Add support for BERT embedding models #5423
iamlemec
merged
21 commits into
ggml-org:master
from
iamlemec:bert
BERT WIP
7286b83d
merge from master
ef10d786
it runs; tokenization is messed up; pooling is wrong for multi batches
0051c82d
add in wordpiece tokenizer
59c1829b
put causal_attn flag in gguf
5f1c21d0
Merge remote-tracking branch 'origin/master' into bert
e0e14e31
Merge remote-tracking branch 'upstream/master' into bert
7218c7b6
Update convert-hf-to-gguf.py
e3efcf13
add causal attention gguf key
96d37f8d
use ctx_output for tok_norm of BERT and BLOOM
e78388d3
bert : add some missing graph callbacks
b14c457f
fix up model sizing and result acquisition
68758083
hard-code token_type = 0
d080bebc
Merge branch 'bert' of github.com:iamlemec/llama.cpp into bert
3a1895d7
ggerganov
commented
on 2024-02-09
style fixes
961e98f2
undo attempted type_embd simplify
56afb2f6
bert : simplify token type embedding access
ab49e9ee
flake8 : add W503 to ignore list
6972e7e9
ggerganov
commented
on 2024-02-11
minor : code style normalization
8fbefed1
ggerganov
approved these changes
on 2024-02-11
avoid use of ggml_graph_get_tensor
e379e8c1
Merge branch 'bert' of github.com:iamlemec/llama.cpp into bert
61bab478
Reviewers
Assignees
No one assigned
Labels
enhancement
model
Review Complexity : High
Milestone
No milestone
Login to write a write a comment.
Login via GitHub