I am trying to follow the example from the official MongoDB documentation.
It mentioned that the search vector was based on the string "historical heist" using the OpenAI text-embedding-ada-002 model. However, when I tried to validate this with my own OpenAI API key, I got different results.
Why?
curl https://api.openai.com/v1/embeddings \
-H "Content-Type: application/json" \
-H "Authorization: Bearer sk-xxxxx" \
-d '{
"input": "historical heist",
"model": "text-embedding-ada-002",
"encoding_format": "float"
}'
My results are the following:
"data": [
{
108 "object": "embedding",
"index": 0,
19 "embedding": [
8 -0.02273331,
97 -0.027646419,
0.0030520316,
3 -0.033631727,
1 -0.026207773,
4 0:00:01 --:- 0.0016854883,
-:-- 0:0 -0.017983066,
0:0 0.006311038,
1 -0.0013012275,
201 -0.024823416,
54 0.015227924,
0.005751188,
0.0004003777,
-0.008095773,
The example results are the following:
'queryVector': [-0.020156775, -0.024996493, 0.010778184, ....
There was a discussion on the official OpenAI forum. People are saying that they get a different embedding vector for the identical input.
The first reply from Curt Kennedy suggests that there seems to be a difference between the OpenAI API endpoint and the Azure API endpoint for the same model.
Embeddings only return vectors. The vector is the same for the same input, same model, and the same API endpoint. But we have seen differences between the OpenAI endpoint and the Azure endpoint for the same model. So a pick an endpoint and stick with it to avoid any differences.
Later on, Curt Kennedy adds that some variance is expected.
Some variation is expected because of the random timing in the GPU’s, and that floating point is not associative, and they are likely taking the last hidden layer and scaling out to the unit hyper-sphere, which would magnify the error for hidden states close to the origin.
I wrote the following Python code to make a test:
# Imports
import os
import numpy as np
from openai import OpenAI
# Initialize OpenAI client
client = OpenAI(
api_key=os.getenv("OPENAI_API_KEY"),
)
# Vector from MongoDB documentation
vector_from_mongodb = np.array([-0.020156775, -0.024996493, 0.010778184, -0.030058576, -0.03309321, 0.0031229265, -0.022772837, 0.0028351594, 0.00036870153, -0.02820117, 0.016245758, 0.0036232488, 0.0020519753, -0.0076454473, 0.0073380596, -0.007377301, 0.039267123, -0.013433489, 0.01428371, -0.017279103, -0.028358135, 0.0020160044, 0.00856761, 0.009653277, 0.0107912645, -0.026683854, 0.009594415, -0.020182934, 0.018077003, -0.015709465, 0.003310956, 0.0014878864, -0.015971072, -0.002411684, -0.029561523, -0.030450987, -0.013106481, -0.005385822, -0.018652538, 0.012642129, -0.005189617, 0.018835662, -0.0048102876, -0.0261214, -0.016167276, -0.007972456, 0.0023381072, -0.010058766, -0.009012341, 0.008358325, 0.018665617, 0.02163485, -0.012975678, -0.010745483, -0.002571918, -0.014479915, 0.007226877, 0.015003128, 0.013165343, -0.028279653, 0.0053727417, -0.020588424, -0.017383745, 0.023518417, 0.01262905, -0.011922712, 0.007638907, -0.0073249796, -0.014859244, -0.00001101736, 0.017043658, 0.010111088, 0.0074623227, 0.009555174, 0.008338705, -0.002240005, -0.0010603234, -0.004973792, 0.003391073, 0.021543289, 0.013341927, 0.0005980159, 0.010693162, 0.005336771, 0.016062634, 0.005768421, 0.005186347, 0.039790336, 0.0021942237, -0.0026275094, 0.010431555, 0.0042151334, -0.0050359233, 0.025768232, -0.021451725, 0.01833861, -0.01836477, -0.013433489, 0.030006256, -0.014793842, 0.017475309, 0.0020585153, -0.012975678, -0.017266022, -0.01593183, -0.014257549, 0.0010676811, -0.007887433, -0.0045911926, 0.00012303676, -0.0014976967, 0.03552615, 0.0065630507, -0.037435878, 0.011929252, -0.00939167, 0.016768971, 0.01223664, 0.007789331, -0.037200432, 0.013145722, 0.00896002, 0.021857215, 0.010333453, 0.021582529, -0.007089534, -0.007154935, -0.02485261, 0.0040254686, -0.00088864425, 0.023466095, -0.020719228, -0.006690584, -0.021006994, -0.018286288, 0.025545865, -0.0096598165, 0.008803056, -0.023021365, -0.040078104, 0.015408617, 0.017043658, -0.011242535, 0.0063537657, -0.026618453, 0.0071614753, -0.014623798, 0.00067322777, -0.00083427917, -0.028070368, 0.03714811, -0.004529061, 0.0054087127, 0.0028727653, 0.008384486, 0.010026066, -0.006190262, -0.0002493436, 0.0029953935, -0.026226042, -0.018417092, 0.009941043, 0.0036494094, -0.00982332, 0.013551212, 0.02574207, -0.0022645304, -0.0006004685, 0.012805633, -0.024303235, 0.008194821, -0.014179068, -0.02977081, 0.003095131, -0.0015941641, 0.029953934, 0.0052680993, 0.025388902, -0.031392768, -0.021386323, 0.014898485, 0.022419669, 0.00897964, 0.013243824, 0.006854088, 0.0066415328, -0.003839074, -0.01877026, 0.021216279, -0.015055449, -0.0015508354, 0.013211124, -0.008783435, 0.0052157775, -0.68938524, -0.01221702, -0.04125533, -0.016232677, 0.020039052, -0.0026422248, -0.0037050007, 0.0064682183, -0.0047579664, 0.0032749851, -0.0035382267, 0.031942144, -0.00035643874, -0.011628405, -0.043086577, -0.0196074, -0.0066088317, -0.014872325, 0.028331975, 0.010294212, -0.013930541, 0.031994462, -0.018626377, 0.017462227, 0.026343765, -0.010274592, 0.0046827546, -0.029430721, -0.011746128, 0.0024362097, 0.0023054064, 0.0027730279, -0.002406779, 0.003917556, 0.059436977, 0.008665713, -0.0018901062, 0.06037876, 0.017880797, 0.05185039, 0.0067102043, -0.020300657, 0.005604917, 0.018704858, 0.012073136, 0.0144145135, 0.012413224, -0.0074819434, 0.015801027, -0.0061412104, 0.008613391, -0.0039077457, -0.0036232488, 0.008469507, 0.014087505, 0.0124066835, 0.019267311, -0.002573553, 0.005055544, -0.009417831, -0.009103903, 0.011150973, -0.012046975, 0.0058567133, -0.0053727417, 0.018260127, -0.005588567, 0.015591742, 0.007495024, -0.02567667, 0.024211673, 0.021386323, -0.012890656, -0.016114954, 0.009515933, 0.009679437, 0.025532786, -0.0076454473, -0.02575515, 0.008319084, -0.0068410076, -0.017082898, -0.026173722, -0.0049901423, 0.01918883, -0.008646091, -0.031759016, 0.014820003, 0.011850771, 0.01836477, 0.012700991, -0.0011437106, 0.005058814, 0.0151993325, -0.0060692686, 0.027416352, 0.0037344315, 0.0013546307, 0.018325528, -0.03152357, -0.008809595, 0.014649959, -0.008345244, 0.0066415328, -0.005523165, 0.0043492066, -0.0015892589, 0.0048855, 0.034453563, -0.03837766, 0.0068410076, -0.0042151334, -0.0067429054, 0.0055689462, -0.011733048, -0.0212032, 0.016847452, -0.0022220195, 0.0059351954, -0.00449963, 0.02251123, -0.01020265, 0.023361452, -0.0032455544, 0.016180357, 0.0049443613, -0.0064747585, -0.03259616, 0.012321662, 0.020104453, 0.009954124, -0.019411195, 0.0048102876, -0.000392614, 0.012184318, 0.0044276887, 0.005634348, -0.020562263, 0.015722545, -0.005179807, -0.0067952266, 0.0027861083, 0.0024198592, -0.0020585153, 0.0018525004, -0.045100946, -0.010176489, -0.012956058, 0.0013497255, 0.0105361985, 0.003796563, -0.0106016, -0.013126101, 0.0050359233, 0.015003128, -0.0075800456, -0.015722545, -0.01755379, -0.00978408, -0.02940456, 0.017606111, 0.016612006, -0.016912855, 0.025441224, 0.0054741143, 0.00448001, 0.009470152, 0.015382457, -0.008332164, -0.019123428, 0.024564842, 0.016860534, 0.008286383, -0.007141855, 0.006559781, 0.016625088, -0.01840401, -0.011602244, -0.00489858, -0.0073184394, -0.008809595, -0.0018459603, -0.01629808, -0.005542786, 0.0064257076, 0.010379234, 0.014663039, 0.034872133, -0.013355007, 0.027285548, 0.011654565, -0.004032009, 0.02323065, -0.02653997, -0.0009941043, 0.002946342, 0.010667001, 0.008345244, 0.018626377, 0.04821406, 0.031392768, 0.010281132, 0.026069079, 0.002735422, 0.01182461, -0.01593183, 0.006585941, -0.010071847, 0.024564842, -0.0025261368, 0.004293615, -0.0068606283, -0.0066448026, -0.0074100015, -0.0014347476, 0.021530207, -0.010418476, 0.018495573, -0.0034924455, -0.014165987, -0.004784127, -0.012472086, 0.004417878, -0.0030313642, -0.010084927, -0.010954768, 0.01508161, 0.0010047321, 0.0042347535, -0.03345946, -0.00027346043, 0.014793842, -0.019882087, 0.012772933, 0.021490967, 0.0031932332, 0.0093589695, 0.00090172456, 0.0048102876, 0.0070045115, -0.0045584915, 0.015840268, 0.024342475, -0.0091300635, 0.0039796876, 0.003796563, 0.025022654, -0.008103259, -0.025022654, 0.03021554, -0.008201361, -0.0070502926, 0.0011821339, 0.021072397, 0.004849529, -0.02495725, 0.012184318, 0.0019228071, -0.007226877, 0.020562263, 0.018861823, -0.0017593032, 0.01345965, 0.0022727058, 0.003023189, -0.026971621, -0.0030558899, 0.017723834, -0.01998673, -0.010608139, 0.011491061, -0.025179617, 0.0069652707, 0.003924096, 0.021177039, 0.0045650317, -0.0009973744, 0.007586586, -0.004032009, -0.008129419, -0.010091467, -0.04279881, 0.019790525, 0.01595799, 0.0044309585, -0.0033747226, -0.018665617, -0.012818714, -0.016206518, 0.014113666, -0.0020912162, 0.01427063, -0.020248337, -0.0112752365, -0.020588424, -0.011039791, 0.008744194, -0.015147011, 0.0022269245, -0.010438096, -0.0017772885, -0.028750544, -0.008861917, -0.016991336, 0.033668745, 0.034636687, 0.009888723, 0.0023953337, 0.006991431, -0.003346927, 0.003103306, -0.0044571194, 0.011249076, 0.0033779927, 0.00012446742, -0.0027027212, -0.025859794, -0.011942333, 0.02694546, 0.028227331, 0.0064289775, -0.03385187, -0.020719228, 0.00489531, 0.10663077, 0.041752383, -0.021700252, -0.008103259, 0.0049574412, -0.01675589, -0.020182934, -0.006585941, 0.007684688, -0.002859685, 0.027023941, 0.00856107, 0.0037017306, 0.016978256, 0.025885954, -0.010372694, 0.0025964435, 0.011706887, 0.021360163, -0.021674091, -0.024983412, 0.0034074234, 0.0032030435, 0.022262705, -0.01266829, -0.002249815, 0.032779284, -0.0034303141, -0.016101874, -0.005156916, -0.0212032, 0.005362931, 0.009077743, -0.013917461, -0.0017315074, 0.010980929, -0.019450437, 0.013865139, 0.028227331, -0.008757275, -0.0033649125, -0.012857955, 0.011039791, 0.009764459, 0.00029594224, -0.026317604, 0.025048813, 0.037749805, -0.025807472, -0.005425063, 0.021791814, -0.010012985, -0.00066995766, -0.016952096, 0.0031147513, -0.016598927, 0.0084368065, 0.004787397, -0.0064355177, 0.0015164997, -0.021216279, -0.023845425, 0.013969782, -0.011255615, 0.0042576445, -0.024250913, -0.009908343, -0.02289056, -0.023361452, -0.010987469, -0.013394248, 0.0032553647, -0.019018786, 0.021438645, 0.029587684, -0.010490417, 0.01263559, -0.018417092, -0.008731114, 0.01875718, -0.0072399573, -0.029090632, -0.017736914, -0.04031355, -0.019712042, 0.012772933, -0.030320182, -0.022341188, -0.02041838, 0.011752668, 0.028829027, -0.017043658, 0.024996493, 0.006334145, -0.0024263994, -0.0077370093, 0.017802317, 0.017396826, 0.030398665, 0.011464901, 0.03016322, -0.014558396, -0.0036690298, -0.009954124, -0.006703664, -0.00035705187, -0.014519156, 0.0075342646, -0.00896656, 0.040078104, 0.024420958, -0.016886694, -0.00092543266, -0.0017494928, 0.01672973, 0.016533526, 0.002648765, 0.0187441, -0.0055460557, 0.004735076, 0.03186366, 0.0003435628, 0.007495024, 0.023453014, -0.012504786, -0.0074557825, -0.0027844731, -0.04570264, 0.010477337, 0.0030101088, -0.015670223, 0.03351178, -0.020261416, 0.00050849747, -0.009653277, -0.023466095, -0.007396921, -0.011909632, 0.003436854, -0.02979697, -0.039031677, -0.014584557, 0.0019555078, 0.0042216736, -0.0060594585, -0.023400694, -0.00023462824, -0.017763074, -0.016180357, 0.0132372845, -0.020496862, -0.007390381, -0.0058697937, -0.0096598165, 0.0039796876, -0.019306554, -0.012622509, -0.0012287326, 0.010863206, 0.024368636, 0.027730279, 0.016795132, 0.019908248, -0.006343955, 0.0014592733, -0.005425063, 0.019450437, 0.004532331, -0.031889822, 0.008476048, 0.019712042, -0.00047906674, -0.0028286192, 0.011883471, -0.012426305, 0.0041497317, 0.001756033, -0.0013603533, -0.008031317, -0.010281132, -0.0071222344, -0.026330685, -0.007920134, -0.026866978, -0.03026786, -0.0015328501, 0.027442513, -0.005922115, 0.005186347, 0.003436854, 0.036703378, -0.0053204205, 0.013165343, 0.0016939015, -0.0041431915, -0.017213702, -0.012439385, -0.015212413, 0.014532236, 0.0093589695, -0.0053400407, 0.017422987, -0.028881347, -0.014179068, 0.011307937, 0.040104263, -0.007593126, -0.000631943, -0.0003404971, -0.0055198953, -0.00063030794, -0.004852799, -0.0024214943, -0.029718488, 0.023322212, 0.011079031, 0.012988758, 0.0071614753, -0.034034993, -0.01551326, 0.004012388, 0.006442058, 0.032386873, 0.0076519875, 0.0465921, 0.01757995, -0.0135381315, -0.016978256, 0.024983412, 0.0003280299, 0.0026209692, 0.022380428, -0.010640841, 0.0027648527, -0.007959375, -0.005922115, 0.0075342646, -0.03597088, -0.018874902, 0.03510758, -0.015356296, 0.004597733, -0.0015328501, -0.019947488, -0.013446569, 0.020614585, -0.0056016473, 0.035186063, 0.0005248479, -0.030712591, -0.019136509, 0.004202053, -0.010339993, 0.014754602, 0.0072922786, -0.015460939, 0.027494833, -0.02974465, -0.0033616424, 0.0105819795, -0.028881347, 0.01720062, -0.0073707607, 0.0054479535, -0.0019522378, -0.018103164, -0.009110443, -0.024630243, 0.005624538, 0.01879642, -0.019345794, -0.0027681228, -0.015971072, 0.022354268, -0.0038194535, 0.018901063, -0.017357586, -0.02493109, 0.006703664, -0.0021173768, -0.005667049, -0.004535601, -0.016441964, 0.0034172337, -0.02447328, -0.003310956, -0.02078463, -0.011589164, 0.013263445, -0.014728441, -0.0187441, -0.019476596, 0.013224204, 0.015238573, -0.012380524, 0.00019058435, 0.010778184, 0.025022654, -0.036127847, 0.01470228, -0.007671608, 0.032857765, 0.002982313, 0.009829861, 0.0072203367, -0.0028237142, 0.025990596, -0.029012151, 0.0016955365, 0.012033895, -0.0049901423, -0.013629694, 0.0072464976, 0.0012704261, 0.0018868363, 0.017043658, 0.00448001, -0.009555174, -0.016520444, 0.02570283, -0.00939167, 0.01998673, 0.002001289, -0.023662299, 0.0041072206, -0.024839528, -0.007396921, -0.0034793653, -0.032020625, -0.0036003583, -0.010719323, 0.022995204, -0.01757995, -0.0043851775, -0.023884665, -0.018430172, -0.009018881, 0.00091562246, -0.0055689462, -0.012537487, 0.016455043, 0.03264848, 0.018560974, 0.014623798, 0.0025555675, -0.0060986993, 0.0058272826, -0.008462967, -0.012720612, -0.0042576445, -0.027207067, 0.014152907, -0.0029610575, 0.010241891, -0.011222915, -0.01140604, -0.022197304, -0.003433584, -0.0056899395, 0.004372097, 0.061896075, -0.005846903, -0.011863851, 0.004535601, -0.0074819434, 0.016847452, -0.0012647035, 0.021085477, 0.02409395, -0.030137058, -0.0012197399, 0.009607496, -0.008220982, -0.007893973, -0.007893973, 0.007972456, 0.010012985, 0.009143144, 0.0044734697, 0.015264734, -0.0032520946, 0.002208939, 0.011968493, -0.0012998568, -0.0114322, -0.056454662, -0.013217663, 0.0017593032, -0.00244275, -0.021399405, -0.010732403, 0.00694565, 0.0033207664, 0.0025539326, 0.01102671, -0.012589809, 0.010706242, -0.012413224, 0.01427063, -0.000049970913, -0.0056016473, 0.027965724, 0.018652538, -0.009535554, 0.0068867886, 0.004699105, -0.001245083, -0.009071202, -0.0032946058, -0.03756668, 0.034453563, -0.00408106, 0.013361547, -0.0065107294, 0.009300108, -0.016415803, 0.0059973267, -0.017422987, 0.0048822295, 0.022158062, -0.025611266, 0.01022227, -0.0061771814, -0.014218308, -0.00044636594, -0.019110348, -0.013747416, -0.013629694, -0.021896457, -0.0051634563, -0.020509942, -0.018731019, 0.0043328563, -0.032386873, -0.023086766, 0.0196074, 0.20614585, -0.014649959, -0.009712138, 0.01345965, -0.010928608, 0.0196074, 0.015814107, 0.017383745, -0.0024656404, 0.021399405, 0.013668935, -0.0063864663, -0.0015303975, -0.0012924991, -0.0030575248, -0.015539421, -0.009692517, -0.012190859, -0.02287748, 0.002936532, 0.00069325697, 0.013158802, -0.0070110518, -0.013629694, 0.01585335, -0.019829765, 0.013747416, 0.016036473, 0.011693806, 0.0071483953, -0.010156869, -0.013799738, -0.00034703725, -0.010706242, -0.02289056, 0.0039339066, -0.0015835363, -0.014532236, 0.012445925, -0.00009779583, 0.0053335004, 0.0055329753, -0.005281179, -0.007475403, 0.00040385488, -0.012942977, -0.015277814, 0.012956058, 0.00006162057, 0.007056833, -0.02571591, -0.018731019, -0.0061771814, 0.034427404, 0.0010570535, 0.0079528345, 0.024172433, 0.021386323, -0.019803606, -0.006821387, -0.011262156, 0.026605371, -0.0036951904, -0.008207901, -0.019698963, 0.042981934, -0.026212962, 0.00856761, 0.015173172, 0.0024149541, -0.0008036222, -0.005752071, -0.02898599, -0.008443347, -0.0064224373, -0.014479915, 0.036467932, -0.00086820626, 0.026396086, 0.002001289, -0.0074361623, -0.0086918725, -0.007835112, 0.021464806, 0.0008984545, -0.02489185, 0.019515838, 0.026644614, -0.0137212565, 0.00448982, 0.004211863, -0.022380428, -0.014100585, -0.01629808, 0.0074884836, 0.02652689, 0.011634945, 0.049626734, -0.023583818, -0.0021958589, -0.015735626, 0.02733787, 0.0036428692, -0.031261966, -0.012674831, 0.006196802, -0.009535554, 0.016886694, 0.010771644, -0.021490967, 0.014100585, -0.007063373, 0.00043778197, -0.012151618, -0.0058894143, 0.009182385, -0.005768421, -0.013995943, 0.004725266, -0.01347273, -0.020797709, -0.018037762, 0.020274498, 0.011595704, 0.0017364125, -0.02248507, 0.005954816, 0.0062196925, -0.014257549, -0.025127295, 0.015356296, 0.005179807, 0.021726413, -0.0034499345, -0.017082898, 0.019803606, 0.005209238, 0.0005939283, -0.0035807376, -0.011661106, 0.006559781, 0.0033207664, 0.0017233322, -0.00059924216, -0.000341519, -0.0140221035, 0.00084286317, -0.003306051, -0.005634348, -0.00816212, -0.009319728, -0.024447119, -0.014950806, -0.024564842, 0.0137212565, -0.010084927, 0.000044886958, -0.0033943432, 0.0025359471, 0.012478625, -0.023086766, 0.014519156, 0.020876192, -0.023282971, -0.0030804155, -0.014545316, -0.16805595, 0.01262905, 0.020719228, -0.012413224, 0.026592292, -0.0024198592, 0.041072205, 0.002658575, -0.013708176, -0.0068867886, -0.0018639456, 0.000031627806, -0.043452825, -0.028018046, -0.0105819795, 0.01266829, -0.009450532, 0.008292923, 0.0058534434, -0.006782146, 0.032229908, 0.0005955633, -0.0023103117, 0.003140912, 0.00037687673, -0.0049247406, -0.008070557, 0.017279103, -0.012759852, -0.011608784, -0.019450437, 0.016167276, 0.02248507, 0.030529467, 0.015905669, 0.0061150496, -0.016834373, 0.017344505, 0.006667693, -0.005461034, 0.0066742334, 0.01998673, 0.024591003, -0.007717389, 0.0096598165, 0.03225607, 0.018626377, -0.020248337, 0.0017740185, 0.012589809, 0.0014927916, -0.040235065, 0.01713522, 0.016206518, 0.017776156, 0.024734886, 0.0040516295, -0.009627116, 0.002001289, -0.010496957, -0.0121058365, -0.017266022, 0.008279843, -0.02122936, -0.01349889, -0.02251123, 0.004820098, -0.000071533, -0.022628954, 0.015238573, -0.01833861, -0.016572766, -0.0031523572, -0.008064018, 0.019973649, 0.0089207785, -0.03228223, 0.0040647094, -0.004784127, -0.0017920039, -0.0013775212, 0.047246117, 0.0030804155, -0.010660461, 0.02982313, 0.006088889, -0.019371955, -0.024447119, -0.011687267, -0.013708176, 0.017187541, -0.018286288, 0.019267311, 0.0011960318, 0.0046271635, 0.016886694, 0.0069129495, 0.00029062838, 0.013629694, -0.016494283, -0.017069818, 0.0058240127, 0.013943622, 0.001675916, 0.01347273, 0.023335291, 0.008129419, 0.0047187256, 0.032099105, 0.0007701039, 0.0068344674, 0.0004672127, -0.00610851, 0.026396086, -0.010738943, 0.024591003, 0.008220982, -0.019908248, 0.024682565, -0.009404751, 0.0594893, -0.009731758, -0.022628954, 0.013865139, -0.016049553, 0.0033371167, -0.107572556, -0.022341188, 0.008050937, -0.0089731, 0.004983602, 0.010771644, -0.013034539, -0.013368088, -0.0071287747, 0.0091758445, -0.017409906, -0.022118822, -0.011170594, -0.010908987, 0.050490037, 0.014584557, 0.018312449, 0.0014968792, -0.0057161, 0.024342475, -0.02699778, 0.020091372, -0.00094587065, -0.021347083, -0.003711541, 0.0016677409, -0.030738752, 0.040208906, 0.008109799, -0.017527629, -0.0009058122, 0.017776156, 0.0052779093, -0.0046206233, 0.0067952266, -0.01226934, -0.009162764, -0.01595799, 0.021582529, -0.027390191, -0.00011210243, -0.003145817, 0.01672973, -0.009999905, 0.003832534, -0.01793312, -0.0004868332, 0.027573315, 0.001756033, -0.012112376, -0.009718678, 0.0025473924, -0.027547155, -0.019084187, 0.010693162, 0.025558947, -0.02168717, -0.0068802484, -0.010869746, -0.028698223, -0.0051634563, -0.012131997, -0.014963887, 0.022210384, 0.01510777, -0.0026504, -0.013577373, 0.0058599836, 0.011281776, -0.0009393305, -0.00204053, 0.030110897, -0.029326078, 0.006491109, -0.01671665, 0.0006049648, -0.024342475, -0.008325624, 0.03722659, -0.007710849, -0.0055656764, -0.02043146, -0.015317055, -0.015212413, 0.002815539, 0.022262705, 0.00818828, 0.021778734, -0.0037409717, -0.02485261, 0.0033779927, 0.013217663, -0.0059319255, -0.018940303, 0.02409395, 0.015761785, -0.009672897, 0.011301396, -0.011582624, 0.0029725027, -0.015343216, -0.00735114, -0.075761214, 0.016821291, 0.0028040938, 0.0017233322, 0.01595799, -0.0054741143, -0.007096074, -0.011641486, -0.003554577, 0.009829861, -0.037828285, 0.024983412, 0.003793293, -0.010895907, -0.011916172, -0.017893879, 0.029640006, 0.0027452323, 0.004977062, 0.0138913, 0.0132830655, 0.010725862, 0.014205228, -0.003839074, 0.020470701, 0.0048626093, -0.010967849, 0.035343025, -0.004568302, -0.007665068, 0.0040091183, -0.02367538, -0.006821387, 0.012112376, -0.0012475356, -0.02041838, -0.030869557, -0.004865879, 0.036127847, 0.019528918, 0.00087147637, 0.0016366751, -0.006072539, -0.012380524, -0.016886694, 0.0014224849, 0.0058632535, 0.0053138803, 0.024525601, -0.008227522, 0.016167276, 0.021373244, -0.019855926, -0.011602244, -0.012223559, 0.009116983, 0.00448001, 0.0027027212, 0.0112294555, -0.025048813, 0.005958086, 0.005578757, 0.012040435, -0.019528918, -0.008096718, -0.023439934, 0.00047497914, 0.0073315194, 0.025061894, -0.016455043, 0.003992768, 0.002038895, -0.0003484679, 0.004444039, -0.014846164, 0.0018263398, 0.017305264, -0.0047154557, -0.006729825, 0.011288317, -0.009764459, -0.03220375, -0.015369376, 0.009594415, 0.031078842, 0.020967754, -0.007802411, 0.022354268, -0.010778184, 0.01833861, 0.004581382, 0.0072399573, 0.010673542, -0.012112376, -0.023073684, 0.0066448026, -0.027887244, 0.0063504954, 0.012956058, 0.032151427, -0.018103164, 0.0048855, -0.018286288, -0.036938824, -0.012354363, 0.020039052, 0.004921471, -0.03790677, 0.0212686, 0.02982313, 0.015434778, 0.0041039507, -0.016245758, 0.012171238, -0.006415897, 0.0072464976, -0.0024362097, -0.025218857, -0.021399405, 0.036860343, 0.0056572384, 0.017004417, 0.03432276, -0.013825899, 0.028724384, 0.008528369, 0.018652538, -0.02443404, -0.025637427, 0.006497649, -0.015447859, 0.01917575, -0.016520444, -0.008678793, -0.021072397, 0.015840268, -0.006324335, 0.025925195, -0.03594472, 0.0384823, 0.01308032, 0.0054217926, 0.00448328, -0.027207067, -0.016847452, 0.0036003583, 0.01061468, -0.019816685, -0.004659864, 0.023387613, -0.005461034, 0.004326316, 0.0037278912, -0.007540805, 0.00860031, 0.0015524705, 0.020039052, -0.0028367946, 0.0049509015, 0.009162764, 0.009705598, 0.013982862, 0.004852799, 0.0061869915, -0.0083910255, 0.012975678, -0.034558207, -0.029064473, -0.03058179, -0.019450437, 0.01062122, -0.014179068, -0.010012985, 0.007874353, -0.014126746, -0.009731758, -0.03398267, -0.000115883464, -0.0029725027, -0.024290156, 0.012864495, -0.00937859, -0.035264544, 0.0027959184, 0.012982218, -0.012609429, 0.0065270797, 0.010712783])
# Loop to generate embeddings and calculate dot products
for _ in range(3):
# Generate an embedding
response = client.embeddings.create(
model="text-embedding-ada-002",
input="historical heist",
encoding_format="float",
)
# Extract the embedding vector from the response
vector = response.data[0].embedding
# Calculate the dot product between the generated embedding and the vector from MongoDB
dot_product = np.dot(vector, vector_from_mongodb)
# Print the first 5 elements of the embedding
print(f"\nEmbedding vector (first 5 elements): {vector[:5]}")
# Print the dot product value
print(f"Dot product: {dot_product}\n")
print("------------------------------------")
After I ran it, I got the following output:
Embedding vector (first 5 elements): [-0.02273331, -0.027646419, 0.0030520316, -0.033631727, -0.026207773]
Dot product: 0.9853400963855683
------------------------------------
Embedding vector (first 5 elements): [-0.02273331, -0.027646419, 0.0030520316, -0.033631727, -0.026207773]
Dot product: 0.9853400963855683
------------------------------------
Embedding vector (first 5 elements): [-0.02273331, -0.027646419, 0.0030520316, -0.033631727, -0.026207773]
Dot product: 0.9853400963855683
------------------------------------
As you can see:
The discussion on the forum went on about whether this small difference between embedding vectors for the identical input is problematic or not. In other words, should you care about the difference or not? I don't think you should, as long as you get meaningful solutions using them.
Moreover, I would care even less if you got identical embedding vectors for a given API endpoint, model, and input. Don't compare an embedding vector you get when using the OpenAI API endpoint and MongoDB API endpoint. Compare an embedding vector you get when using the OpenAI API endpoint multiple times. Compare an embedding vector you get when using the MongoDB API multiple times. The difference in embedding vectors between the OpenAI API endpoint and MongoDB API endpoint might be of a technical nature, as Curt Kennedy explained on the official OpenAI forum.