@first-pass

14 presentations

Subscribe
LatestPopular
5:55

Average two texts into one input, and a 3B model still ranks both next words

@first-pass · 2 weeks ago

5:48

Kimi Delta Attention can rotate its memory in a single layer

@first-pass · 2 weeks ago

5:58

GPT-6 Astra shows its work in a fifth of Claude Sonnet 5’s steps

@first-pass · 2 weeks ago

5:58

FP8 RL training switches off the penalty on garbled answers

@first-pass · 2 weeks ago

5:56

Sandbagging leaves the right answer readable inside the model

@first-pass · 2 weeks ago

5:10

Self-consistency costs 4.64x the energy split across eight calls

@first-pass · 2 weeks ago

5:21

On-policy distillation punishes the student for stopping

@first-pass · 2 weeks ago

5:13

Grouped Value Attention matches GQA on a 47% smaller KV cache

@first-pass · 2 weeks ago

5:56

NCP-ArchPreview hits OLMo-3-7B loss at 51.3% of the tokens

@first-pass · 1 month ago

5:43

Puro-2B trains 2 billion parameters from scratch on consumer cards

@first-pass · 1 month ago

5:51

Training Leaves Traces: verifying model lineage from weights alone

@first-pass · 1 month ago

5:17

Let's Scale Step by Step — predicting the learning rate for a 155B MoE

@first-pass · 1 month ago

2:41

A 7-point Terminal-Bench jump from 1,200 examples

@first-pass · 1 month ago

2:41

A 7-point Terminal-Bench jump from 1,200 examples

@first-pass · 1 month ago