Data cleaning, custom tokenizer, 167-run speedrun, scaling probes, a 2.67B pretrained on one B200 in 38h, then SFT and reasoning. ~300 W&B runs.
Julián Núñez PRO
juliannunezb
AI & ML interests
None yet
Recent Activity
updated a dataset about 2 months ago
juliannunezb/crisp-v8 updated a collection about 2 months ago
LLM training from scratch: 90M → 2.7B on one GPU updated a collection about 2 months ago
LLM training from scratch: 90M → 2.7B on one GPUOrganizations
None yet
models 18
juliannunezb/crisp3b-v8-ckpts
Text Generation • Updated
juliannunezb/crisp3b-v7-ckpts
Text Generation • Updated
juliannunezb/crisp3b-sft-ckpts
Updated
juliannunezb/crisp3b-reason-ckpts
Updated
juliannunezb/llm-500m-sft-v1
Updated
juliannunezb/llm-500m-pretrain-v1
Updated
juliannunezb/llama-3b-checkpoints
Updated
juliannunezb/local50m-combo3-grpo-bootstrap-v1
Updated
juliannunezb/local50m-combo3-dpo-v1
Updated
juliannunezb/local50m-combo3-sft-v1
Updated
datasets 12
juliannunezb/crisp-v8
Updated • 6
juliannunezb/crisp-reason-cold
Viewer • Updated • 35k • 45
juliannunezb/crisp-sft-chat
Viewer • Updated • 330k • 63
juliannunezb/crisp-165b
Updated • 990
juliannunezb/crisp-128b
Updated • 54
juliannunezb/mixed-pretrain-100b-gpt2
Updated • 163
juliannunezb/midtrain-mix-5b-gpt2
Updated • 74
juliannunezb/ultrafeedback-gpt2-dpo
Viewer • Updated • 62k • 56
juliannunezb/smoltalk-gpt2-sft
Viewer • Updated • 1.04M • 57
juliannunezb/mixed-pretrain-10b-gpt2
Viewer • Updated • 1 • 610 • 1