config.json, Dream-org/Dream-v0-Base-7B
repo
Fields
id
sources/arch-hf-config-dreamPublisher
Hugging Face
Kind
repo
Accessed
2026-10-03
notes
28 layers, hidden 3584, intermediate 18944, num_attention_heads 28, num_key_value_heads 4 (so head_dim 128), max_position_embeddings 131072, max_window_layers 28, mask_token_id 151666, rope_theta 1e6, use_sliding_window false. GQA head count is retained even though the model is bidirectional and recomputes rather than caching.
Notes
28 layers, hidden 3584, intermediate 18944, num_attention_heads 28, num_key_value_heads 4 (so head_dim 128), max_position_embeddings 131072, max_window_layers 28, mask_token_id 151666, rope_theta 1e6, use_sliding_window false. GQA head count is retained even though the model is bidirectional and recomputes rather than caching.
Cross-references
This record references no other records.