TestSubject-1 / README.md
CyanMonkey's picture
step 8800 | ppl=10.75
d53103e verified
|
Raw History Blame Contribute Delete
429 Bytes
metadata
language: en
tags:
  - causal-lm
  - gqa
  - qk-norm
  - rope
  - swiglu
  - refresh-gate
  - rose-x1
license: apache-2.0

TestSubject-1

Rose X1.5 SLM (148.8M params).

Architecture

Property Value
Layers 30
Hidden 640
Heads 10 (kv=5)
QK Norm True
XSA True
Refresh Gate [9, 19]
Params 148.826M

Training

  • Tokens: 4,611,481,600
  • Val PPL: 10.75
  • Optimizer: muon_adamw