LFM2.5-ColBERT Filters 151 Tools to Top Five for Faster LLM Calls · Digg

/Tech1h ago

LFM2.5-ColBERT Filters 151 Tools to Top Five for Faster LLM Calls

345965540534.3K

OK#200|@LATEINTERACTION

Original post

Liquid AI@liquidai

Storing too many tools in your context window increases latency and can lead to wrong tool selection.

In this demo, we used LFM2.5-ColBERT-350M as a filter to only select the five most relevant tools among 151 options.

It's fast and reliable, even without any specific fine-tuning.

Try the demo on @huggingface! https://huggingface.co/spaces/LiquidAI/colbert-tool-selection

And learn more on our blog: https://www.liquid.ai/blog/lfm2-5-retrievers

1:55 PM · Jun 19, 2026 · 20.2K Views

Sentiment

Users praise LFM2.5-ColBERT for efficiently narrowing 151 tools to the top five because it cuts latency, avoids context bloat, and fixes bottlenecks in agent workflows without fine-tuning.

Pos

93.7%

Neg

6.3%

14 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Related links

colbert tool selection

HUGGINGFACE.COVia

lfm 2 5 retrievers

LIQUID.AI

Via

Posts from X

Most Activity

Most Activity

VIEWS518

Ronald Mannak@ronaldmannak

@liquidai Why LFM2.5-ColBERT-350M and not LFM2.5-Embedding-350M? Isn’t a single vector good enough for tool selection? It’s simpler to implement too I assume.

16h|Views 518Likes 1

LIKES3

Antoine Chaffin@antoine_chaffin

@maximelabonne @iacopo_poli

1d|Views 458Likes 3

RETWEETS29

Maxime Labonne@maximelabonne

LFM2.5-ColBERT-350M is a surprisingly reliable smart tool selector.

We gave it 151 tools, and it consistently surfaces the 5 most relevant ones based on the user prompt.

This saves tokens and improves accuracy. Ideal for hmmmm agentic edge models? 👀

1d|Views 14.5KLikes 256Bookmarks 210

Brian Cheong@briancheong

@maximelabonne Small tool selectors feel like the boring infrastructure agents actually need. Less context stuffing, fewer weird tool calls.

1d|Views 271Likes 2

Noctus@noctus91

@maximelabonne 🤔 interesting

1d|Views 200Likes 2

Mike Frager ⧉@fragermk

@maximelabonne @mlech26l Cool 😎 Please drop code!

1d|Views 197Likes 2

Trung Le@trung_rta

@liquidai Tiny but powerful, that’s what I usually describe LFMs

16h|Views 328Likes 1

Quan Nguyen@stablequan

@maximelabonne are you guys planning on having a ColPali-like version on LFM-2.5 VL?

1d|Views 279Likes 1

Alina Fomina@fominaaalina

@maximelabonne agentic edge is the right framing. we tested this as well:

1d|Views 277Likes 1

𝑘𝑒𝑟𝑛𝑒𝑙𝑡𝑟𝑖𝑐𝑘@kernel_trick

@maximelabonne so cool!

1d|Views 222Likes 1

The Canaanite@mysticaltech

@maximelabonne Extraordinary these colbert models.

1d|Views 172Likes 1

Shreyas Karnik@KarnikShreyas

@maximelabonne Very cool

1d|Views 142Likes 1

mystic@gurshabad90

@liquidai W

16h|Views 264

Owlfy.ai@Owlfy_ai

@maximelabonne Totally agree—tool retrieval bottlenecks fly under the radar. Packing 151 tools into context just isn’t scalable. Smart to see a lightweight selector narrowing things down before the agent goes to work.

1d|Views 75Likes 1

Eddie,Jesse Bailey@SwaggaBros1234

@liquidai curious how tight the correlation is between colbert rank and tool correctness. feels like a good spot to find weird model behaviors and test some reranking ideas.

13h|Views 195

Dwid Lee@DwidLee

@maximelabonne it will break the prompt cache

1d|Views 111

Alice The Ai Expert@AliceInfoAi

@liquidai Smarter tool selection, zero latency bloat

8h|Views 93

Xianliang Wu@XianliangWu

@maximelabonne @zleapai 请问这个模型是否可以用来组装 workspace?

1d|Views 87

Satyam Kumar ☄️@STUD_MAN_X

@liquidai How can we test with the API of the LFM 2.5 8B model?

8h|Views 67

Samian@ApplyWiseAi

@liquidai 151 tools is wild. filtering down to 5 keeps things snappy. how does it score ties when tool descriptions overlap hard

15h|Views 53

Load more posts

345965540534.3K