In modern LLM applications like RAG and Agents, the model is constantly fed new context. For example, in RAG, we retrieve relevant documents and stuff them into the prompt.

The issue is that this dynamically retrieved context doesn't always appear at the beginning of the input sequence. Traditional KV caching only reuses a "common prefix," so if the new information isn't at the very start, the cache hit rate plummets, and your GPU ends up recomputing the same things over and over.

CacheBlend changes the game by allowing for the reuse of pre-computed KV caches regardless of their position in the input sequence.

This makes it possible to achieve a 100% KV Cache hit rate in applications like RAG. The performance gains are significant:

• Faster Time-To-First-Token (TTFT): Get your initial response much quicker.
• More Throughput: Serve significantly more users with the same hardware.
• Almost lossless Output Quality: All of this is achieved with little degradation in the model's generation quality.

CacheBlend works by intelligently handling the two main challenges of reusing non-prefix caches:

• Positional Encoding Update: It efficiently updates positional encodings to ensure the model always knows the correct position of each token, even when we're stitching together cached and new data.
• Selective Attention Recalculation: Instead of recomputing everything, it strategically recalculates only the minimal cross-attention needed between the new and cached chunks to maintain perfect generation quality.

An interactive CacheBlend demo is available at: github.com/LMCache/LMCache-Exa…

LMCache-Examples/demo-rag-blending at main · LMCache/LMCache-Examples

Contribute to LMCache/LMCache-Examples development by creating an account on GitHub.

^GitHub

https://www.reuters.com/legal/government/us-senate-strikes-ai-regulation-ban-trump-megabill-2025-07-01/

The violence last week in Kafr Malik, in the West Bank, comes amid a surge in assaults by Israeli settlers. It also set off a chain of violence in the area.

By Fatima AbdulKarim
July 1, 2025 Updated 1:56 p.m. ET

"The attackers threw another firebomb into the bedroom where Mr. Afeef’s newborn nephew was being lulled to sleep, scorching furniture and leaving blackened marks on the floor and walls, the family said. The damage was visible when Times reporters visited on Friday.
[...]
Soon after, Israeli forces arrived and opened fired at Palestinians instead of stopping the rioters, according to multiple witnesses.

The soldiers killed three people, according to the Palestinian health ministry. [...]Nine others were injured, some gravely..."

archive.ph/j6Mce

https://www.nytimes.com/2025/07/01/world/middleeast/israel-west-bank-palestinians-attack.html

The violence last week in Kafr Malik, in the West Bank, comes amid a surge in assaults by Israeli settlers. It also set off a chain of violence in the area.

By Fatima AbdulKarim
July 1, 2025 Updated 1:56 p.m. ET

"The attackers threw another firebomb into the bedroom where Mr. Afeef’s newborn nephew was being lulled to sleep, scorching furniture and leaving blackened marks on the floor and walls, the family said. The damage was visible when Times reporters visited on Friday.
[...]
Soon after, Israeli forces arrived and opened fired at Palestinians instead of stopping the rioters, according to multiple witnesses.

The soldiers killed three people, according to the Palestinian health ministry. [...]Nine others were injured, some gravely..."

archive.ph/j6Mce

https://www.nytimes.com/2025/07/01/world/middleeast/israel-west-bank-palestinians-attack.html

Republican Alaskan Senator Lisa Murkowski said Tuesday that despite voting in favor of the sweeping tax and spending package, she wants the House to return the "One Big Beautiful Bill" to the Senate for further work.

"My hope is that the House is gonna look at this and recognize that we're not there yet," Murkowski told reporters today.

Murkowski's vote was pivotal in the Senate's razor-thin 51–50 passage of the bill. The Alaska senator had been the focus of intense lobbying by GOP leaders, including Senate Majority Leader John Thune, to secure her support amid concerns over Medicaid cuts and food assistance reductions. The bill now heads to the House, where its future remains uncertain.

Big, Beautiful Bill: Republican Lisa Murkowski Urges House Not to Vote on Bill She Just Approved

Sen. Lisa Murkowski backs Trump's megabill but urges House to return it for fixes, citing Medicaid cuts and tax impacts on Alaska.

^{Amanda Castro (Newsweek)}

In modern LLM applications like RAG and Agents, the model is constantly fed new context. For example, in RAG, we retrieve relevant documents and stuff them into the prompt.

The issue is that this dynamically retrieved context doesn't always appear at the beginning of the input sequence. Traditional KV caching only reuses a "common prefix," so if the new information isn't at the very start, the cache hit rate plummets, and your GPU ends up recomputing the same things over and over.

CacheBlend changes the game by allowing for the reuse of pre-computed KV caches regardless of their position in the input sequence.

This makes it possible to achieve a 100% KV Cache hit rate in applications like RAG. The performance gains are significant:

• Faster Time-To-First-Token (TTFT): Get your initial response much quicker.
• More Throughput: Serve significantly more users with the same hardware.
• Almost lossless Output Quality: All of this is achieved with little degradation in the model's generation quality.

CacheBlend works by intelligently handling the two main challenges of reusing non-prefix caches:

• Positional Encoding Update: It efficiently updates positional encodings to ensure the model always knows the correct position of each token, even when we're stitching together cached and new data.
• Selective Attention Recalculation: Instead of recomputing everything, it strategically recalculates only the minimal cross-attention needed between the new and cached chunks to maintain perfect generation quality.

An interactive CacheBlend demo is available at: github.com/LMCache/LMCache-Exa…

LMCache-Examples/demo-rag-blending at main · LMCache/LMCache-Examples

Contribute to LMCache/LMCache-Examples development by creating an account on GitHub.

^GitHub

https://www.pcmag.com/news/fcc-relaxes-rule-to-allow-for-wider-phone-support-on-t-mobiles-cellular

https://www.politico.com/news/2025/07/01/kirsten-gillibrand-zohran-mamdani-00436031

Moderate 3.8 mi out and back
1,332 ft elevation gain
Hiked 6/3/25

Beginning at the former Yoho NP campground, this short but steep trail takes you to one of the many great hoodoos found along the Canadian Rockies. As the trail crosses multiple washout areas, the trail degrades in places but remains manageable. A split towards the end takes you to different views which are both pretty cool, but both washed out a lot/steep.

The view up at the hoodoos from the lower trail terminus. Trees grow in the more stable sections of the hill.

An upstream view of Hoodoo creek taken from the walking bridge. Mountains may be seen in the distance.

This section of the lower hoodoos trail was very degraded, with atleast one foothold wedged into the slide.

https://peertube.mesnumeriques.fr/w/3S2r5g7a2PFq2gAuVuzWh6