Text Generation
Safetensors
nemotron
pretrained
openllm-france
oligou commited on
Commit
52fd60f
·
verified ·
1 Parent(s): 50c5fef

Update README.md

Browse files

fix RoPE theta for context extension

Files changed (1) hide show
  1. README.md +1 -1
README.md CHANGED
@@ -275,7 +275,7 @@ The details of the intitial pretraining phase are listed below. For each subsequ
275
  | Total \# steps | 5,960 |
276
  | Context length | 32,768 |
277
  | Batch size | 128 |
278
- | RoPE theta | 1,000,000 |
279
  | Learning rate schedule | Warmup + Cosine annealing |
280
  | Maximum Learning rate | 6.87e-5 |
281
  | Final Learning rate | 0 |
 
275
  | Total \# steps | 5,960 |
276
  | Context length | 32,768 |
277
  | Batch size | 128 |
278
+ | RoPE theta | 500,000 |
279
  | Learning rate schedule | Warmup + Cosine annealing |
280
  | Maximum Learning rate | 6.87e-5 |
281
  | Final Learning rate | 0 |