You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A chat interface using open source models, eg OpenAssistant or Llama. It is a SvelteKit app and it powers the [HuggingChat app on hf.co/chat](https://huggingface.co/chat).
18
18
19
19
0.[No Setup Deploy](#no-setup-deploy)
20
20
1.[Setup](#setup)
21
21
2.[Launch](#launch)
22
-
3.[Extra parameters](#extra-parameters)
23
-
4.[Deploying to a HF Space](#deploying-to-a-hf-space)
24
-
5.[Building](#building)
22
+
3.[Web Search](#web-search)
23
+
4.[Extra parameters](#extra-parameters)
24
+
5.[Deploying to a HF Space](#deploying-to-a-hf-space)
25
+
6.[Building](#building)
25
26
26
27
## No Setup Deploy
27
28
@@ -70,6 +71,16 @@ npm install
70
71
npm run dev
71
72
```
72
73
74
+
## Web Search
75
+
76
+
Chat UI features a powerful Web Search feature. It works by:
77
+
78
+
1. Generating an appropriate Google query from the user prompt.
79
+
2. Performing Google search and extracting content from webpages.
80
+
3. Creating embeddings from texts using [transformers.js](https://huggingface.co/docs/transformers.js). Specifically, using [Xenova/e5-small-v2](https://huggingface.co/Xenova/e5-small-v2) model.
81
+
4. From these embeddings, find the ones that are closest to the user query using vector similarity search. Specifically, we use `inner product` distance.
82
+
5. Get the corresponding texts to those closest embeddings and perform [Retrieval-Augmented Generation](https://huggingface.co/papers/2005.11401) (i.e. expand user prompt by adding those texts so that a LLM can use this information).
0 commit comments