Post summary

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

Answered

Add the .s-post-summary__answered modifier class to indicate that the post has an accepted answer.

<div class="s-post-summary s-post-summary__answered">
    …
</div>

+24 votes

1 answer

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

retrieval-augmented-generation langchain llm vector-database

Bountied

Include the .s-post-summary--stats-bounty element to indicate that the post has a bounty.

<div class="s-post-summary">
    <div class="s-post-summary--stats s-post-summary--sm-hide">
        <div class="s-post-summary--stats-votes">…</div>
        <div class="s-post-summary--stats-answers">…</div>
        <div class="s-post-summary--stats-bounty">
            +50 <span class="v-visible-sr">bounty</span>
        </div>
    </div>
    …
</div>

+24 votes

1 answer

+ 50 bounty

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

+ 50 bounty

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

retrieval-augmented-generation langchain llm vector-database

Ignored

Including an ignored tag will automatically apply custom ignored styling to the post summary.

<div class="s-post-summary">
    …
    <div class="s-post-summary--tags">
        <a class="s-tag s-tag__ignored" href="…">…</a>
        …
    </div>
</div>

+24 votes

1 answer

+ 50 bounty

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

+ 50 bounty

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

retrieval-augmented-generation

Ignored tag

langchain llm vector-database ai

Watched

Including a watched tag will automatically apply custom watched styling to the post summary.

<div class="s-post-summary">
    …
    <div class="s-post-summary--tags">
        <a class="s-tag s-tag__watched" href="…">…</a>
        …
    </div>
</div>

+24 votes

1 answer

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

retrieval-augmented-generation

Watched tag

langchain llm vector-database ai

Deleted

Include the .s-post-summary__deleted modifier class applies custom deleted styling to the post summary.

<div class="s-post-summary s-post-summary__deleted">
    …
</div>

+24 votes

1 answer

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

retrieval-augmented-generation langchain llm vector-database

State badges

Include the appropriate state badge to indicate the current state of the post.

<!-- Draft -->
<div class="s-post-summary">
    <div class="s-post-summary--content">
        <div class="s-post-summary--sm-show">
            <span class="s-badge s-badge__sm s-badge__info">Draft</span>
        </div>
        <div class="s-post-summary--content-meta">
            …
            <span class="s-badge s-badge__info ml-auto s-post-summary--sm-hide">Draft</span>
        </div>
        …
    </div>
</div>

+24 votes

1 answer

Draft

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

Draft

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

retrieval-augmented-generation langchain llm vector-database

+24 votes

1 answer

Review

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

Review

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

retrieval-augmented-generation langchain llm vector-database

+24 votes

1 answer

Closed

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

Closed

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

retrieval-augmented-generation langchain llm vector-database

+24 votes

1 answer

Archived

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

Archived

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

retrieval-augmented-generation langchain llm vector-database

+24 votes

1 answer

Pinned

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

Pinned

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

retrieval-augmented-generation langchain llm vector-database

Content types

Include the appropriate content type badge to indicate the type of content the post represents.

<!-- Announcement -->
<div class="s-post-summary">
    …
    <div class="s-post-summary--content">
        …
        <div class="s-post-summary--tags">
            <a class="s-post-summary--content-type" href="#">Announcement</a>
            …
        </div>
    </div>
</div>

+24 votes

1 answer

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

Announcement

+24 votes

1 answer

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

How-to guide

+24 votes

1 answer

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

Knowledge article

+24 votes

1 answer

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

Policy

Excerpt sizes

Post summaries can be shown without an excerpt or with an excerpt with one, two, or three lines of text. Exclude the excerpt container to hide the excerpt or apply the appropriate truncation class to the excerpt container. See also Truncation.

Classes

Class	Description
`.v-truncate1`	Truncates the excerpt to 1 line of text.
`.v-truncate2`	Truncates the excerpt to 2 lines of text.
`.v-truncate3`	Truncates the excerpt to 3 lines of text.

Examples

<!-- No excerpt -->
<div class="s-post-summary">…</div>

<!-- 1-line excerpt -->
<div class="s-post-summary">
    …
    <p class="s-post-summary--excerpt v-truncate1">…</p>
    …
</div>

<!-- 2-line excerpt -->
<div class="s-post-summary">
    …
    <p class="s-post-summary--excerpt v-truncate2">…</p>
    …
</div>

<!-- 3-line excerpt -->
<div class="s-post-summary">
    …
    <p class="s-post-summary--excerpt v-truncate3">…</p>
    …
</div>

No excerpt

+24 votes

1 answer

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

retrieval-augmented-generation langchain llm vector-database

Small (1 line)

+24 votes

1 answer

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

retrieval-augmented-generation langchain llm vector-database

Medium (2 lines)

+24 votes

1 answer

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

retrieval-augmented-generation langchain llm vector-database

Large (3 lines)

+24 votes

1 answer

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

retrieval-augmented-generation langchain llm vector-database

Small container

Post summaries adapt to their container size. When shown with a container smaller than 448px, the post summary renders with a compact layout.

<div class="s-post-summary">…</div>

+24 votes

1 answer

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

+24 votes

1 answer

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

+24 votes

1 answer

+ 50 bounty

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

1 answer

+ 50 bounty

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

Answers

Answers to a question can be shown in a post summary. Include the .s-post-summary--answers container to show the answers.

For accepted answers, add the .s-post-summary--answer__accepted modifier class and display the Accepted answer text and icon as shown in the example below.

<div class="s-post-summary">
    …
    <div class="s-post-summary--answers">
        <div class="s-post-summary--answer s-post-summary--answer__accepted">
            …
        </div>
        <div class="s-post-summary--answer">
            …
        </div>
    </div>
</div>

+24 votes

2 answers

reputation bling 1

How to reduce hallucinations and improve source relevance in a RAG pipeline?

24

2 answers

98 views

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.

retrieval-augmented-generation langchain llm vector-database

reputation bling 1

answered 1 hour ago

18

Accepted answer

I have built a Retrieval-Augmented Generation (RAG) system using LangChain, a vector database, and an open-source LLM. While it works reasonably well, the model often hallucinates answers or cites sources that are only tangentially related to the user's query. My chunking strategy is set to a chunk size of 1000 tokens, which seems to be the sweet spot for the model.