(function() { var utmInheritingDomain = "appstore.com", utmRegExp = /(&|\?)utm_[A-Za-z]+=[A-Za-z0-9]+/gi, links = document.getElementsByTagName("a"), utms = [ "utm_medium={{URL – utm_medium}}", "utm_source={{URL – utm_source}}", "utm_campaign={{URL – utm_campaign}}" ]; for (var index = 0; index < links.length; index += 1) { var tempLink = links[index].href, tempParts; if (tempLink.indexOf(utmInheritingDomain) > 0) { tempLink = tempLink.replace(utmRegExp, ""); tempParts = tempLink.split("#"); if (tempParts[0].indexOf("?") < 0 ) { tempParts[0] += "?" + utms.join("&"); } else { tempParts[0] += "&" + utms.join("&"); } tempLink = tempParts.join("#"); } links[index].href = tempLink; } }());

Use Cases


CUSTOMER STORIES
loading

skt.png

SK Telecom Elevates LLM Operations with Friendli Dedicated Endpoints

PROBLEM

Running and operating the custom LLMs requires long hours and increases operational costs.

SOLUTION

Leverages Friendli Dedicated Endpoints to serve and operate their LLMs.

RESULT

Onboarding within a few hours, 3x cost savings, and 5x increase in throughput.

nacloud.svg

NaCloud: Reducing LLM serving costs for a novel writing service.

PROBLEM

Operating a writing service powered by LLMs

Generative AI powered writing service required serving LLMs.

SOLUTION

Use Friendli Container for LLM serving

Friendli Container enabled our client to use Friendli Engine.

RESULT

Cut LLM serving cost instantly

NaCloud was able to cut GPU serving costs.

upstage.svg

Upstage: Upstage’s Solar LLMs are operated cost-efficiently without any operation burden, thanks to Friendli Dedicated Endpoints.

PROBLEM

Operated LLMs cost-efficiently under varying input traffic

Upstage needed to manage large language model serving efficiently under varying input traffic.

SOLUTION

Use Friendli Dedicated Endpoints for running LLMs

To solve their problem, Upstage decided to utilize Friendli Dedicated Endpoints which is easy to use for operating large language models.

RESULT

Cost-efficient LLM offering without any operational burden

As a result, Upstage was able to serve their propriety large language model without any operation hassle.

scatterlab.png

ScatterLab: Zeta blooms with Friendli Engine

PROBLEM

Quality and size of generative model comes with its own cost

The client company wanted their model to produce real-time responses based on current context, which required 17 times more parameters than the original version.

SOLUTION

Use Friendli Engine for Zeta

Scatter Lab adopted Friendli Engine to serve their model. Friendli was able to handle the real-time executions while reducing the cost and the latency dramatically.

RESULT

Reliable service with much improved efficiency

With Friendli Engine, Zeta had launched successfully and is being used in practice. Its enhanced performance of interactive and creative communication is accepting praises while maintaining the cost and latency of the service.

ncsoft-wide.webp

Integration of Friendli Engine with Amazon Sagemaker Jumpstart

PROBLEM

Serving JumpStart Foundation Models incurs performance and cost challenges

It is challenging to serve JumpStart Foundation Models efficiently in Amazon Sagemaker. The models are computationally heavy, incurring high costs and performance problems.

SOLUTION

Friendli Engine has been integrated with Amazon Sagemaker Jumpstart to serve JumpStart Foundation Models

Friendli Engine can be used with NCSOFT VARCO LLMs. Users of VARCO LLMs enjoy high speed and low cost of serving LLMs.

RESULT

Harness the power of Friendli Engine to serve JumpStart Foundation Models

Users can effortlessly utilize NCSOFT VARCO LLMs on Friendli Engine, resulting in cost reduction within Amazon Sagemaker Jumpstart.

friendliai.webp

Training a Large Language Model (LLM) with Friendli training

PROBLEM

Too much cost for large-scale AI training

Normally, training a large-scale model takes a lot of resources. If you take distributed learning, the burden of faults and loads would only increase.

SOLUTION

Automated and optimized training experience

On Friendli Engine, we could enjoy its special support for distributed learning along with various optimization techniques. Friendli Engine also handled the errors and performance problems to ensure sound training.

RESULT

Made large-scale AI simple

Manipulating Friendli’s automatic matching for state-of-the-art training techniques, training a 13 billion parameter model was felt like a breeze.