configdefinitions/src/vespa/col-bert-embedder.def


1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36

namespace=embedding

# Path to tokenizer.json
tokenizerPath model

# Path to model.onnx
transformerModel  model

# Max query tokens for ColBERT
maxQueryTokens int default=32

# Max document query tokens for ColBERT
maxDocumentTokens int default=512

# Max length of token sequence model can handle
transformerMaxTokens int default=512

# Input names
transformerInputIds      string default=input_ids
transformerAttentionMask string default=attention_mask

# special token ids
transformerStartSequenceToken int default=101
transformerEndSequenceToken   int default=102
transformerMaskToken   int default=103

# Output name
transformerOutput string default=contextual

# Settings for ONNX model evaluation
transformerExecutionMode enum { parallel, sequential } default=sequential
transformerInterOpThreads int default=1
transformerIntraOpThreads int default=-4
# GPU device id, -1 for CPU
transformerGpuDevice      int default=0