chi (@chimcis) header
chi (@chimcis) profile picture

chi

@chimcis

Inference @primeintellect, cs @carleton_u

 ottawa / sf
chimcisaac.co
Joined March 2025
  • Tweets1.2K
  • Following1.4K
  • Followers1K
23Media
Photo by chi (@chimcis)Photo by chi (@chimcis)Photo by chi (@chimcis)Photo by chi (@chimcis)Photo by chi (@chimcis)Photo by chi (@chimcis)Photo by chi (@chimcis)Photo by chi (@chimcis)Photo by chi (@chimcis)Photo by chi (@chimcis)
At 96.9%, Prime Inference has the highest cache hit rate of any provider serving Nemotron 3.5 Super VL. This brings the effective input cost to $0.024/M tokens, the lowest of any provider, and ~70% below the largest provider.
Image from Prime Intellect (@PrimeIntellect)
12
14
2
164
14,282
🚀 0 → 120 BILLION tokens/day in just ONE month! Just one month since onboarding Prime Inference to OpenRouter, we've scaled from zero to 120B tokens served daily. Note this is just a small portion of Prime Inference 🔥 openrouter.ai/provider/prime…
Image from Stu Cao (@caoshuxun)
14
19
7
141
24,356
I'm doing DP. I'm doing TP. I'm doing SP. I'm doing PP. I'm doing ZP. I'm doing FSDP. I'm doing 1F1B. I'm doing CP. I'm doing ETP. I'm doing DPA. I'm doing VPP. I'm doing P/D. I'm doing ZBPP. I'm doing DPA. I'm doing HSDP. I'm doing DCP. I'm doing DWDP. I'm doing AFD.
45
35
11
692
60,356
GLM 5.3 is now 30% cheaper on Prime Inference, and also on OpenRouter when you pick Prime Intellect as the provider. We offer not only a top-tier speed but more importantly highly reliable tool call.
Image from Stu Cao (@caoshuxun)
1
4
30
817
"We Bought the Whole GPU, So We're Damn Well Going to Use the Whole GPU"
Flapping Airplanes (@flappyairplanes) profile pictureFlapping Airplanes
@flappyairplanesOct 21
Replying to @flappyairplanes
We’ve developed the world’s most complete megakernel framework. It can perform entire training runs, serve inference, play video games, and even do all three simultaneously in a single kernel.
89
4,588
Today we're releasing prime-rl v0.10.0 and verifiers v0.4.0 We added our new RL algorithm IPO as the default, tailored for async RL stability, as well as performance improvements and new model support
Image from Prime Intellect (@PrimeIntellect)
14
40
14
519
47,187
chi retweeted
BetaKit (@BetaKit) profile picture
BetaKit
@BetaKitOct 19
Exclusive: Toronto and NYC’s @MeckaAI has recruited the growth hackers behind Ottawa-based “IRL sidequest” app @thirdspaceSo to help gather home video for training real-world robots. #CDNtech betakit.com/mecka-ai-recruit…
Mecka AI recruits growth hackers behind “IRL sidequest” app Thirdspace

Mecka AI recruits growth hackers behind “IRL sidequest” app Thirdspace

Toronto- and New York City-based Mecka AI, which aims to supply the training data to deploy robots in the real world, has completed its second acquisition.

betakit.com
15
13
3
60
7,445
DeepSeek-v4.1-Flash-1017 is now available on Prime Inference
Image from Stu Cao (@caoshuxun)
6
9
63
4,086
Prime Intellect has acquired Moondream We’re bringing Moondream’s open vision models and inference optimizations into our stack Train and deploy multimodal models across vision, video, voice, decision models and computer use
101
81
57
1,262
148,406
chat is this real
Prime Intellect (@PrimeIntellect) profile picturePrime Intellect
@PrimeIntellectOct 13
Prime Intellect has acquired Moondream We’re bringing Moondream’s open vision models and inference optimizations into our stack Train and deploy multimodal models across vision, video, voice, decision models and computer use
1
48
980
Introducing Prime Inference: We've served trillions of tokens for RL and dedicated customer deployments To own your intelligence, you need to own your inference Unpacking our inference stack
90
119
53
1,336
235,918
Inference is here
Prime Intellect (@PrimeIntellect) profile picturePrime Intellect
@PrimeIntellectOct 2
Introducing Prime Inference: We've served trillions of tokens for RL and dedicated customer deployments To own your intelligence, you need to own your inference Unpacking our inference stack
13
1
91
4,170
We’re hiring a full-time engineer to help build Prime Agent :) Own features end to end across agent harnesses, cloud agents, and self-improving loops -- integrated with our sandboxes, evals, and hosted training stack. Full-time, in person in SF: primeintellect.ai/careers/85…
Member of Technical Staff - Prime Agent | Prime Intellect

Member of Technical Staff - Prime Agent | Prime Intellect

Help build the open superintelligence stack. Explore careers in research, engineering, and the teams bringing open AI to the world.

primeintellect.ai
27
12
3
292
16,130
Introducing Prime Sandboxes: MicroVM sandboxes purpose-built for RL training. Model training requires running tens of thousands of concurrent sandboxes, leading to complex and costly configuration. We built Prime Sandboxes for our own team. Today we're releasing them publicly.
41
79
31
746
120,915
Last week, I joined @PrimeIntellect as an intern to help optimize their inference stack :) Very grateful to get the chance to work with such a talented team of researchers and engineers Plenty to do and much to learn, super excited for what’s ahead!
Image from chi (@chimcis)
76
12
4
348
19,642
today's my last day @cloudflare i was given the opportunity to work on the Workers AI team this summer, huge shoutout to @kevin_flansburg and @michellechen for taking me under their wing and letting me work on some very exciting problems around optimizing and improving their entire inference and model serving stack! very grateful for all the flexibility i was given to work on whatever was most interesting to me and the time i had to close some of the gaps in my inexperience i also got the chance to work with many talented engineers and researchers on this small (but very mighty!) team: Alex, Xiang, @KastanDay it was a great run! i'll still be in SF this upcoming fall, but no longer at Cloudflare - more news to come :)
31
3
160
7,898
Introducing Prime Flash MoE - a set of Blackwell-optimized CUDA kernels for MoE inference. Fuses routing-aware GEMMs, SwiGLU, quantization & reductions without materializing intermediates. BF16 + MXFP8 paths, benchmarked on B200s. primeintellect.ai/blog/prime…
Prime Flash MoE - Faster MoE Kernels optimized for Blackwell

Prime Flash MoE - Faster MoE Kernels optimized for Blackwell

Prime Flash MoE is a set of Blackwell-optimized CUDA kernels for the MoE feed-forward pass that keep the intermediate activation on-chip — up to 2.4× faster than PyTorch grouped GEMM, with bf16 and...

primeintellect.ai
18
52
10
556
44,286
Given all of these neolabs and neoclouds and neochips the real contrarian opportunity is to start a paleolab and a paleocloud and a paleochip company
4
1
69
3,511
safe and fast models!!!! go check it out on workers ai - has been a fantastic experience getting to work on some of these problems
michelle (@michellechen) profile picturemichelle
@michellechenAug 3
happy agents week - kicking it off with another look behind the scenes of how we run trillion parameter models at scale. featuring incredible MLE work from our team (hiring!) and some advancements for kv cache protections blog.cloudflare.com/smaller-…
Smaller, faster, safer: running Kimi and GLM at scale

Smaller, faster, safer: running Kimi and GLM at scale

Serving frontier models like Kimi and GLM means fighting for GPU memory. Here's how we quantize KV caches, compress model weights, and add integrity checks to serve them faster, cheaper, and safely.

blog.cloudflare.com
4
1
17
2,902
super cool work by ben!
ben (@benbye) profile pictureben
@benbyeJul 14
I built LeMario from scratch, a JEPA world model trained directly on 737K Super Mario frames + controller inputs, with no rewards. It predicts future states and uses CEM to control Mario over short horizons. Based on LeWorldModel by @lucasmaes_. Technical write-up below
6
1,985
Load more