Commit History

Author SHA1 Message Date
  Guillaume Wenzek 1756897d23 simplify _finalize_hypothesis 1 year ago
  Guillaume Wenzek f49763de86 fix beam size and scores 1 year ago
  Guillaume Wenzek c28db8c8ac wip beam-size=2 1 year ago
  Guillaume Wenzek b24dbe3030 batching -> TransformerEmbeddingFrontend_forward 1 year ago
  Guillaume Wenzek 86993cbd00 fix StandardTransformerEncoder 1 year ago
  Guillaume Wenzek 28ed039370 fix MultiheadAttention_forward 1 year ago
  Guillaume Wenzek 81cdf80eb9 WIP: MultiheadAttention_forward 1 year ago
  Guillaume Wenzek eb80195345 use ggml_diag_mask_inf 1 year ago
  Guillaume Wenzek 88b0690a72 split tests files 1 year ago
  Guillaume Wenzek bfbafd9603 fix generation with beam_size=1 1 year ago
  Guillaume Wenzek 45f986055a add naive tweaking of lprobs 1 year ago
  Guillaume Wenzek 7c9b2a1b95 pass correct prefix sequence in test 1 year ago
  Guillaume Wenzek c7b89f32f4 disable flash attn because of cross attention 1 year ago
  Guillaume Wenzek dcb9535666 wip: generate_sequence 1 year ago
  Guillaume Wenzek 78e7c9a311 fix TransformerEmbeddingFrontend 1 year ago
  Guillaume Wenzek 2238cea072 SinusoidalPositionEncoder + WIP: TransformerEmbeddingFrontend 1 year ago
  Guillaume Wenzek 2fb09f34fb generate fairseq2.cpp 1 year ago
  Guillaume Wenzek f1f33dbec1 has_layer + transformer decoder 1 year ago
  Guillaume Wenzek e1faba0c5d text encoder 1 year ago
  Guillaume Wenzek b81061704c StandardTransformerEncoderLayer_forward 1 year ago
  Guillaume Wenzek e7c3b7a4ba allow flash attn 1 year ago
  Guillaume Wenzek b07a08102a MultiheadAttention_forward 1 year ago
  Guillaume Wenzek be236b8584 remove size/init functions 1 year ago
  Guillaume Wenzek fa85f05545 test flash_attn 1 year ago
  Guillaume Wenzek 06d4ed1475 ggml.ne != np.shape 1 year ago
  Guillaume Wenzek c2e6384e29 nb 1 year ago
  Guillaume Wenzek 3f5912b973 forward 1 year ago
  Guillaume Wenzek 772f90dfdc load_fairseq2_ggml_file 1 year ago
  Guillaume Wenzek 506dee42d8 move layers to fairseq2.cpp 1 year ago