 Guillaume Wenzek
|
353a001d64
fix merge
|
il y a 1 an |
|
Guillaume Wenzek
|
f2ef995b95
format/isort
|
il y a 1 an |
|
Guillaume Wenzek
|
20fea08676
fairseq2 churn ebf637dc52b42850e086eebe2d60a45a89881aad
|
il y a 1 an |
|
Guillaume Wenzek
|
2f71737adc
skip tests that can't run locally
|
il y a 1 an |
|
Guillaume Wenzek
|
d5b035f230
fix PositionalEmbedding with cache
|
il y a 1 an |
|
Guillaume Wenzek
|
ddff1a0644
add failing test: test_PositionalEmbedding_forward_with_cache
|
il y a 1 an |
|
Guillaume Wenzek
|
8c074387a9
s2tt generate the sample if needed
|
il y a 1 an |
|
Guillaume Wenzek
|
a771adc782
nicer ctypes_utils
|
il y a 2 ans |
|
Guillaume Wenzek
|
f6d810543d
fix incremental decoding !
|
il y a 2 ans |
|
Guillaume Wenzek
|
522b97234e
WIP: simple failing test case
|
il y a 2 ans |
|
Guillaume Wenzek
|
cc23e2b1c7
prepare tests for flash_attn
|
il y a 2 ans |
|
Guillaume Wenzek
|
18c919b0b8
update flash_attn tests
|
il y a 2 ans |
|
Guillaume Wenzek
|
eb7810b81f
force little-endian
|
il y a 2 ans |
 Ning
|
d9061b89b3
Update test_unity_cpp.py
|
il y a 2 ans |
|
Ning
|
1589c529bf
unity.cpp speech_encoder_frontend+speech_encoder+adaptor (#81)
|
il y a 2 ans |
|
Guillaume Wenzek
|
e38442d1f0
rename mask to attn_mask in MultiheadAttention_forward
|
il y a 2 ans |
|
Guillaume Wenzek
|
6fbb465f2b
generate_sequence return full results
|
il y a 2 ans |
|
Guillaume Wenzek
|
f49763de86
fix beam size and scores
|
il y a 2 ans |
|
Guillaume Wenzek
|
c28db8c8ac
wip beam-size=2
|
il y a 2 ans |
|
Guillaume Wenzek
|
3d69d4975f
batching -> StandardTransformerDecoder
|
il y a 2 ans |
|
Guillaume Wenzek
|
b24dbe3030
batching -> TransformerEmbeddingFrontend_forward
|
il y a 2 ans |
|
Guillaume Wenzek
|
86993cbd00
fix StandardTransformerEncoder
|
il y a 2 ans |
|
Guillaume Wenzek
|
28ed039370
fix MultiheadAttention_forward
|
il y a 2 ans |
|
Guillaume Wenzek
|
81cdf80eb9
WIP: MultiheadAttention_forward
|
il y a 2 ans |
|
Guillaume Wenzek
|
f2b5007277
start testing with batch size > 1
|
il y a 2 ans |
|
Guillaume Wenzek
|
eb80195345
use ggml_diag_mask_inf
|
il y a 2 ans |
|
Guillaume Wenzek
|
88b0690a72
split tests files
|
il y a 2 ans |
|
Guillaume Wenzek
|
bfbafd9603
fix generation with beam_size=1
|
il y a 2 ans |
|
Guillaume Wenzek
|
45f986055a
add naive tweaking of lprobs
|
il y a 2 ans |
|
Guillaume Wenzek
|
7c9b2a1b95
pass correct prefix sequence in test
|
il y a 2 ans |
