La recherche à l’ère de l’IA
A presentation at FinistDevs in in Brest, France by David Pilato
Search a new era David Pilato | @dadoonet
Elasticsearch You Know, for Search
These are not the droids you are looking for.
GET /_analyze { “char_filter”: [ “html_strip” ], “tokenizer”: “standard”, “filter”: [ “lowercase”, “stop”, “snowball” ], “text”: “These are <em>not</em> the droids you are looking for.” }
“char_filter”: “html_strip” These are <em>not</em> the droids you are looking for. These are not the droids you are looking for.
“tokenizer”: “standard” These are not the droids you are looking for. These are not the droids you are looking for
“filter”: “lowercase” These are not the droids you are looking for these are not the droids you are looking for
“filter”: “stop” These are not the droids you are looking for these are not the droids you are looking for droids you looking
“filter”: “snowball” These are not the droids you are looking for these are not the droids you are looking for droids you droid you looking look
These are <em>not</em> the droids you are looking for. { “tokens”: [{ “token”: “droid”, “start_offset”: 27, “end_offset”: 33, “type”: “<ALPHANUM>”, “position”: 4 },{ “token”: “you”, “start_offset”: 34, “end_offset”: 37, “type”: “<ALPHANUM>”, “position”: 5 }, { “token”: “look”, “start_offset”: 42, “end_offset”: 49, “type”: “<ALPHANUM>”, “position”: 7 }]}
Semantic search ≠ Literal matches
Elasticsearch You Know, for Vector Search
What is a Vector ?
Example: 1-dimensional vector Character Vector [ 1 ] ] Realistic
[ Embeddings represent your data Cartoon 1
represent different data aspects Human Character Vector [ 1, 1 Realistic Cartoon ] ] Machine
[ Multiple dimensions 1, 0
is grouped together Human Character Vector [ 1.0, 1.0 1.0, 0.0 Realistic Cartoon [ 1.0, 0.8 1.0, 1.0 [ 1.0, 1.0 ] ] ] ] ]
Machine
[ [ Similar data
Vector search ranks objects by similarity (~relevance) to the query Human Rank Query 1 Realistic Cartoon 2 3 4 5 Machine Result
How do you index vectors ?
Architecture of Vector Search
dense_vector field type PUT ecommerce { “mappings”: { “properties”: { “description”: { “type”: “text” } “desc_embedding”: { “type”: “dense_vector” } } } }
Data Ingestion and Embedding Generation POST /ecommerce/_doc { “_id”:”product-1234”, “product_name”:”Summer Dress”, “description”:”Our best-selling…”, “Price”: 118, “color”:”blue”, “fabric”:”cotton”, “fabric”:”cotton” “desc_embedding”:[0.452,0.3242,…], } “desc_embedding”:[0.452,0.3242,…] } “img_embedding”:[0.012,0.0,…] } Source data POST /ecommerce/_doc
Co m m er ci With Elastic ML al { } Source data { } “_id”:”product-1234”, “product_name”:”Summer Dress”, “description”:”Our best-selling…”, “Price”: 118, “color”:”blue”, “fabric”:”cotton”, POST /ecommerce/_doc “_id”:”product-1234”, “product_name”:”Summer Dress”, “description”:”Our best-selling…”, “Price”: 118, “color”:”blue”, “fabric”:”cotton”, “desc_embedding”:[0.452,0.3242,…]
Eland Imports PyTorch Models Co m m er ci al $ eland_import_hub_model —url https://cluster_URL —hubmodel-id BERT-MiniLM-L6 —tasktype text_embedding —start BERT-MiniLM-L6 Select the appropriate model Load it Manage models
Elastic’s range of supported NLP models Co m m er ci ● Fill mask model Mask some of the words in a sentence and predict words that replace masks ● Named entity recognition model NLP method that extracts information from text ● Text embedding model Represent individual words as numerical vectors in a predefined vector space ● Text classification model Assign a set of predefined categories to open-ended text ● Question answering model Model that can answer questions given some or no context ● Zero-shot text classification model Model trained on a set of labeled examples, that is able to classify previously unseen examples Full list at: ela.st/nlp-supported-models al
How do you search vectors ?
Architecture of Vector Search
knn query GET /ecommerce/_search { “query” : { “bool”: { “must”: [{ “knn”: { “field”: “desc_embbeding”, “query_vector”: [0.123, 0.244,…] } }], “filter”: { “term”: { “department”: “women” } } } } }, “size”: 10
knn query (with Elastic ML Co m m er ci al GET /ecommerce/_search { “query” : { “bool”: { “must”: [{ “knn”: { “field”: “desc_embbeding”, “query_vector_builder”: { “text_embedding”: { “model_text”: “summer clothes”, “model_id”: <text-embedding-model> } } } }], “filter”: { “term”: { “department”: “women” } } } }, “size”: 10 } ) Transformer model
semantic_text field type PUT /_inference/text_embedding/e5-small-multilingual { “service”: “elasticsearch”, “service_settings”: { “num_allocations”: 1, “num_threads”: 1, “model_id”: “.multilingual-e5-small_linux-x86_64” } } POST ecommerce/_doc { “description”: “Our best-selling…” } ne w in 8. 15 PUT ecommerce { “mappings”: { “properties”: { “description”: { “type”: “text”, “copy_to”: [ “desc_embedding” ] } “desc_embedding”: { “type”: “semantic_text”, “inference_id”: “e5-small-multilingual” } } } } GET ecommerce/_search { “query”: { “semantic”: { “field”: “desc_embedding” “query” : “I’m looking for a red dress for a DJ party” }}}
Architecture of Vector Search
Choice of Embedding Model Start with Off-the Shelf Models Extend to Higher Relevance ●Text data: Hugging Face (like Microsoft’s E5 ●Apply hybrid scoring ) ●Images: OpenAI’s CLIP ●Bring Your Own Model: requires expertise + labeled data
Problem training vs actual use-case
But how does it really work?
Similarity Human q cos(θ) = d1 d2 Realistic θ q⃗ × d ⃗ | q⃗ | × | d |⃗ _score = 1 + cos(θ) 2
Similarity: cosine (cosine) θ Similar vectors θ close to 0 cos(θ) close to 1 1+1 _score = =1 2 θ Orthogonal vectors θ close to 90° cos(θ) close to 0 1+0 _score = = 0.5 2 θ Opposite vectors θ close to 180° cos(θ) close to -1 1−1 _score = =0 2
Similarity: Dot Product (dot_product or max_inner_product) q⃗ × d ⃗ = | q⃗ | × cos(θ) × | d |⃗ q d θ | q⃗ | × co s (θ ) 1 + dot_ product(q, d) scorefloat = 2 0.5 + dot product(q, d) _scorebyte = 32768 × dims
Similarity: Euclidean distance (l2_norm) y 2 n i (x ∑ 1 i= − y i) q l2_normq,d = y1 d x1 y2 x2 n ∑ i=1 (xi − yi) 1 _score = 1 + (l2_normq,d )2 x 2
Brute Force
Hierarchical Navigable Small Worlds (HNSW One popular approach HNSW: a layered approach that simplifies access to the nearest neighbor Tiered: from coarse to fine approximation over a few steps Balance: Bartering a little accuracy for a lot of scalability ) Speed: Excellent query latency on large scale indices
Scaling Vector Search Vector search Best practices
- Needs lots of memory
- Avoid searches during indexing
- Indexing is slower
- Exclude vectors from _source
- Merging is slow
- Reduce vector dimensionality 4. Use byte rather than float
- Continuous improvements in Lucene + Elasticsearch
Scalar Quantization Ela s 8.14 ticsea def rch aul t float32 Recall: High Precision: High Rescore: Likely Not Needed + Full RAM Required int8 int4 bit Recall: Good Precision: Good Oversampling: Moderate Recall: Low Precision: Low Oversampling: Needed Recall: Bad Precision: Bad Oversampling: Needed Rescore: Reasonable Rescore: may be slower Rescore: Expensive and Limiting 4X RAM Savings 8X RAM Savings 32X RAM Savings
Scalar Quantization -> Better Binary Quantization float32 int8 int4 bit : BBQ 32X RAM savings. Faster & more accurate than Product Quantization BBQ*
Memory required 100M vectors? Only 12GB!?! One single node.
Benchmarketing
https://djdadoo.pilato.fr/
https://github.com/dadoonet/music-search/
Elasticsearch You Know, for Hybrid Search
Hybrid scoring Term-based score Linear Combination manual boosting Vector similarity score Combine
GET ecommerce/_search { “query” : { “bool” : { “must” : [{ “match”: { “description”: { “query”: “summer clothes”, “boost”: 0.1 } } },{ “knn”: { “field”: “desc_embbeding”, “query_vector”: [0.123, 0.244,…], “boost”: 2.0, “filter”: { “term”: { “department”: “women” } } } }], “filter” : { “range” : { “price”: { “lte”: 30 } } } } } } summer clothes pre-filter post-filter
PUT starwars { “mappings”: { “properties”: { “text.tokens”: { “type”: “sparse_vector” } } } “These are not the droids you are looking for.”, } “Obi-Wan never told you what happened to your father.” GET starwars/_search { “query”:{ “sparse_vector”: { “field”: “text.tokens”, “query_vector”: { “lucas”: 0.50047517, “ship”: 0.29860738, “dragon”: 0.5300422, “quest”: 0.5974301, … } } } }
ELSER Elastic Learned Sparse EncodER sparse_vector Not BM25 or (dense) vector Sparse vector like BM25 Stored as inverted index Co m m er ci al
Hybrid ranking Dense vector score Reciprocal Rank Fusion (RRF blend multiple ranking methods Combine ) Term-based score Sparse vector score
Reciprocal Rank Fusion (RRF D set of docs R set of rankings as permutation on 1..|D| k - typically set to 60 by default Dense Vector r(d) k+r(d) A 1 1 B 0.7 C D Score r(d) k+r(d) 61 C 1,341 1 61 2 62 A 739 2 62 0.5 3 63 F 732 3 63 0.2 4 64 G 192 4 64 0.01
=
= +
E Doc 5 65 H 183 5 65 ) Score
Doc BM25 Doc RRF Score A 1/61 1/62 0,0325 C 1/63 1/61 0,0323 B 1/62 0,0161 F 1/63 0,0159 D 1/64 0,0156
GET index/_search { “retriever”: { “rrf”: { “retrievers”: [{ “standard” { “query”: { “match”: {…} } } },{ “standard” { “query”: { “sparse_vector”: {…} } } },{ “knn”: { … } } ] } } } Hybrid Ranking BM25f + Sparse Vector + Dense Vector Co m m er ci al
ChatGPT Elastic and LLM
Gen AI Search engines
LLM opportunities and limits your question one answer your question GAI / LLM : public internet data
Retrieval Augmented Generation your question the right answer your question + context window GAI / LLM public internet data your business data documents images audio
Demo Elastic Playground
Elasticsearch You Know, for Semantic Search
Search a new era David Pilato | @dadoonet
La recherche ne se contente plus de l’approche maintenant traditionnelle basée sur la fréquence des termes (TF/IDF ou BM25) mais plus sur la tendance actuelle du machine learning où les nouveaux modèles ont ouvert une nouvelle dimension pour la recherche.
Cette conférence donne un aperçu de :
- La recherche “Classique” et ses limitations
- Qu’est qu’un modèle de machine learning et comment vous pouvez l’utiliser
- Comment utiliser la recherche vectorielle ou la recherche hybride dans Elasticsearch
- Comment ChatGPT d’OpenAI ou les “large language models” (LLMs) similaires viennent jouer naturellement avec Elastic
La démo principale montre comment générer des embeddings à partir de musiques puis comment trouver la musique qui s’approche le plus d’une musique que nous fredonnons 🎶🎸🎻.
Resources
The following resources were mentioned during the presentation or are useful additional information.
Buzz and feedback
Here’s what was said about this presentation on social media.
