back to llama-model, add cbs
|
3 months ago |
|
Piotr Wilkin
|
890fa2c1e3
WE HAVE OUTPUT!
|
3 months ago |
|
Piotr Wilkin
|
e590a75905
Cleanup complete, now for the recurrent memory management...
|
3 months ago |
|
Piotr Wilkin
|
2b0673c315
Cleanup ggml_delta_net
|
3 months ago |
|
Piotr Wilkin (ilintar)
|
72c98b0c7d
Merge pull request #1 from ggml-org/xsn/qwen3next_experiment
|
3 months ago |
 Xuan Son Nguyen
|
e83ef74733
one less magic number
|
3 months ago |
|
Xuan Son Nguyen
|
f643b957f4
refactor softplus fn
|
3 months ago |
|
Xuan Son Nguyen
|
46110e0630
split q_proj/gate
|
3 months ago |
|
Piotr Wilkin
|
9832f2934a
Remove comments as half of them are wrong anyways
|
3 months ago |
