jobergum

vespa-fbench -n 1 -c 0 -q /tmp/query.txt -p 5 -s 120 localhost 8080

*** HTTP keep-alive statistics ***
connection reuse count -- 7297
***************** Benchmark Summary *****************
clients:                       1
ran for:                      24 seconds
cycle time:                    0 ms
lower response limit:          0 bytes
skipped requests:              0

jobergum

<engine>
      <proton>
        <tuning>
          <searchnode>
            <summary>
              <io><read>directio</read></io>
              <store>
                <cache>
                  <maxsize-percent>5</maxsize-percent>
                  <compression><type>none</type></compression>

jobergum

schema passage {
   document passage {
     field doc_id type long {} #duplicated for every passage extracted from doc
     field title type string {} #duplicated for every passage from doc
     field passage type string {}
     field embedding type tensor<float>(x[768]) {}
    }
    rank-profile hybrid {
       inputs {
         query(q): tensor<float>(x[768])

jobergum

import json
import numpy as np

for i in range(0,10000000):
  doc = {
    "put": "id:foo:grouping::%i" % i,
    "fields": {
      "a1": np.random.randint(0,10),
      "a2": np.random.randint(0,10),
      "a3": np.random.randint(0,10),

jobergum

{
trace: {
children: [
{
message: "Using query profile 'default' of type 'root'"
},
{
children: [
{
timestamp: 0,

jobergum

<?xml version="1.0" encoding="utf-8" ?>
<services version="1.0">
 <admin version="2.0">
  <!-- Logserver where logs are forwarded from all nodes -->
  <logserver hostalias="admin0" />
   <!-- Configserver stores state in zookeeper a-->
  <configservers>
    <configserver hostalias="admin1" />
    <configserver hostalias="admin2" />
    <configserver hostalias="admin3" />

jobergum

{
  "yql": "select id from vector where ([{\"targetHits\": 100, \"approximate\":true}]nearestNeighbor(doc_embedding,q1_embedding);",
  "hits": 100
  "ranking.features.query(q1_embedding): [0.21,0.12,....],
  "ranking.features.query(q2_embedding): [0.21,0.12,....],
  "ranking.features.query(threshold): 0.01,
  "ranking.profile": "threshold-something-else" 
}

jobergum

<container id='feed' version='1.0'>
    <document-api/>
    <nodes>
      <node hostalias="feednode1"/>
      <node hostalias="feednode2"/>
      <node hostalias="feednode3"/>
    </nodes>
</container>

<container id='default' version='1.0'>

jobergum

import torch 
from transformers import BertPreTrainedModel
from transformers import BertModel

class SentenceEncoder(BertPreTrainedModel):

    def __init__(self,config):
        super().__init__(config)
        self.bert = BertModel(config)
        self.init_weights()

jobergum

[
    {
        "question": "who ran the fastest 40 yard dash in the nfl",
        "answers": [
            "Jakeem Grant",
            "John Ross"
        ],
        "ctxs": [
            {
                "id": 6036353,