NeMo
English
nvidia
steerlm
llama3
zhilinw commited on
Commit
4e7b7a8
1 Parent(s): 0eded9d

Update model_config.yaml

Browse files
Files changed (1) hide show
  1. model_config.yaml +2 -7
model_config.yaml CHANGED
@@ -40,11 +40,6 @@ num_query_groups: 8
40
  tokenizer:
41
  library: huggingface
42
  type: meta-llama/Meta-Llama-3-70B
43
- model: /dataset/models/llama2-13b/llama-tokenizer.model
44
- vocab_file: null
45
- merge_file: null
46
- tokenizer_model: /dataset/models/llama2-13b/llama-tokenizer.model
47
- sentencepiece_legacy: false
48
  use_fast: true
49
  native_amp_init_scale: 4294967296
50
  native_amp_growth_interval: 1000
@@ -194,7 +189,7 @@ nsys_profile:
194
  gen_shape: false
195
  optim:
196
  name: distributed_fused_adam
197
- lr: 1.03e-07
198
  weight_decay: 0.01
199
  betas:
200
  - 0.9
@@ -203,7 +198,7 @@ optim:
203
  name: CosineAnnealing
204
  warmup_steps: 1
205
  constant_steps: 300
206
- min_lr: 1.02e-07
207
  bucket_cap_mb: 200
208
  overlap_grad_sync: false
209
  contiguous_grad_buffer: true
 
40
  tokenizer:
41
  library: huggingface
42
  type: meta-llama/Meta-Llama-3-70B
 
 
 
 
 
43
  use_fast: true
44
  native_amp_init_scale: 4294967296
45
  native_amp_growth_interval: 1000
 
189
  gen_shape: false
190
  optim:
191
  name: distributed_fused_adam
192
+ lr: 1e-07
193
  weight_decay: 0.01
194
  betas:
195
  - 0.9
 
198
  name: CosineAnnealing
199
  warmup_steps: 1
200
  constant_steps: 300
201
+ min_lr: 0.9999e-07
202
  bucket_cap_mb: 200
203
  overlap_grad_sync: false
204
  contiguous_grad_buffer: true