morphllm's profile picture. The Fastest, Flawless AI Code Edits.
Better coding agents in 8 lines. 10,500 tok/sec.
http://morphllm.com | http://discord.gg/AdXta4yxEK

Morph

@morphllm

The Fastest, Flawless AI Code Edits. Better coding agents in 8 lines. 10,500 tok/sec. http://morphllm.com | http://discord.gg/AdXta4yxEK

置頂
morphllm's tweet image.

divorce someone who says:

morphllm's tweet image. divorce someone who says:

marry someone who says: - i love you - i'm proud of you - i'm grateful for you - yes you can buy another domain - no 3 new domains isn't too many - i wish we could afford more domains - should we buy another domain?



before and after fast apply

morphllm's tweet image. before and after fast apply
morphllm's tweet image. before and after fast apply

the latency of most legit Morph requests is now faster than throwing a model not found error when the model name is invalid


Morph fast apply at pass 1: 98% Search and replace takes 40% longer on average

string_replace is god awful in cursor, plz fix i've literally never had it be actually useful - agent just thinks a bunch then fails to do anything

aidenybai's tweet image. string_replace is god awful in cursor, plz fix

i've literally never had it be actually useful - agent just thinks a bunch then fails to do anything


codegen is the defacto non text way models will express themselves


Morph is accelerating faster than any inference engine in history. One month ago: 4,200 tokens per second. Today: 10,000 tokens per second. This is just the beginning


Let’s be clear. apply() loops and diff timeouts are not solutions. They’re bandaids on a broken foundation. You’ve seen the screenshots: “Thought for 1s. Edit failed because of a diff timeout.” “String to replace not found” Morph was engineered differently. It doesn’t choke…

morphllm's tweet image. Let’s be clear.  

apply() loops and diff timeouts are not solutions. They’re bandaids on a broken foundation. You’ve seen the screenshots: “Thought for 1s. Edit failed because of a diff timeout.” “String to replace not found” 

Morph was engineered differently. It doesn’t choke…

🔥 gpus do indeed go brrrrrr

morphllm's tweet image. 🔥 gpus do indeed go brrrrrr

Morph 已轉發

We're hitting 10,542 tokens/sec per request on Nvidia B200s Batch size=1. P95 latency: 384ms. 8k input/8k output. Single-request throughput—no batching tricks.

tejasybhakta's tweet image. We're hitting 10,542 tokens/sec per request on Nvidia B200s
Batch size=1. P95 latency: 384ms.
8k input/8k output. Single-request throughput—no batching tricks.

Morph 已轉發

We added support for @morphllm to our Fragments repo from @e2b. Morph's LLM is specialized fast apply and codegen model. Instead of generating the whole file from scratch every time you want to change something, you can just use Morph to generate the actual changes and have…


The best sandboxes deserve the best code gen experience. The e2b product and developer experience is amazing and we’re thrilled to work together!

We added support for @morphllm to our Fragments repo from @e2b. Morph's LLM is specialized fast apply and codegen model. Instead of generating the whole file from scratch every time you want to change something, you can just use Morph to generate the actual changes and have…



United States 趨勢

Loading...

Something went wrong.


Something went wrong.