prjkt

11 karma · joined November 8, 2020

submissions comments

prjkt··on Turning GLM-5.3-Flash into a Jev-like decision model

how is Jev cheaper if I can run locally. 0.5% prefill, 0.1% decode, 99.4% cached, latency is <20ms

prjkt··on Plan mode is dead

"This is discussion; no code changes yet."

Press "Tab"

prjkt··on Claude Code sends 33k tokens before reading the prompt; OpenCode sends 7k

It's like Microsoft banning Vim users that use Azure

prjkt··on Nvidia, CoreWeave, and Nebius: Inside the Circular Financing of the GPU Boom

> If there's such a dramatic hole in demand, who are NVIDIA selling their compute to?

NVIDIA itself is also training foundation models (and open-sourcing them). If there is excess compute available, NVIDIA can increase the scale of such models.

prjkt··on AI overly affirms users asking for personal advice

In that case, what tokenizer version? What was the temperature set to? topk? topp? FP32? FP16? Quantized? Hopper? Blackwell?

prjkt··on Counter-Strike: A billion-dollar game built in a dorm room

What is stopping us from all using IRC again ?

prjkt··on DeepSeek open source DeepEP – library for MoE training and Inference

does pytorch count

prjkt··on Why DeepSeek had to be open source

Source:

- Training SW [x]

- Inference SW [x]

- Evaluation SW [x]

- Data [x]

Output:

- Weights []

DeepSeek is closed-source with *open-weights*

prjkt··on Kimi K1.5: Scaling Reinforcement Learning with LLMs

These types of "repositories" should contain some kind of flag/indication that it contains no source code, similar to when a repo is archived