id stringlengths 36 36 | url stringclasses 504
values | title stringclasses 475
values | section stringlengths 0 148 | content_type stringclasses 3
values | has_code bool 2
classes | text stringlengths 147 22.5k | vector listlengths 384 384 |
|---|---|---|---|---|---|---|---|
c5960210-25f6-5e85-8167-5fa539431353 | /documentation/embeddings/cohere/ | Cohere | Qdrant Cloud Inference | documentation | true | Cohere > Qdrant Cloud Inference
On Qdrant Cloud, Qdrant can call Cohere for you. Instead of a vector, send a `Document` or an `Image` with the model name prefixed with `cohere/`. Pass your Cohere API key in the `cohere-api-key` request header or in the `options` object. Qdrant does not store the key, so send it with e... | [
0.06337107717990875,
-0.03376389667391777,
-0.012732903473079205,
-0.05534694716334343,
0.04763276129961014,
-0.0462915375828743,
0.08592049777507782,
0.06534139066934586,
0.05110174044966698,
-0.020990312099456787,
-0.012394881807267666,
0.014286343939602375,
0.04308692738413811,
-0.03643... |
c59aa7dd-e7d6-5e4b-b5fd-ee6d793b422b | /documentation/manage-data/points/ | Points | Retrieve Points | documentation | true | Points > Retrieve Points
There is a method for retrieving points by their ids.
REST API ([Schema](https://api.qdrant.tech/api-reference/points/get-points)):
```python
client.retrieve(
collection_name="{collection_name}",
ids=[0, 3, 100],
)
```
This method has additional parameters `with_vectors` and `with_p... | [
0.03238925337791443,
-0.045172031968832016,
0.005365281365811825,
-0.05586690083146095,
0.04215031862258911,
-0.009617788717150688,
0.030621081590652466,
0.0467522107064724,
0.0716380700469017,
-0.006235166918486357,
0.041728295385837555,
0.037231262773275375,
0.011452145874500275,
-0.0165... |
c5f1c584-857f-5de4-a0dd-f9d5efa677be | /documentation/tutorials-search-engineering/using-relevance-feedback/ | Relevance Feedback Retrieval in Qdrant | Adding Relevance Feedback Query to a Search Pipeline > 2. Getting Feedback on Initial Retrieval | documentation | true | Relevance Feedback Retrieval in Qdrant > Adding Relevance Feedback Query to a Search Pipeline > 2. Getting Feedback on Initial Retrieval
Now we get feedback from our `mxbai-embed-large-v1` feedback model on the top 3 results for the query *"recommendations API how to use"*.
The feedback model rescores them according ... | [
0.03307340294122696,
-0.03588993102312088,
-0.016200190410017967,
-0.08288207650184631,
0.0640493705868721,
-0.004062462132424116,
0.0564175508916378,
0.05100293830037117,
0.037204645574092865,
-0.022619329392910004,
0.028973761945962906,
0.0504063256084919,
0.06863046437501907,
-0.0563181... |
c5f43877-6f7c-5b20-b32d-dbb357d7e767 | /documentation/tutorials-develop/code-search/ | Semantic Search for Code | Data preparation | documentation | false | Semantic Search for Code > Data preparation
Chunking the application sources into smaller parts is a non-trivial task. In
general, functions, class methods, structs, enums, and all the other language-specific
constructs are good candidates for chunks. They are big enough to
contain some meaningful information, but sm... | [
0.00941197108477354,
-0.034333642572164536,
-0.04751301929354668,
-0.05418112128973007,
0.049837879836559296,
-0.06095869094133377,
0.0681150034070015,
0.013350386172533035,
0.05382901802659035,
-0.010098767466843128,
0.02586214803159237,
0.004324928391724825,
0.07415693998336792,
-0.03579... |
c5f93eb6-659e-5a01-a9d0-fa59143b1763 | /course/multi-vector-search/module-3/muvera/ | MUVERA | Applying MUVERA Multi-Stage Retrieval > Setting Up MUVERA with ColModernVBERT | course | true | MUVERA > Applying MUVERA Multi-Stage Retrieval > Setting Up MUVERA with ColModernVBERT
FastEmbed 0.7.2+ includes MUVERA as a postprocessing technique that transforms variable-length multi-vector sequences into fixed-dimensional single vectors. You can apply MUVERA to any late interaction model, including **ColModernVB... | [
0.0720297172665596,
-0.031153496354818344,
-0.03827741742134094,
-0.04312124475836754,
0.06211290508508682,
-0.01518571749329567,
0.059857144951820374,
-0.01461366843432188,
0.05060767009854317,
0.0049368636682629585,
0.008648151531815529,
0.018862176686525345,
0.05030038580298424,
-0.0244... |
c6010c89-a83e-5bd6-8f2f-43bce5ce3951 | /course/essentials/day-0/pitstop-project/ | Project: Building Your First Vector Search System | Success Criteria | course | false | Project: Building Your First Vector Search System > Success Criteria
You’ll know you’ve succeeded when:
Collection created without errors
Search returns results ranked by similarity score
Filtered search works and returns appropriate subsets
You can explain why certain items are more similar than others | [
0.05986963212490082,
-0.07382901757955551,
-0.03086930513381958,
-0.013591239228844643,
0.07822294533252716,
-0.040967296808958054,
0.06882085651159286,
0.021829284727573395,
0.058066584169864655,
-0.04200231283903122,
0.018275098875164986,
0.052892640233039856,
0.08007089048624039,
-0.026... |
c6041e26-1d99-559a-b5aa-110f54c6c1f7 | /articles/memory-consumption/ | Minimal RAM to Serve 1M Vectors | article | false | Minimal RAM to Serve 1M Vectors
When it comes to measuring the memory consumption of our processes, we often rely on tools such as `htop` to give us an indication of how much RAM is being used. However, this method can be misleading and doesn't always accurately reflect the true memory usage of a process.
There are m... | [
-0.0019867797382175922,
-0.06856079399585724,
-0.027398237958550453,
-0.03308168426156044,
0.0715131163597107,
-0.013353388756513596,
0.048580948263406754,
0.035174522548913956,
0.126548171043396,
-0.02451981231570244,
0.0065820664167404175,
0.017737699672579765,
0.058668944984674454,
-0.0... | |
c60e6da5-d3cc-5548-8d3f-023f46474c05 | /course/beginners/module-5/analyst-queries/ | Analyst Queries | course | false | Analyst Queries
Module 5
One collection, three named vectors. Two ways to ask here, plus the centroid query from [Clustering Risk Signals](/course/beginners/module-5/clustering-risk-signals/). | [
0.058089278638362885,
-0.01631196215748787,
-0.04792853444814682,
-0.06630963087081909,
0.05423327162861824,
-0.032520752400159836,
0.024479834362864494,
0.07602182775735855,
0.09296973049640656,
-0.023548997938632965,
0.002668805420398712,
0.03975536674261093,
0.04423054680228233,
-0.0288... | |
c627cf20-4a36-5b8c-a19c-7a0c513ffe68 | /documentation/manage-data/indexing/ | Indexing | Payload Index > Parameterized Index > On-Disk Payload Index | documentation | true | Indexing > Payload Index > Parameterized Index > On-Disk Payload Index
*Available as of v1.11.0*
Payload indexes are always persisted to disk. By default, they are also loaded into the `pinned` [memory tier](/documentation/ops-configuration/memory-tiers/). This keeps the index on the heap, so payload values can be ac... | [
0.01955161616206169,
-0.034727200865745544,
-0.027808723971247673,
-0.02270839549601078,
0.06842838227748871,
-0.05564070865511894,
0.0102504538372159,
0.027495378628373146,
0.09005054086446762,
-0.011026032269001007,
0.008478250354528427,
0.011799167841672897,
0.027487333863973618,
-0.012... |
c62ec53d-d3b3-598e-86aa-7596aa74b0df | /course/essentials/day-1/embedding-models/ | Points, Vectors and Payloads | Vector Types in Qdrant > Multivectors | course | true | Points, Vectors and Payloads > Vector Types in Qdrant > Multivectors
While most models produce one vector per input, advanced techniques like late-interaction models (e.g., ColBERT) generate a set of vectors, often one for each token. Qdrant's multivector lets you store this whole matrix on a single point.
 with its metadata.
- Without a predefined similarity threshold for our semantic cache, we'll always get something out of the collection as long as our cache is not empty.
- With a predefined on... | [
0.049122802913188934,
-0.0704626739025116,
-0.05859598144888878,
-0.02845938876271248,
0.07592358440160751,
-0.031459957361221313,
0.02682151272892952,
-0.0014661939349025488,
0.04202746972441673,
-0.06210574507713318,
0.016975149512290955,
0.02313668467104435,
0.06930842995643616,
-0.0171... |
c66d4f03-8a33-5179-bb75-02b9ccc5bb13 | /documentation/tutorials-operations/blue-green-deployment/ | Blue-Green Cluster Deployment | documentation | false | Blue-Green Cluster Deployment
| Time: 30 min | Level: Intermediate |
| :---- | :---- |
A blue-green deployment runs two clusters in parallel: one live cluster serving production traffic (blue), and one staging cluster (green) where you prepare and validate changes. Once the green cluster is ready, you switch applicat... | [
0.003645658493041992,
-0.021290408447384834,
-0.04706481099128723,
-0.04460017383098602,
0.08990763127803802,
-0.010019355453550816,
0.023240653797984123,
0.03907901048660278,
0.06774917244911194,
-0.020720841363072395,
0.013207875192165375,
0.012258066795766354,
-0.01551405992358923,
-0.0... | |
c67a0c5c-20e0-5c89-8489-2f62ff61f1ed | /documentation/private-cloud/api-reference/ | API Reference | qdrant.io/v1 > Resource Types > QdrantClusterRestore | documentation | false | API Reference > qdrant.io/v1 > Resource Types > QdrantClusterRestore
QdrantClusterRestore is the Schema for the qdrantclusterrestores API
_Appears in:_
- [QdrantClusterRestoreList](#qdrantclusterrestorelist)
| Field | Description | Default | Validation |
| --- | --- | --- | --- |
| `apiVersion` _string_ | `qdrant.io... | [
0.05911583453416824,
-0.0383131243288517,
-0.020189259201288223,
-0.058462485671043396,
0.07159540802240372,
-0.03597844019532204,
0.045303381979465485,
0.030929643660783768,
0.042984772473573685,
-0.06313032656908035,
0.04534079506993294,
0.012791257351636887,
0.051155202090740204,
0.0061... |
c67b2031-defa-5510-90ee-96fd6ae01c74 | /course/multi-vector-search/module-2/how-colpali-works/ | How ColPali Models Work | Patch-Based Image Processing | course | false | How ColPali Models Work > Patch-Based Image Processing
Vision transformers, the foundation of ColPali, don't process entire images at once. Instead, they divide images into a grid of fixed-size patches - think of it like a checkerboard overlaid on your document.
The ColPali pipeline starts with a document. It is usua... | [
0.020336061716079712,
-0.037315599620342255,
-0.06886164098978043,
-0.08075990527868271,
0.07344160974025726,
0.005533057264983654,
0.03288567066192627,
0.04232580214738846,
0.06588432192802429,
-0.018757455050945282,
0.031816039234399796,
-0.01783035509288311,
0.0590074360370636,
-0.03334... |
c6993043-3e64-524a-935a-cb0ee1355808 | /documentation/tutorials-build-essentials/video-anomaly-edge-part-3/ | Video Anomaly Detection Part 3: Scoring, Governance, and Deployment | Getting Started | documentation | false | Video Anomaly Detection Part 3: Scoring, Governance, and Deployment > Getting Started
Before continuing, make sure you have completed Parts 1 and 2 and have the following running:
- Docker stack up (`docker compose up`)
- Qdrant Cloud collection populated with baseline embeddings
- At least one edge device registered... | [
0.049979548901319504,
-0.044680021703243256,
-0.05296996235847473,
-0.036672402173280716,
0.050029728561639786,
-0.034750375896692276,
0.03657762333750725,
0.031879279762506485,
0.08687632530927658,
-0.04377865791320801,
0.0492326021194458,
0.02650303579866886,
0.005177680402994156,
-0.033... |
c6bddf6d-2c17-5d1e-9353-08bb6c841f4c | /documentation/tutorials-operations/secure-and-expose-hybrid-cloud/ | Securing and Exposing a Hybrid Cloud Cluster | Expose the Cluster Through an Ingress > Configure the backend connection | documentation | true | Securing and Exposing a Hybrid Cloud Cluster > Expose the Cluster Through an Ingress > Configure the backend connection
Traefik needs two pieces of configuration to reach an HTTPS backend: which protocol scheme to use, and which `ServersTransport` to trust it with. Both go **on the Qdrant Service, not the Ingress**, a... | [
0.07062897831201553,
-0.03692369535565376,
-0.07425964623689651,
-0.010389287956058979,
0.04698863625526428,
-0.03608423098921776,
0.06721635162830353,
0.048924412578344345,
0.007784695830196142,
-0.028960904106497765,
0.019275417551398277,
0.002374705858528614,
0.05617409572005272,
-0.023... |
c70ad672-c16d-5a3a-ad9b-3cd14c5cb255 | /documentation/tutorials-build-essentials/semantic-cache/ | Semantic Caching for RAG | Keep Cached Answers Fresh | documentation | true | Semantic Caching for RAG > Keep Cached Answers Fresh
This step is optional, consider it if in your use case a cached answer goes stale, when it gets old, or when the document behind it changes.
Expire old answers and invalidate answers when a document changes
- **It gets old.** The lookup already skips answers past ... | [
0.025748373940587044,
-0.03103344701230526,
-0.026205727830529213,
-0.0512247197329998,
0.0782385766506195,
-0.015305520966649055,
0.007766843773424625,
-0.03157191351056099,
0.05118434876203537,
-0.05183560401201248,
-0.003033512271940708,
0.018954414874315262,
0.08474278450012207,
-0.020... |
c719450b-0d23-556d-83fa-633576107b77 | /course/essentials/day-3/sparse-retrieval-demo/ | Demo: Keyword Search with Sparse Vectors | Sparse Neural Retrieval > SPLADE: Sparse Lexical and Expansion Model > Create & Insert SPLADE++ Sparse Vectors with FastEmbed | course | true | Demo: Keyword Search with Sparse Vectors > Sparse Neural Retrieval > SPLADE: Sparse Lexical and Expansion Model > Create & Insert SPLADE++ Sparse Vectors with FastEmbed
The FastEmbed library provides **SPLADE++**; one of the latest models in the SPLADE family.
> **Update:** Since the release of [Qdrant Cloud Inferenc... | [
0.07689864188432693,
-0.03322818502783775,
-0.03081245720386505,
-0.05555162951350212,
0.08275357633829117,
-0.04391859471797943,
0.08019650727510452,
0.011447297409176826,
0.05712444335222244,
-0.02161170355975628,
0.026862051337957382,
0.03135208413004875,
0.04268871992826462,
-0.0262421... |
c728cde6-badd-55d9-9043-3defebee0430 | /course/essentials/day-3/hybrid-search-demo/ | Demo: Implementing a Hybrid Search System | Step 6: Distribution-Based Score Fusion (DBSF) | course | true | Demo: Implementing a Hybrid Search System > Step 6: Distribution-Based Score Fusion (DBSF)
RRF isn't the only fusion method available. DBSF normalizes scores from each query and sums them across different retrievers:
```python
def dbsf_search(query: str) -> list[models.ScoredPoint]:
response = client.query_points... | [
0.02938949689269066,
-0.020546725019812584,
-0.06471924483776093,
-0.06806638091802597,
0.040713902562856674,
-0.022553252056241035,
0.06757079064846039,
0.02384447120130062,
0.06966975331306458,
-0.024491295218467712,
0.04476412758231163,
-0.001785856089554727,
0.047273293137550354,
0.010... |
c72a00e5-9f10-5ae8-92e4-101b1a8d685d | /documentation/frameworks/mastra/ | Mastra | Methods > createIndex() | documentation | false | Mastra > Methods > createIndex()
| Name | Type | Description | Default Value |
|------------|------------------------------------------|-------------------------------------------------|--------------|
| `indexName` | `string` ... | [
0.05013890191912651,
-0.03429778665304184,
-0.03267049044370651,
-0.02813141420483589,
0.032912787050008774,
-0.02798602543771267,
0.0472126342356205,
0.040631309151649475,
0.062278229743242264,
-0.029463201761245728,
0.0482616052031517,
0.042587582021951675,
0.052057284861803055,
0.004646... |
c74e017f-bcab-514d-9beb-150e75781ef7 | /documentation/faq/qdrant-fundamentals/ | Qdrant Fundamentals | Collections > What's the recommended batch size for uploading vectors? | documentation | false | Qdrant Fundamentals > Collections > What's the recommended batch size for uploading vectors?
There is no universal recommended batch size. The optimum depends on your vector dimensionality, payload size, cluster configuration, and available memory. You should benchmark different batch sizes against your own setup to f... | [
-0.017887623980641365,
-0.03082067333161831,
-0.04249320924282074,
-0.05475945025682449,
0.04645794630050659,
-0.05377096310257912,
0.050674375146627426,
0.05662321299314499,
0.10734591633081436,
-0.04962196573615074,
0.03570409491658211,
0.03661646321415901,
0.02470400370657444,
-0.045125... |
c76f07cd-475f-5fb0-aae2-9d611fe6023c | /articles/late-interaction-models/ | Late Interaction Retrieval with Dense Token Embeddings | Try the Experiment Yourself | article | false | Late Interaction Retrieval with Dense Token Embeddings > Try the Experiment Yourself
In a real-world scenario, you might take it a step further by first calculating the token embeddings and then performing pooling to obtain the single vector representation. This approach allows you to complete everything in a single p... | [
0.061438217759132385,
-0.04140973463654518,
-0.014085977338254452,
-0.05449448525905609,
0.08462561666965485,
-0.06395230442285538,
0.051317714154720306,
0.06090252473950386,
0.06066344305872917,
-0.012840754352509975,
0.025086084380745888,
0.012420251965522766,
0.03766463324427605,
-0.038... |
c789686b-86a5-57b3-ba46-c5c5a914573a | /documentation/tutorials-search-engineering/sparse-embeddings-ecommerce-part-5/ | Fine-Tuning Sparse Embeddings for E-Commerce Search | Part 5: From Research to Product | What Changed From the Series Code | documentation | true | Fine-Tuning Sparse Embeddings for E-Commerce Search | Part 5: From Research to Product > What Changed From the Series Code
The pipeline from Parts 2-4 is the same underneath. The toolkit wraps it with:
**Automatic data handling.** Pass a CSV, JSON, JSONL, or HuggingFace dataset path. The loader auto-detects columns (... | [
0.03231661394238472,
-0.08751777559518814,
-0.07928619533777237,
-0.03531046211719513,
0.07460004836320877,
-0.02952464669942856,
0.04732989892363548,
0.004557115025818348,
0.04799873009324074,
-0.021241644397377968,
0.020660506561398506,
0.046120285987854004,
0.04186021536588669,
-0.02376... |
c78b23f0-946e-51db-a178-238b43fd82ce | /documentation/ops-optimization/read-write-contention/ | Read-Write Contention | Read More | documentation | false | Read-Write Contention > Read More
- [Optimizer](/documentation/ops-optimization/optimizer/) covers all optimizer settings referenced in this guide, including how to monitor deferred points.
- [Low-Latency Search](/documentation/search/low-latency-search/) covers delayed fan-outs and other techniques for reducing searc... | [
0.017278552055358887,
-0.03627467527985573,
-0.07839133590459824,
-0.017026128247380257,
0.06909022480249405,
-0.02854819782078266,
0.05031860992312431,
0.03112190030515194,
0.06813264638185501,
-0.023038377985358238,
0.04174768552184105,
0.003277079900726676,
0.017271410673856735,
-0.0284... |
c7a348f4-5c65-5146-b315-7b1188403bd9 | /documentation/search/text-search/text-filtering/ | Text Filtering | Full-Text Filtering > Text Processing | documentation | false | Text Filtering > Full-Text Filtering > Text Processing
To enable efficient full-text filtering, Qdrant processes text strings by breaking them down into individual tokens (words) and applying several normalization steps. This process ensures that searches are more flexible and can match variations of words. At query t... | [
0.047425586730241776,
-0.049288973212242126,
-0.031153766438364983,
-0.04657401889562607,
0.05862896889448166,
-0.003953021951019764,
0.04220075532793999,
0.02475176751613617,
0.06582172214984894,
-0.03108852542936802,
0.03613845631480217,
0.019432948902249336,
0.054957062005996704,
-0.038... |
c7a514a8-dd44-563d-8ee0-3db9fb787ece | /documentation/deploy-intro/ | Deploy Qdrant | Managed Cloud | documentation | false | Deploy Qdrant > Managed Cloud
- [Managed Cloud](/documentation/cloud/index.md) — Run Qdrant as a managed service on AWS, GCP, or Azure with automatic scaling, backups, and zero-downtime upgrades.
- [Create a Cluster](/documentation/cloud/create-cluster/index.md) — Launch a free or standard cluster on your preferred cl... | [
0.017983317375183105,
-0.00896463729441166,
-0.059628959745168686,
-0.03843430429697037,
0.0765627920627594,
-0.04622562229633331,
0.06331571936607361,
0.04149686545133591,
0.04777687042951584,
-0.03865349292755127,
0.04072980582714081,
0.007696166634559631,
0.018998900428414345,
-0.035771... |
c7b46fb0-7d31-5d38-b4d0-de27db04c8b4 | /documentation/edge/edge-synchronization-guide/ | Synchronize with a Server | Synchronizing Qdrant Edge with a Server > 4. Periodically Update the Immutable Edge Shard | documentation | true | Synchronize with a Server > Synchronizing Qdrant Edge with a Server > 4. Periodically Update the Immutable Edge Shard
You can periodically update the immutable Edge Shard with changes from the server using partial snapshots, as described in [Update Qdrant Edge with Server-Side Changes](/documentation/edge/edge-data-sy... | [
0.04560060054063797,
-0.02810833230614662,
-0.037238795310258865,
-0.008311323821544647,
0.05634888634085655,
-0.053437963128089905,
0.08037663996219635,
0.020671825855970383,
0.09428456425666809,
-0.03622271865606308,
0.039265427738428116,
0.04055659472942352,
0.03589840605854988,
-0.0176... |
c7cbe839-3edb-5b05-b2b2-05d102f51aa1 | /articles/muvera-embeddings/ | MUVERA: Making Multivectors More Performant | How MUVERA Embeddings are created > Dimensionality reduction through Random Projection | article | false | MUVERA: Making Multivectors More Performant > How MUVERA Embeddings are created > Dimensionality reduction through Random Projection
The paper reports results from experiments with `k_sim` values of 4 and 5. Practically, this means 16 or 32 clusters,
respectively. If the original token vectors have a dimensionality of... | [
0.0573226734995842,
-0.03272614628076553,
-0.05216129496693611,
-0.07371910661458969,
0.07857669144868851,
-0.015498349443078041,
0.054039500653743744,
0.030744927003979683,
0.0901755839586258,
-0.021602408960461617,
0.029768628999590874,
0.027859244495630264,
0.05519065260887146,
-0.03966... |
c7e4eed0-74c3-5b54-bf36-7b97ea02ddc7 | /course/essentials/day-0/qdrant-cloud/ | Qdrant Setup | Connect from Python | course | true | Qdrant Setup > Connect from Python
Store credentials in an `.env` file at the root of your working directory or in colab:
```env
QDRANT_URL=https://YOUR-CLUSTER.cloud.qdrant.io:6333
QDRANT_API_KEY=YOUR_API_KEY
```
Load the credentials from the environment and create a Qdrant client:
```python
from qdrant_client imp... | [
0.028795963153243065,
-0.05470507964491844,
0.001906614052131772,
-0.03749675303697586,
0.05815558508038521,
-0.06806293874979019,
0.049277860671281815,
0.05603095889091492,
0.07040899246931076,
-0.06369247287511826,
0.010785747319459915,
-0.021060803905129433,
0.10675323754549026,
-0.0238... |
c8022298-9f8b-51a4-8f73-31abff6337b6 | /documentation/embeddings/gemini/ | Gemini | Reduce Vector Size | documentation | true | Gemini > Reduce Vector Size
Both models use Matryoshka Representation Learning, so `output_dimensionality` can shorten the vector with little loss in quality. Google recommends 768, 1536, or 3072. Set the collection's `size` to the same value, and pass the same `output_dimensionality` when you embed queries:
```pytho... | [
0.025214308872818947,
0.01629769802093506,
-0.03810146450996399,
-0.0533723458647728,
0.04000562056899071,
-0.01388662401586771,
0.03596428781747818,
0.030577735975384712,
0.08432947844266891,
-0.044990744441747665,
0.04292837902903557,
0.026892106980085373,
0.051878899335861206,
-0.024390... |
c80485c5-92e5-5dbf-85ef-fb8ef96e6f54 | /course/multi-vector-search/module-3/quantization-techniques/ | Vector Quantization Techniques | What's Next | course | false | Vector Quantization Techniques > What's Next
In this lesson, you learned how quantization dramatically reduces memory usage and improves brute force search performance for multi-vector representations. Key takeaways:
- ColModernVBERT's memory footprint: **512 KB per document** without quantization
- Scalar quantizatio... | [
0.04728211462497711,
-0.053509533405303955,
-0.03635613992810249,
-0.027192221954464912,
0.06725175678730011,
-0.006564488634467125,
0.044066719710826874,
0.023246388882398605,
0.11343137919902802,
-0.025315377861261368,
0.008732525631785393,
0.01473048236221075,
0.031504154205322266,
-0.0... |
c817409f-442f-520c-8e3a-b723743cf2cb | /documentation/ops-optimization/read-write-contention/ | Read-Write Contention | Step 1: Prevent Reads from Large Unindexed Segments | documentation | true | Read-Write Contention > Step 1: Prevent Reads from Large Unindexed Segments
*Requires Qdrant v1.17.1 or later. `prevent_unoptimized` is an experimental feature.*
Under heavy write load, the optimizer can fall behind. When that happens, searches need to perform a full scan over large unindexed segments, which increase... | [
0.028264060616493225,
-0.03787493705749512,
-0.07098197937011719,
-0.019972356036305428,
0.04156774654984474,
-0.027070146054029465,
0.046547453850507736,
0.0073004779405891895,
0.07377099990844727,
-0.021601809188723564,
0.0366443507373333,
0.016880791634321213,
0.0725642740726471,
-0.041... |
c82f4759-dd31-5846-ad1a-a88e148b9ce6 | /documentation/tutorials-operations/large-scale-search/ | Large-Scale Search | Memory Usage > Size of vector index | documentation | true | Large-Scale Search > Memory Usage > Size of vector index
Vector index is a bit more complicated, as it is not a simple matrix.
Internally, it is stored as a list of connections in a graph, and each connection is a 4-byte integer.
The number of connections is defined by the `M` parameter of the HNSW index, and in our... | [
0.03114798478782177,
-0.047352757304906845,
-0.055436328053474426,
-0.014682423323392868,
0.062180306762456894,
-0.02815508469939232,
0.0455172173678875,
0.045271217823028564,
0.0879712849855423,
-0.023716386407613754,
0.011355162598192692,
0.004900911822915077,
0.031128358095884323,
-0.02... |
c8654107-e7e5-5bad-8bf7-25a443882fdb | /documentation/migration-guidance/data-integrity/ | Data Integrity | 5. Point ID Verification | documentation | true | Data Integrity > 5. Point ID Verification
Check for duplicate or orphaned point IDs:
```py
# Scroll through all points and collect IDs
all_ids = []
next_offset = None
while True:
records, next_offset = client.scroll(
collection_name="your_collection",
limit=1000,
offset=next_offset,
... | [
0.07550296932458878,
-0.025047579780220985,
-0.033454205840826035,
-0.05730807036161423,
0.027542248368263245,
-0.016276873648166656,
0.06666859239339828,
0.05553939938545227,
0.05822975933551788,
0.004091536160558462,
0.018310977146029472,
-0.0030671798158437014,
0.03518650308251381,
0.01... |
c8732681-4062-5b06-a752-6d7882e5ea74 | /documentation/migration-recovery-options/ | Migration and Recovery | Comparison > Infrastructure Limits | documentation | false | Migration and Recovery > Comparison > Infrastructure Limits
Constraints on source system, target cloud, cluster shape, and Qdrant version compatibility.
| | Migration Tool | Snapshots | Cloud Backups |
|---|---|---|---|
| **Flexibility** | High | Moderate | Low |
| **Source system** | [Supported Sources](/documentati... | [
0.03598032519221306,
-0.052795425057411194,
-0.04649096354842186,
-0.04615681245923042,
0.04665959253907204,
0.0006681836093775928,
0.10761726647615433,
0.009431242942810059,
0.07223907858133316,
-0.021410366520285606,
0.03377939760684967,
-0.03130296990275383,
0.032703664153814316,
-0.028... |
c87334b0-59c0-5301-9827-5359ad0eeb93 | /articles/universal-io/ | Universal I/O in Qdrant | Mirroring Remote Object Storage to Local Disk | article | false | Universal I/O in Qdrant > Mirroring Remote Object Storage to Local Disk
Even with the async bridge, a network read is still far slower than a read from memory or local disk, but with a generic I/O interface already in place, we can build a tiered solution on top of it: a backend that caches data by mirroring remote fi... | [
0.06320440769195557,
-0.046076253056526184,
-0.0701342225074768,
-0.019912252202630043,
0.08038105070590973,
-0.010511573404073715,
0.04306148365139961,
0.010186385363340378,
0.07569585740566254,
0.016368230804800987,
-0.011283897794783115,
-0.0019319037673994899,
0.06948033720254898,
0.01... |
c8a544cc-66c9-53b3-bb87-eae4819b312b | /course/essentials/day-7/unstructured/ | Integrating with Unstructured.io | Unstructured Enterprise Data Processing > Key Features | course | false | Integrating with Unstructured.io > Unstructured Enterprise Data Processing > Key Features
**VLM Partitioner Capabilities**:
- Intelligent document layout recognition
- Table and figure extraction with context preservation
- Image understanding and text hierarchy analysis
- Multi-modal content processing
**Smart Chunk... | [
0.004742584191262722,
-0.017236914485692978,
-0.07697192579507828,
-0.03298464044928551,
0.06994350254535675,
-0.021378422155976295,
0.05977591499686241,
0.021857352927327156,
0.062475044280290604,
0.0017969525652006269,
-0.0023818344343453646,
-0.020622558891773224,
0.059784404933452606,
... |
c8b4df16-92ff-5a99-a871-23982bc82294 | /documentation/embeddings/gemini/ | Gemini | Setup | documentation | true | Gemini > Setup
Install the Google Gen AI SDK and the Qdrant client:
- Python: `pip install google-genai qdrant-client`
- TypeScript: `npm install @google/genai @qdrant/js-client-rest`
Create both clients and a collection sized for the model. The Gemini client reads your API key from the `GEMINI_API_KEY` environment ... | [
0.010362782515585423,
-0.015455438755452633,
-0.011957431212067604,
-0.050149932503700256,
0.050597403198480606,
-0.03333204239606857,
0.06444961577653885,
0.04182915762066841,
0.04391258955001831,
-0.04894702509045601,
0.05690321698784828,
0.03231566771864891,
0.06803152710199356,
-0.0153... |
c8bf0f4a-7486-546c-a35b-062ea8ad1bc1 | /documentation/cloud-security/ | Cloud Security | Security Considerations > Hybrid Cloud | documentation | false | Cloud Security > Security Considerations > Hybrid Cloud
For Qdrant Hybrid Cloud, the same security considerations as Managed Cloud apply. The key difference is that since the data plane is running in your own infrastructure, you as the customer are responsible for the security of the underlying infrastructure.
Qdrant... | [
-0.0042041148990392685,
0.003200895618647337,
-0.08272427320480347,
-0.05429708585143089,
0.0314781628549099,
-0.041287630796432495,
0.09486325085163116,
0.05802278593182564,
0.05517800524830818,
-0.026813555508852005,
0.021255044266581535,
0.003133055055513978,
0.04861973971128464,
-0.037... |
c8cc4f5b-eb89-5123-ac31-4292a1ca1a36 | /documentation/search/hybrid-queries/ | Hybrid Queries | Hybrid Search > Reciprocal Rank Fusion (RRF) > Weighted RRF | documentation | true | Hybrid Queries > Hybrid Search > Reciprocal Rank Fusion (RRF) > Weighted RRF
_Available as of v1.17.0_
By default, each query is assigned an equal weight. In reality, one retriever is often stronger than the other for a given workload. For example, a dense retriever may dominate on natural-language queries, while BM2... | [
0.02590886317193508,
-0.04388043284416199,
-0.05831313505768776,
-0.05605040863156319,
0.07659940421581268,
-0.034835100173950195,
0.03047865629196167,
0.03229224309325218,
0.04072810709476471,
-0.025617538020014763,
0.0092724384739995,
0.0056745982728898525,
0.04494433104991913,
-0.012838... |
c8df2ace-d824-5bcf-9441-c768a3af2bee | /documentation/tutorials-search-engineering/pdf-retrieval-at-scale/ | Multivector Document Retrieval | Challenges of Scaling VLLMs > Our Solution | documentation | false | Multivector Document Retrieval > Challenges of Scaling VLLMs > Our Solution
We recommend reducing the number of vectors in a PDF page representation for the **first-stage retrieval**. After the first stage retrieval with a reduced amount of vectors, we propose to **rerank** retrieved subset with the original uncompres... | [
0.029967570677399635,
-0.05094961076974869,
-0.0360419899225235,
-0.05369469150900841,
0.10319055616855621,
0.012780645862221718,
0.028718000277876854,
0.05295217037200928,
0.08491283655166626,
-0.004754760768264532,
0.02094111777842045,
-0.00634279940277338,
0.036308083683252335,
-0.02910... |
c8e2fc67-6004-57c5-8cf7-657426f99895 | /documentation/tutorials-operations/embedding-model-migration/ | Migrate to a New Embedding Model | Migrate Using Named Vectors | documentation | false | Migrate to a New Embedding Model > Migrate Using Named Vectors
If your collection uses [named vectors](/documentation/manage-data/points/#named-vectors) and your deployment is running version 1.18 or later, you can migrate to a new embedding model without creating a second collection. Instead, [add the new model as an... | [
0.06250682473182678,
-0.06028888002038002,
-0.015012514777481556,
-0.019170686602592468,
0.05409522354602814,
-0.029607480391860008,
0.038059018552303314,
0.049784496426582336,
0.03027699701488018,
-0.03230899199843407,
0.0366157703101635,
0.03944719582796097,
0.018665960058569908,
-0.0266... |
c90f20dd-6798-5059-9c66-9162b85aab61 | /documentation/cloud/inference/ | Inference | Enabling/Disabling Inference | documentation | false | Inference > Enabling/Disabling Inference
Inference is enabled by default for all new clusters created after July 7, 2025. You can enable it for existing clusters directly from the Inference tab of the Cluster Detail page in the Qdrant Cloud Console. Activating inference triggers a restart of your cluster to apply the ... | [
0.025442959740757942,
0.0008367799455299973,
-0.014095335267484188,
-0.06119117513298988,
0.0624215342104435,
-0.044260576367378235,
0.03389398008584976,
0.067903071641922,
0.06924506276845932,
-0.04183543846011162,
0.029407285153865814,
0.027504613623023033,
0.06371164321899414,
-0.062346... |
c929478e-aed3-5b50-8087-fd666a089e71 | /documentation/tutorials-search-engineering/query-decomposition/ | Query Decomposition for Multi-Hop Questions | How It Works | documentation | true | Query Decomposition for Multi-Hop Questions > How It Works
Start with the clients and a cap on the number of hops.
```python
from openai import OpenAI
from qdrant_client import QdrantClient, models
from your_embedding_model import embed # must match the model your collection uses
llm = OpenAI(api_key="<your-api-ke... | [
0.04270564392209053,
-0.04908164590597153,
-0.0798444077372551,
-0.03540824353694916,
0.07092830538749695,
-0.02865409106016159,
0.06636816263198853,
-0.0025507176760584116,
0.06322287768125534,
-0.02822640910744667,
0.04288134723901749,
-0.014982225373387337,
0.05523965507745743,
-0.01191... |
c9512725-3fc9-518d-b10a-12f1b09ea7d3 | /documentation/ops-configuration/configuration/ | Configuration | Configuration Options | documentation | true | Configuration > Configuration Options
The following YAML example describes the available configuration options.
See [Memory Tiers](/documentation/ops-configuration/memory-tiers/) for details on the `memory` options shown below.
```yaml
log_level: INFO
# Logging configuration
# Qdrant logs to stdout. You may configu... | [
0.029736263677477837,
-0.00984545424580574,
-0.029980076476931572,
-0.0615774504840374,
0.047178588807582855,
-0.04417039453983307,
0.041627187281847,
0.06640350818634033,
0.07725855708122253,
0.0007990925805643201,
0.008721756748855114,
0.03163115307688713,
0.07098614424467087,
-0.0339344... |
c95ade52-1456-5c9b-8b60-cfccb016d83b | /course/essentials/day-5/pitstop-project/ | Project: Building a Recommendation System | Your Mission | course | false | Project: Building a Recommendation System > Your Mission
Build a complete recommendation system using Qdrant’s Universal Query API with dense, sparse, and ColBERT multivectors in one request.
**Estimated Time:** 90 minutes | [
0.000894861645065248,
-0.040327269583940506,
-0.02028096839785576,
-0.026893066242337227,
0.06288815289735794,
-0.03050605021417141,
0.06799620389938354,
0.08680935949087143,
0.08403254300355911,
-0.02839965745806694,
0.06329656392335892,
0.03356301411986351,
0.06351914256811142,
-0.056687... |
c964bece-376a-5610-a12b-86192d4665ed | /documentation/tutorials-operations/embedding-model-migration/ | Migrate to a New Embedding Model | Check Retrieval Quality Before Switching | documentation | false | Migrate to a New Embedding Model > Check Retrieval Quality Before Switching
- **Decide the bar first.** Pick what metric to look at, and how big the gains have to be to justify the switch to a new embedding model. Different metrics can rank the two models differently.
- **Look at individual queries.** Average can ofte... | [
0.04708687961101532,
-0.05089753121137619,
-0.06044415757060051,
-0.012963718734681606,
0.03273357078433037,
-0.002758743241429329,
0.03839626908302307,
-0.000035498385841492563,
0.06257744133472443,
-0.03030528500676155,
0.027118029072880745,
0.021132096648216248,
0.054516810923814774,
-0... |
c96eea36-1b06-54a6-b0c2-66fe24aa7bc5 | /course/beginners/module-3/fusion-strategies/ | Fusion Strategies | course | false | Fusion Strategies
Module 3
The difference between the two strategies is what each does with magnitude. RRF knows only that a document came first, second, or third, so a runaway top match and a photo finish look identical to it. DBSF rescales each retriever's scores onto a comparable range before combining them, which... | [
0.024556774646043777,
-0.02175232209265232,
-0.04210883378982544,
-0.05526014044880867,
0.041084252297878265,
-0.03459076210856438,
0.067080557346344,
0.014486514963209629,
0.037411004304885864,
-0.025516605004668236,
0.021242011338472366,
-0.014342593960464,
0.051932573318481445,
0.026658... | |
c97437b7-8e64-5b6d-9cba-60be5b2788a4 | /documentation/ops-monitoring/monitoring/ | Monitoring & Telemetry | Metrics > Node Metrics /metrics | documentation | false | Monitoring & Telemetry > Metrics > Node Metrics /metrics
[^metrics-per-collection]: When `/metrics?per_collection=true` is used, these metrics include a `collection` label. See [Per-Collection API Metrics](#per-collection-api-metrics).
The output does not include metrics for the collection info, listing, and snapshot... | [
-0.029017770662903786,
-0.0021004134323447943,
-0.015988154336810112,
-0.025224434211850166,
0.04020224139094353,
-0.024907929822802544,
0.05297219008207321,
0.07579604536294937,
0.08760277926921844,
-0.08149351179599762,
0.043904099613428116,
-0.001187986577861011,
0.05677903816103935,
0.... |
c995a85b-d313-570f-8b8a-978165f92d0a | /documentation/manage-data/quantization/ | Quantization | Binary Quantization > Binary Quantization as Hamming Distance | documentation | false | Quantization > Binary Quantization > Binary Quantization as Hamming Distance
The additional benefit of this method is that you can efficiently emulate Hamming distance with dot product.
Specifically, if original vectors contain `{-1, 1}` as possible values, then the dot product of two vectors is equal to the Hamming ... | [
0.0773092657327652,
-0.07449324429035187,
-0.02475564368069172,
-0.028665170073509216,
0.07019398361444473,
-0.0055152541026473045,
0.04273740574717522,
0.05863054469227791,
0.12110765278339386,
-0.013495950028300285,
0.043419647961854935,
0.00023613261873833835,
0.05956635624170303,
-0.00... |
c9b039d3-728e-54fe-90ee-f862e1bea180 | /documentation/manage-data/storage/ | Storage | Versioning | documentation | false | Storage > Versioning
To ensure data integrity, Qdrant performs all data changes in 2 stages.
In the first step, the data is written to the Write-ahead-log(WAL), which orders all operations and assigns them a sequential number.
Once a change has been added to the WAL, it will not be lost even if a power loss occurs.
T... | [
0.023868277668952942,
-0.06367558240890503,
-0.011690271086990833,
-0.0019591832533478737,
0.03221255540847778,
-0.01596543751657009,
0.025515923276543617,
0.033480457961559296,
0.06771355867385864,
-0.007277133874595165,
0.007824944332242012,
0.03619673103094101,
0.05742607265710831,
-0.0... |
c9b8c289-64db-5cd8-a2c0-b445e9a963fd | /course/beginners/ | Qdrant Beginner Course | Who It's For | course | false | Qdrant Beginner Course > Who It's For
Anyone new to vector search who wants to understand the fundamentals. No prior experience with Qdrant or vector search engines required. | [
0.013513129204511642,
-0.06942352652549744,
-0.04816001281142235,
-0.03358950838446617,
0.04623255878686905,
-0.015241263434290886,
0.04888209328055382,
0.0173869039863348,
0.06725767254829407,
-0.035016510635614395,
0.010925296694040298,
0.03426340967416763,
0.04140171408653259,
-0.040639... |
c9e7d2f6-1ec5-50ba-907e-7fcb286bcc90 | /course/essentials/ | Qdrant Essentials Course | Syllabus | course | false | Qdrant Essentials Course > Syllabus
- title: "Day 0: Setup and First Steps"
content: |
- Qdrant Cloud Setup
- Implementing a Basic Vector Search
- Project: Building Your First Vector Search System
→ Start Day 0
- title: "Day 1: Vector Search Fundamentals"
content: |
- Points, Vectors and Payloads... | [
0.05392317846417427,
-0.03525004908442497,
-0.055444274097681046,
-0.008046996779739857,
0.07170073688030243,
-0.03930958732962608,
0.06409305334091187,
0.038594525307416916,
0.07491695880889893,
-0.04477108269929886,
0.035305846482515335,
0.042365144938230515,
0.042130760848522186,
-0.070... |
c9fdf99f-7ef2-5000-a648-89177545667b | /documentation/cloud-quickstart/ | Cloud Quickstart | 3. Connect to Qdrant Cloud | documentation | true | Cloud Quickstart > 3. Connect to Qdrant Cloud
Import the qdrant client and create a connection to your Qdrant Cloud cluster using your cluster URL and API key.
```python
from qdrant_client import QdrantClient
from qdrant_client.models import Distance, VectorParams, PointStruct, Document
# connect to Qdrant Cloud
cli... | [
0.013408901169896126,
-0.025891786441206932,
-0.05028524622321129,
-0.02907995507121086,
0.05928521975874901,
-0.03604049235582352,
0.04731188341975212,
0.04235337674617767,
0.08098379522562027,
-0.04536811634898186,
0.016529861837625504,
0.0059299468994140625,
0.036178313195705414,
-0.024... |
ca3b4e25-11e0-5084-b26b-47c1c4c16a28 | /documentation/search-patterns/vector-search-filtering/ | Filtering in Vector Search: When to Use What | Common Pitfalls | documentation | false | Filtering in Vector Search: When to Use What > Common Pitfalls
- **Floats need `range`, not `match`.** Qdrant rejects a `match` condition with a float value such as `11.99`. To find an exact stored value, set both bounds to it: `"range": { "gte": 11.99, "lte": 11.99 }`. If values carry rounding errors, use a small win... | [
0.022204076871275902,
-0.04513055086135864,
-0.04646967351436615,
-0.014514032751321793,
0.06786196678876877,
-0.02432972379028797,
0.02872743271291256,
0.010881011374294758,
0.07990215718746185,
-0.019146787002682686,
0.0450853668153286,
0.01982150413095951,
0.05308573693037033,
-0.019831... |
ca3be733-3d29-5e71-a3b9-e20565297246 | /documentation/embeddings/openai/ | OpenAI | Setup | documentation | true | OpenAI > Setup
Install the OpenAI SDK and the Qdrant client:
- Python: `pip install openai qdrant-client`
- TypeScript: `npm install openai @qdrant/js-client-rest`
Create both clients and a collection sized for the model. The OpenAI client reads your API key from the `OPENAI_API_KEY` environment variable.
```python... | [
0.043809641152620316,
-0.04916858673095703,
0.010097018443048,
-0.04722065478563309,
0.06552950292825699,
-0.03677983209490776,
0.05067921057343483,
0.06982345134019852,
0.039466455578804016,
-0.019635286182165146,
0.059621937572956085,
-0.001445929054170847,
0.07385069131851196,
-0.046108... |
ca42d195-707c-5dd6-b8ea-522981c91382 | /documentation/examples/rag-chatbot-vultr-dspy-ollama/ | Private RAG Information Extraction Engine | documentation | false | Private RAG Information Extraction Engine
| Time: 90 min | Level: Advanced | | |
|--------------|-----------------|--|----|
Handling private documents is a common task in many industries. Various businesses possess a large amount of
unstructured data stored as huge files that must be processed and analyzed. Indu... | [
0.02976413257420063,
-0.0190220195800066,
-0.07172596454620361,
-0.023763606324791908,
0.06903215497732162,
-0.012607521377503872,
0.071317657828331,
0.0727333202958107,
0.05354416370391846,
-0.029897179454565048,
-0.008533905260264874,
0.024454936385154724,
0.06408990174531937,
-0.0420237... | |
ca55f28b-0d8e-57ea-90fe-000ed2476328 | /documentation/tutorials-build-essentials/video-anomaly-edge-part-2/ | Video Anomaly Detection Part 2: Edge-to-Cloud Pipeline | Edge Triage: Why Imperfect Is the Point | documentation | false | Video Anomaly Detection Part 2: Edge-to-Cloud Pipeline > Edge Triage: Why Imperfect Is the Point
The two-tier architecture only works *because* the edge is imperfect and the cloud cleans up after it. If the edge were perfect, you wouldn't need the cloud. If the edge were random, you wouldn't save any bandwidth. The sw... | [
0.01373966969549656,
-0.019645383581519127,
-0.042081769555807114,
-0.04995756596326828,
0.04994017630815506,
-0.012548697181046009,
0.050544388592243195,
-0.017093777656555176,
0.08065363764762878,
-0.020693251863121986,
0.014192324131727219,
0.015216955915093422,
0.06501754373311996,
0.0... |
ca5bec4c-2c03-5660-88e5-fa4d1d1e2b37 | /articles/turboquant-quantization/ | TurboQuant in Qdrant | What Qdrant Adds > SIMD Acceleration > 1-bit: RaBitQ bit-plane scoring | article | false | TurboQuant in Qdrant > What Qdrant Adds > SIMD Acceleration > 1-bit: RaBitQ bit-plane scoring
The 1-bit scoring path follows **[RaBitQ](https://arxiv.org/abs/2405.12497)** as-is, without specific modifications. The data is one bit per coordinate (the sign of the rotated coordinate), bit-packed at 128 dimensions per 16... | [
0.048946648836135864,
-0.03697175160050392,
-0.03586771339178085,
-0.036769840866327286,
0.055095456540584564,
-0.01544139813631773,
0.04037635028362274,
0.0498892106115818,
0.046605754643678665,
-0.03310105949640274,
0.08605240285396576,
0.00490422360599041,
0.05713961645960808,
-0.018967... |
ca6251fd-2071-5157-bc77-5e3dbe621cf1 | /documentation/data-synchronization/with-postgres/ | Keeping Postgres and Qdrant in Sync | Tier 3: Change Data Capture with Debezium + Redpanda > Debezium Connector Config | documentation | true | Keeping Postgres and Qdrant in Sync > Tier 3: Change Data Capture with Debezium + Redpanda > Debezium Connector Config
```json
{
"name": "products-connector",
"config": {
"connector.class": "io.debezium.connector.postgresql.PostgresConnector",
"database.hostname": "postgres",
"database.... | [
0.05572304129600525,
-0.06617993861436844,
-0.015460110269486904,
-0.03762667626142502,
0.012993823736906052,
-0.037361226975917816,
0.07084605097770691,
0.056740593165159225,
0.05877160280942917,
-0.04043467342853546,
0.00992578361183405,
0.040556855499744415,
0.0789991095662117,
-0.02067... |
ca647026-8915-5820-8c4c-3e6470495673 | /course/beginners/module-4/references-and-further-reading/ | References and Further Reading | course | false | References and Further Reading
Module 4
- [Sizing Tool](https://sizing.qdrant.tech) How much memory and disk a collection needs, before you commit to a node.
- [What Is RAG](/articles/what-is-rag-in-ai/) The retrieval and generation pattern in depth.
- [Qdrant Cloud](https://cloud.qdrant.io/) Create a free cluster be... | [
0.023534370586276054,
-0.01958424225449562,
-0.043040309101343155,
-0.04790601506829262,
0.10382191836833954,
-0.05121764540672302,
0.05197853595018387,
0.08253496885299683,
0.0769505575299263,
-0.060915328562259674,
0.03559155762195587,
0.04286457598209381,
0.04722556471824646,
-0.0384263... | |
ca81ab6d-9291-5a1f-8c85-82d6acea909e | /articles/turboquant-quantization/ | TurboQuant in Qdrant | What Is TurboQuant? | article | false | TurboQuant in Qdrant > What Is TurboQuant?
TurboQuant ([Zandieh et al., 2025](https://arxiv.org/abs/2504.19874)) is a rotation-based vector quantization algorithm in the Product Quantization (PQ) family, with a clean theoretical recipe:
1. **Apply a random orthogonal rotation** to every vector. This redistributes per... | [
0.027822047472000122,
-0.036013413220644,
-0.05399115011096001,
-0.03366083651781082,
0.05981927365064621,
0.016616519540548325,
0.06270699948072433,
-0.007864536717534065,
0.07122522592544556,
-0.02285691723227501,
0.031259648501873016,
0.02734539657831192,
0.05829434096813202,
-0.0335048... |
ca88b0ad-a375-5360-b566-ec517e635bce | /articles/what-is-a-vector-database/ | What is a Vector Database? | Hybrid Search > Dense vs. Sparse Vectors > 2. Sparse Vectors | article | true | What is a Vector Database? > Hybrid Search > Dense vs. Sparse Vectors > 2. Sparse Vectors
Sparse vectors operate differently. They focus only on the essentials. In most sparse vectors, a large number of elements are zeros. When a feature or token is present, it’s marked; otherwise, zero.
In the image, you can see a s... | [
0.02987859770655632,
-0.03492671623826027,
-0.07489103823900223,
-0.03797821328043938,
0.043479204177856445,
-0.026307320222258568,
0.059614572674036026,
0.0006708164582960308,
0.09153490513563156,
-0.05258999019861221,
0.026861252263188362,
0.037447601556777954,
0.042185649275779724,
0.00... |
cab6eb75-da39-5690-a0f9-7b7821234d37 | /documentation/hybrid-cloud/platform-deployment-options/ | Deployment Platforms | Next Steps | documentation | false | Deployment Platforms > Next Steps
Once you've got a Kubernetes cluster deployed on a platform of your choosing, you can begin setting up Qdrant Hybrid Cloud. Head to our Qdrant Hybrid Cloud [setup guide](/documentation/hybrid-cloud/hybrid-cloud-setup/) for instructions. | [
0.013119947165250778,
-0.017871182411909103,
-0.0427522212266922,
-0.05401674285531044,
0.0671016201376915,
-0.05738729611039162,
0.04596953094005585,
0.0048399511724710464,
0.044829271733760834,
-0.026895392686128616,
0.051768627017736435,
0.010736600495874882,
0.04800272360444069,
-0.021... |
cabce4b8-e215-551e-aaf2-4740dd2fb0ff | /documentation/private-cloud/changelog/ | Changelog | 1.9.11 (2026-07-08) | documentation | false | Changelog > 1.9.11 (2026-07-08)
| Component | Version |
|-------------------------|---------|
| qdrant-kubernetes-api | v1.37.1 |
| operator | 2.33.0 |
| qdrant-cluster-manager | v0.3.19 |
| qdrant-cluster-exporter | 1.8.0 |
Latest validated Qdrant version: 1.18.2
* Updated operato... | [
0.033893197774887085,
-0.017403440549969673,
-0.060614485293626785,
-0.049316007643938065,
0.046037573367357254,
-0.048132847994565964,
0.05894503742456436,
0.03972364962100983,
0.0459265261888504,
-0.03707318753004074,
0.036823172122240067,
0.04327520728111267,
0.04830623045563698,
-0.064... |
cac60c4b-ca00-56ca-8d02-2725b1ccdd88 | /course/essentials/day-2/what-is-hnsw/ | HNSW Indexing Fundamentals | What's Next | course | false | HNSW Indexing Fundamentals > What's Next
Now you understand how HNSW makes vector search fast and scalable. Next we'll combine fast search with complex filters using Qdrant’s filter‑aware HNSW.
Learn more: [HNSW in Qdrant Documentation](/documentation/manage-data/indexing/#vector-index)
Ready to see how HNSW handles... | [
0.03849826008081436,
-0.01854587346315384,
-0.04468362033367157,
-0.03850489854812622,
0.03319460526108742,
-0.03076723963022232,
0.07328078895807266,
0.033841054886579514,
0.049549560993909836,
0.0011150925420224667,
0.032759685069322586,
0.05533068999648094,
0.06391656398773193,
-0.00593... |
cac699d2-e29d-585a-967d-e29658506366 | /documentation/tutorials-search-engineering/branch-aware-search/ | Branch-Aware Search | Step 2: Update a File | documentation | true | Branch-Aware Search > Step 2: Update a File
A second commit changes a file that already exists.
```text
root ●───────● seq 1: update pricing.md
seq 0 seq 1
```
The old point stays in the collection. We tag it as superseded so a later read skips it, with an `overwritten_in` entry: a `{by, seq}` record... | [
0.04421331733465195,
-0.06643571704626083,
-0.05181242153048515,
-0.017029402777552605,
0.06236598268151283,
-0.016616418957710266,
-0.012873800471425056,
0.007867320440709591,
0.04463782534003258,
-0.03142085298895836,
0.028329260647296906,
0.0714230015873909,
0.05123833939433098,
-0.0223... |
cac89540-0a91-5fc7-86b4-21d1994a97e0 | /documentation/migration-guidance/diagnosing-discrepancies/ | Diagnosing Discrepancies | documentation | false | Diagnosing Discrepancies
When verification catches a problem, you need to determine whether it's a data issue (something went wrong during migration) or a configuration issue (the data is correct but the systems behave differently). This page provides a diagnostic decision tree and vendor-specific gotchas. | [
0.03639481961727142,
-0.027757683768868446,
-0.05403749272227287,
-0.05862665921449661,
0.0006053417455404997,
-0.014707624912261963,
0.03733455762267113,
0.05796173959970474,
0.04950855299830437,
0.0013539752690121531,
-0.015732640400528908,
0.05819627270102501,
0.061686575412750244,
-0.0... | |
cad0cbaa-a4ac-5bf5-a62b-14eaa35d3ec7 | /articles/minicoil-en-es/ | miniCOIL EN-ES: Sparse Neural Retrieval Across the Language Barrier | Results | article | false | miniCOIL EN-ES: Sparse Neural Retrieval Across the Language Barrier > Results
The published checkpoint is at [Jocana/minicoil-en-es](https://huggingface.co/Jocana/minicoil-en-es).
> **License note:** Both the checkpoint and the training code are released under CC BY-NC 4.0: the concept vocabulary is built from the MU... | [
0.05825500190258026,
-0.023107197135686874,
-0.026732932776212692,
-0.054012689739465714,
0.06976176798343658,
-0.0249253548681736,
0.07535579055547714,
0.024095388129353523,
0.049117520451545715,
-0.046310994774103165,
0.03783031180500984,
0.03532729670405388,
0.08023466169834137,
-0.0229... |
cae95921-ad22-5755-8f38-c53bc6f98b92 | /articles/bm42/ | BM42: Attention-Based Sparse Embeddings for Hybrid Search | Practical examples | article | true | BM42: Attention-Based Sparse Embeddings for Hybrid Search > Practical examples
| Trait | BM25 | SPLADE | BM42 |
|-------------------------|--------------|--------------|--------------|
| Interpretability | High ✅ | Ok 🆗 | High ✅ |
| Document Inference sp... | [
0.05507489666342735,
-0.024626579135656357,
-0.061646077781915665,
-0.06673400849103928,
0.05155908688902855,
-0.08351040631532669,
0.0687294453382492,
0.06514563411474228,
0.03697501868009567,
-0.02028505504131317,
0.033129263669252396,
0.008818614296615124,
0.037890512496232986,
-0.02046... |
caf75881-4803-5545-aa03-2c2077b7b27e | /documentation/tutorials-develop/code-search/ | Semantic Search for Code | Data preparation > Parsing the codebase | documentation | true | Semantic Search for Code > Data preparation > Parsing the codebase
While our example uses Rust, you can use our approach with any other language.
You can parse code with a [Language Server Protocol](https://microsoft.github.io/language-server-protocol/) (**LSP**)
compatible tool. You can use an LSP to build a graph of... | [
0.047907739877700806,
-0.031006913632154465,
-0.05472717434167862,
-0.022607658058404922,
0.0747494250535965,
-0.03594409301877022,
0.05152440816164017,
-0.010547318495810032,
0.022466719150543213,
-0.03835654631257057,
0.04228431358933449,
0.006471778731793165,
0.06933044642210007,
-0.037... |
caf8d6a4-e573-5220-ac44-deb652ccba15 | /documentation/embeddings/gemini/ | Gemini | Embed Images, Video, Audio, and PDFs | documentation | true | Gemini > Embed Images, Video, Audio, and PDFs
`gemini-embedding-2` embeds files into the same vector space as text, so a collection of PDFs can be searched with the text query from [Embed Queries](#embed-queries). This example embeds a PDF and stores it in the same collection:
```python
with open("filename.pdf", "rb"... | [
0.0025416354183107615,
-0.034215573221445084,
-0.06648280471563339,
-0.07501581311225891,
0.05798781290650368,
0.00866877194494009,
0.04380238428711891,
0.051487259566783905,
0.07903996109962463,
-0.03203999251127243,
-0.002716783666983247,
-0.005118196364492178,
0.03441786766052246,
-0.02... |
cafcfbad-612f-5c12-b98f-40197f219404 | /documentation/search-patterns/vector-search-filtering/ | Filtering in Vector Search: When to Use What | Index Every Field You Filter On | documentation | true | Filtering in Vector Search: When to Use What > Index Every Field You Filter On
Qdrant uses payload indexes in three ways: to find matching points without scanning every payload, to estimate how many points a filter matches so the [query planner](#let-the-query-planner-choose) can pick a strategy, and to build the extr... | [
0.03239299729466438,
-0.043088898062705994,
-0.05246119946241379,
-0.008912711404263973,
0.07236957550048828,
-0.04424828290939331,
0.05068860203027725,
0.02479039877653122,
0.07508692890405655,
-0.03659022971987724,
0.043156906962394714,
0.03137515112757683,
0.042010076344013214,
-0.00778... |
cb324650-d5f2-50d1-88ce-fb8d53d62625 | /course/multi-vector-search/module-1/maxsim-distance/ | MaxSim Distance Metric | What's Next | course | false | MaxSim Distance Metric > What's Next
MaxSim gives us a way to compare multi-vector representations, capturing fine-grained semantic matching that single-vector search cannot achieve. But as we've seen, it also introduces significant computational challenges - particularly the asymmetry that breaks traditional indexing... | [
0.059816282242536545,
-0.017252089455723763,
-0.011505703441798687,
-0.00923210196197033,
0.07613281905651093,
-0.044900741428136826,
0.035897139459848404,
0.0020088818855583668,
0.07226203382015228,
-0.03366711363196373,
0.037727441638708115,
0.03454167768359184,
0.09219387918710709,
-0.0... |
cb3f55e3-9ff3-57aa-ac9d-60f1e21bcdd8 | /documentation/faq/qdrant-fundamentals/ | Qdrant Fundamentals | Search > How does Qdrant handle real-time data updates and search? | documentation | false | Qdrant Fundamentals > Search > How does Qdrant handle real-time data updates and search?
Qdrant supports live updates for vector data, with newly inserted, updated and deleted vectors available for immediate search. The system uses full-scan search on unindexed segments during background index updates. | [
0.021973084658384323,
-0.031281255185604095,
-0.05537772923707962,
-0.007133152335882187,
0.032497942447662354,
-0.010713807307183743,
0.08737155050039291,
0.007516634650528431,
0.08123031258583069,
-0.022758441045880318,
0.02153264358639717,
0.03941383957862854,
0.05468872934579849,
-0.04... |
cb43456f-1ea5-5989-b959-ca6c2d34e0b9 | /documentation/embeddings/jina-embeddings/ | Jina AI | Qdrant Cloud Inference | documentation | true | Jina AI > Qdrant Cloud Inference
On Qdrant Cloud, Qdrant can call Jina AI for you. Instead of a vector, send a `Document` or an `Image` with the model name prefixed with `jinaai/`. Pass your Jina AI API key in the `jina-api-key` request header or in the `options` object. Qdrant does not store the key, so send it with ... | [
0.050291333347558975,
-0.03506160154938698,
-0.02500828541815281,
-0.05362781509757042,
0.05037451535463333,
-0.02557949535548687,
0.08403325825929642,
0.056328028440475464,
0.03974737972021103,
-0.024864936247467995,
0.026914650574326515,
0.014243505895137787,
0.0528600700199604,
-0.02342... |
cb49b241-e259-56e5-8734-5a2403dcc19c | /documentation/search/explore/ | Explore | Discovery API > Context Search | documentation | true | Explore > Discovery API > Context Search
Conversely, in the absence of a target, a rigid integer-by-integer function doesn't provide much guidance for the search when utilizing a proximity graph like HNSW. Instead, context search employs a function derived from the [triplet-loss](/articles/triplet-loss/) concept, whic... | [
0.05206403136253357,
-0.04435179755091667,
-0.03191680088639259,
-0.0018291698070243,
0.06154138967394829,
0.007184233516454697,
0.03342888504266739,
0.03435344994068146,
0.05589170381426811,
-0.031442176550626755,
0.028217248618602753,
0.024128466844558716,
0.06432731449604034,
-0.0087561... |
cb51def3-f2ca-5041-99b8-86f802c8acc2 | /course/essentials/day-1/distance-metrics/ | Distance Metrics | Quick rule of thumb > Manhattan Distance | course | true | Distance Metrics > Quick rule of thumb > Manhattan Distance
- **Common Use Cases:** Sparse data, robust outlier handling.
- **Focus:** Grid-based distance (sum of absolute differences).
Manhattan Distance is similar to Euclidean Distance but calculates distance as if moving along grid lines (horizontal and vertical).... | [
0.02396082691848278,
0.004963743966072798,
-0.06855757534503937,
-0.03923133760690689,
0.06255163252353668,
0.01605510339140892,
0.04851670190691948,
0.030341913923621178,
0.11288508027791977,
-0.014572963118553162,
0.03950750082731247,
-0.020663274452090263,
0.04633789137005806,
0.0189935... |
cb5c2ff0-18e9-5deb-8b48-2cfa9ad23921 | /documentation/search-patterns/choose-embedding-model/ | How to Choose an Embedding Model: Evaluation & Tradeoffs | documentation | false | How to Choose an Embedding Model: Evaluation & Tradeoffs
How do you choose the right embedding model to achieve the best search quality? Public benchmarks such as the [Massive Text Embedding Benchmark (MTEB)](https://huggingface.co/spaces/mteb/leaderboard) can help you narrow down the options, but their datasets rarel... | [
0.030632851645350456,
-0.024897824972867966,
-0.03635677695274353,
-0.0384419709444046,
0.03477274253964424,
-0.07919999957084656,
0.061022114008665085,
0.06719082593917847,
0.023474933579564095,
-0.017957456409931183,
0.026306666433811188,
0.044610656797885895,
0.05913830175995827,
-0.088... | |
cb84720a-fcfe-569a-8067-a278d951cbac | /documentation/search/text-search/hybrid-search/ | Hybrid Search | documentation | true | Hybrid Search
[Hybrid search](/documentation/search/hybrid-queries/#hybrid-search) enables you to combine semantic and lexical search in a single query, returning results that match the semantic meaning, the exact keywords, or both. This is useful when you don't know whether the user is looking for a specific keyword ... | [
0.018235651776194572,
-0.032184332609176636,
-0.06835775822401047,
-0.04500818997621536,
0.06689427047967911,
-0.026641789823770523,
0.054843686521053314,
0.005729103460907936,
0.04255714640021324,
-0.01651477813720703,
0.03922823816537857,
0.01409050915390253,
0.06508836150169373,
0.00311... | |
cb954c00-be72-595d-aa06-3b8b41604711 | /documentation/migrate-to-qdrant/from-pgvector/ | From pgvector | Concept Mapping | documentation | false | From pgvector > Concept Mapping
## What You Need from Postgres
- **Connection URL** — a standard Postgres connection string
- **Table name** — the table containing your vector data
| pgvector | Qdrant | Notes |
| :--- | :--- | :--- |
| Table | Collection | One-to-one mapping |
| Row | Point | Each row becomes a poin... | [
0.0603814534842968,
-0.0702102929353714,
-0.016376828774809837,
-0.023722456768155098,
0.04205533117055893,
-0.0050337910652160645,
0.07606224715709686,
0.06364616751670837,
0.0668325275182724,
-0.012685299851000309,
0.02452319674193859,
0.022373700514435768,
0.02620181441307068,
-0.013662... |
cb9c0d65-c3ac-5946-9013-51642a1562c5 | /course/essentials/day-0/building-simple-vector-search/ | Implementing a Basic Vector Search | Step 7: Retrieve Collection Details | course | true | Implementing a Basic Vector Search > Step 7: Retrieve Collection Details
Now that we've inserted vectors, let's confirm they're stored correctly by getting collection information:
```python
collection_info = client.get_collection(collection_name)
print("Collection info:", collection_info)
```
Expected output: Detail... | [
0.02551724761724472,
-0.06316330283880234,
-0.015953905880451202,
-0.05602008104324341,
0.009316776879131794,
-0.0429084338247776,
0.03385327011346817,
0.06975022703409195,
0.05805816873908043,
-0.027725547552108765,
0.02617011032998562,
0.07097205519676208,
0.051642101258039474,
-0.027761... |
cba28f62-1b2b-574e-81a8-81c663beddc5 | /documentation/cloud/cluster-monitoring/ | Monitor Clusters | Grafana Dashboard > Cluster System Metrics /sys_metrics | documentation | false | Monitor Clusters > Grafana Dashboard > Cluster System Metrics /sys_metrics
| Name | Type | Meaning |
|-----------------------------------... | [
-0.00818535964936018,
-0.005814723204821348,
-0.013401646167039871,
0.0014891226310282946,
0.06966397911310196,
-0.058795489370822906,
0.05289404094219208,
0.05995044484734535,
0.05539000779390335,
-0.04849591851234436,
0.03892534598708153,
-0.006520943716168404,
0.04456590861082077,
-0.01... |
cba66024-6cd5-5084-a284-e9c4c1a145d1 | /course/essentials/day-0/ | Day 0: Setup and First Steps | course | false | Day 0: Setup and First Steps
Day 0
Get started with Qdrant by setting up your environment and building your first vector search application.
---
## Today’s path
1. Qdrant Cloud Setup
2. Implementing a Basic Vector Search
3. Project: Building Your First Vector Search System
By the end, you’ll have a working Qdrant... | [
0.035597752779722214,
-0.08200935274362564,
-0.005366060417145491,
-0.002209355589002371,
0.07458962500095367,
-0.03716910257935524,
0.038652919232845306,
0.0022099846974015236,
0.05836254358291626,
-0.031675443053245544,
0.01583150029182434,
0.017420955002307892,
0.04152360185980797,
-0.0... | |
cba8120a-d1ae-5320-8dda-ad3c12dec22c | /documentation/ops-configuration/memory-tiers/ | Memory Tiers | Default Tiers | documentation | false | Memory Tiers > Default Tiers
If you don't explicitly set `memory` on a structure, Qdrant defaults to the following tiers:
| Data structure | Default tier |
|---|---|
| Dense vectors | `cached` |
| HNSW vector index | `cached` |
| Quantized vectors | Depends on the placement of the original dense vectors: `pinned` if ... | [
0.07031657546758652,
-0.02827649936079979,
-0.026391267776489258,
-0.015622097998857498,
0.04177621752023697,
-0.0494450218975544,
0.07313425838947296,
0.060336556285619736,
0.09327098727226257,
-0.02076551504433155,
0.015581978484988213,
0.03008667193353176,
0.003982964903116226,
-0.00575... |
cbb9c9c9-fb67-5518-b177-6020214c17fa | /documentation/tutorials-search-engineering/sparse-embeddings-ecommerce-part-2/ | Fine-Tuning Sparse Embeddings for E-Commerce Search | Part 2: Training SPLADE on Modal | Prerequisites | documentation | true | Fine-Tuning Sparse Embeddings for E-Commerce Search | Part 2: Training SPLADE on Modal > Prerequisites
- Python 3.10 or later, [uv](https://docs.astral.sh/uv/), and a [Modal](https://modal.com/) account.
- A Qdrant cluster URL and API key, stored as a Modal secret named `qdrant`. The repository's Modal functions refer... | [
0.06379425525665283,
-0.04293780401349068,
-0.022456049919128418,
-0.01299832109361887,
0.030298760160803795,
-0.04013655334711075,
0.05371320620179176,
0.08886396139860153,
0.07975991070270538,
-0.03491733595728874,
0.029942484572529793,
0.0020268799271434546,
0.042427908629179,
-0.023973... |
cbc624f7-8b6a-5d58-8799-f6f51869a6fa | /documentation/fastembed/fastembed-splade/ | Working with SPLADE | Observations | documentation | false | Working with SPLADE > Observations
- The relative order of importance is quite useful. The most important tokens in the sentence have the highest weights.
- **Term Expansion:** The model can expand the terms in the document. This means that the model can generate weights for tokens that are not present in the document... | [
0.03000311739742756,
-0.060485582798719406,
-0.04663467034697533,
-0.04197012633085251,
0.05079035460948944,
-0.019590364769101143,
0.05911024287343025,
0.02470131777226925,
0.07886165380477905,
-0.03458833321928978,
0.012574935331940651,
0.017114292830228806,
0.030338257551193237,
-0.0222... |
cbcb1a91-46d2-5276-aeb3-e42251277626 | /course/multi-vector-search/module-3/evaluating-pipelines/ | Evaluating Search Pipelines | Unified Collection Architecture | course | true | Evaluating Search Pipelines > Unified Collection Architecture
The traditional approach creates a separate collection for each configuration - tedious and storage-intensive. A better approach: **store all vector representations in a single collection** using named vectors.
### Four Named Vectors
We'll store four diff... | [
0.05122193694114685,
-0.045268330723047256,
-0.05167173594236374,
-0.029588889330625534,
0.07249819487333298,
-0.006179286632686853,
0.05416073650121689,
0.05791202560067177,
0.05085258185863495,
-0.005781822372227907,
-0.006799985654652119,
0.03822054713964462,
0.04871528223156929,
-0.022... |
cbce7275-883d-5ea7-bfa5-745b5f68ebc0 | /articles/modern-sparse-neural-retrieval/ | Modern Sparse Neural Retrieval: From DeepCT to SPLADE++ | Sparse Neural Retrieval: As If Keyword-Based Retrievers Understood Meaning | article | false | Modern Sparse Neural Retrieval: From DeepCT to SPLADE++ > Sparse Neural Retrieval: As If Keyword-Based Retrievers Understood Meaning
**Keyword-based (lexical) retrievers** like BM25 provide a good explainability.
If a document matches a query, it’s easy to understand why: query terms are present in the document,
and... | [
0.05654965341091156,
-0.06087157502770424,
-0.044798873364925385,
-0.06685721129179001,
0.04276198521256447,
-0.04760614037513733,
0.052639976143836975,
0.012612240388989449,
0.06384111195802689,
-0.002804310992360115,
0.005905601661652327,
0.014434189535677433,
0.0635107010602951,
-0.0288... |
cbd500f2-43d2-51e1-8804-f472408f576f | /documentation/search-tuning/when-a-reranker-is-worth-it/ | When Is a Reranker Worth It? | Compare with the Best First Stage | documentation | false | When Is a Reranker Worth It? > Compare with the Best First Stage
Compare the reranker against the strongest first stage you can build. Qdrant's default reciprocal rank fusion (RRF) is already a solid baseline, and fusion tuned on your own labels is stronger. A reranker measured against the default can look like a win ... | [
0.041116394102573395,
-0.03813061863183975,
-0.04351237788796425,
-0.010032507590949535,
0.057785458862781525,
-0.04873230308294296,
0.0363541841506958,
0.0378146655857563,
0.0646635890007019,
-0.03874577209353447,
0.027071122080087662,
-0.032316043972969055,
0.03836032375693321,
-0.019265... |
cbe36a85-4da9-53a3-9ed2-465cd8f72cad | /documentation/manage-data/bulk-upload/ | Bulk Upload | Parallelize Across Multiple Threads | documentation | false | Bulk Upload > Parallelize Across Multiple Threads
A single upload thread rarely saturates the server. Split your dataset across two to four concurrent threads, each sending its own stream of batches. This keeps Qdrant's internal write workers busy across shards and reduces total upload time.
If [your collection has m... | [
0.009192711673676968,
-0.015151643194258213,
-0.04545920714735985,
-0.009044058620929718,
0.09096048772335052,
-0.02874952182173729,
0.0541214793920517,
0.027771100401878357,
0.07292678207159042,
-0.023776531219482422,
0.02038874477148056,
0.024182653054594994,
0.04023845121264458,
-0.0129... |
cbf9471a-8d3e-5e07-9e79-d587404f622d | /articles/sparse-vectors/ | Understanding SPLADE and Sparse Vectors | Practical implementation in Python > 2. Create a collection with sparse vector support | article | true | Understanding SPLADE and Sparse Vectors > Practical implementation in Python > 2. Create a collection with sparse vector support
The following code creates a collection with one sparse vector named `text`. Sparse vectors need no size or distance settings, and Qdrant keeps the sparse index in RAM by default. To move it... | [
0.05419493094086647,
-0.053828202188014984,
-0.01184283010661602,
-0.03620636463165283,
0.0444604828953743,
-0.054726552218198776,
0.05587455630302429,
0.022460853680968285,
0.07071276009082794,
-0.02297333814203739,
0.016466684639453888,
0.04679636284708977,
0.033471111208200455,
0.001653... |
cbfaa1ec-46ad-5491-8f5f-6524b05bcf73 | /documentation/migration-recovery-options/ | Migration and Recovery | Comparison > Performance & Cost | documentation | false | Migration and Recovery > Comparison > Performance & Cost
How each option handles indexing on the target, data transfer speed, network cost, and load placed on the source cluster.
| | Migration Tool | Snapshots | Cloud Backups |
|---|---|---|---|
| **Indexing on target** | Full re-index (slow) | None (index transfers ... | [
0.018961992114782333,
-0.03749062120914459,
-0.04986565187573433,
-0.018384438008069992,
0.03438703715801239,
0.019663210958242416,
0.048638343811035156,
0.012577595189213753,
0.07176169753074646,
-0.03586408123373985,
0.01605568639934063,
0.017900822684168816,
0.0327874980866909,
-0.00241... |
cc017060-3b3a-5d3d-9ffb-d6d2eb667216 | /documentation/tutorials-search-engineering/food-discovery-recommendations/ | Multimodal Food Search Demo | Different search modes > Choosing a recommendation strategy | documentation | false | Multimodal Food Search Demo > Different search modes > Choosing a recommendation strategy
Passing only negatives can work as an outlier detection mechanism. The dataset was supposed to contain only food photos, but it does not. Passing food photos as negatives returns the most unlike images, which in this dataset are ... | [
0.04852517694234848,
-0.032585807144641876,
-0.04659000039100647,
-0.01639021374285221,
0.05642135068774223,
0.00813329592347145,
0.06349688023328781,
-0.010964294895529747,
0.042378902435302734,
-0.019058771431446075,
0.024948948994278908,
0.01814711093902588,
0.0451316200196743,
-0.01345... |
cc073088-f2dc-5a5f-8a1c-93ff3e57bfce | /documentation/cloud-account-setup/ | Account Setup | Enterprise Single-Sign-On (SSO) | documentation | false | Account Setup > Enterprise Single-Sign-On (SSO)
Qdrant Cloud supports enterprise single sign-on (SSO) for [Premium Tier](/documentation/cloud-premium/) customers. SSO covers Cloud Console sign-in only. Applications and users that send requests to a cluster still authenticate with [Database API keys](/documentation/clo... | [
0.020634274929761887,
-0.030362673103809357,
-0.03723364695906639,
-0.022706855088472366,
0.01848851703107357,
-0.05404721200466156,
0.054341502487659454,
0.0316847525537014,
0.045069947838783264,
-0.03162158653140068,
0.020279986783862114,
0.009374504908919334,
0.04755284637212753,
-0.032... |
cc0887af-8fe8-504c-bd3d-f9f4c88748bc | /documentation/migrate-to-qdrant/from-faiss/ | From FAISS | What You Need from FAISS | documentation | false | From FAISS > What You Need from FAISS
- **Index file path** — path to your FAISS index file
- **Distance metric** — the metric used when the index was built (`l2`, `inner product`, etc.)
Important: Only non-quantized FAISS index types are supported. Quantized indexes (e.g., IndexIVFPQ) do not store the original vecto... | [
0.02420974150300026,
-0.030702365562319756,
-0.01708718203008175,
-0.008412100374698639,
0.05673157423734665,
-0.023911835625767708,
0.045440878719091415,
0.026442084461450577,
0.0931241437792778,
-0.03808887302875519,
0.031095363199710846,
-0.00255294400267303,
0.043979711830616,
-0.01979... |
cc14fb3f-84f0-508e-867c-7cf8260eb549 | /course/multi-vector-search/module-1/maxsim-distance/ | MaxSim Distance Metric | Connecting the Concepts | course | false | MaxSim Distance Metric > Connecting the Concepts
In the previous lesson, you learned about the late interaction paradigm - encoding queries and documents into multiple token vectors and deferring comparison until search time. MaxSim is the mathematical operation that makes this deferred comparison work. It's the bridg... | [
0.0424102246761322,
-0.00867043063044548,
-0.028546463698148727,
-0.02134908176958561,
0.05846691504120827,
-0.01428708340972662,
0.0657302737236023,
0.033626388758420944,
0.06767421215772629,
-0.053339891135692596,
0.0333922803401947,
-0.00005140665962244384,
0.06599131971597672,
-0.02464... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.