BluScriptCP / logs /csv /CP_Training_log2_config.txt
unparallelled's picture
Upload folder using huggingface_hub
7625259 verified
Raw History Blame Contribute Delete
544 Bytes
Configuration (bluscriptCP):
d_model: 1536
n_layers: 24
q_heads: 12
kv_heads: 4 (G=3, head_dim=128)
ffn_hidden: 4096
context_length: 4096
vocab_size: 50304 (tying=1)
B: 1
T: 4096
world_size: 4 (dp=2 x cp=2)
CP_ATTN_MODE: ring (load_balance=HeadTail)
global_batch: 8192
grad_accum_steps: 1
Parameters: 681328128
Parameters per GPU: 681328128
Max Learning Rate: 0.0006
Min Learning Rate: 6e-05
max_steps: 1
warmup_steps: 91
checkpointing: 0
run: 2
GPU Memory Used (rank 0): 6068.9 MB / 143155.9 MB