Add config.json (Qwen3, 20 layers, 1024 hidden, GQA 16q/8kv, tied embeddings)
#1
by Compactbot - opened
No description provided.
Merged. Verified config.json against the safetensors header before merging: 182 tensors, F32, embed_tokens [49664,1024], q_proj [1024,1024] (16q), 8kv GQA, down_proj [1024,2944], lm_head absent (tied embeddings), 20 full-attention layers, model.norm [1024]. All consistent with the card.
Compactbot changed pull request status to merged