Expand description
Llama 4 dedicated stack stub — not generic GQA.
Real checkpoints use MoE + a non-generic attention graph (llama.cpp
LLM_ARCH_LLAMA4). Required tensors (names from pinned llama.cpp
LLM_TENSOR_NAMES / llama4.cpp), not implemented here:
token_embd.weight,output_norm.weight,output.weight- Per layer:
blk.{i}.attn_norm.weight,blk.{i}.ffn_norm.weight - MoE FFN:
blk.{i}.ffn_gate_inp.weight,ffn_gate_exps.weight,ffn_up_exps.weight,ffn_down_exps.weight,ffn_exp_probs_b.bias - Llama-4-specific attention projections (not plain GQA
attn_q/attn_k)
Fail-closed via [Self::reject] until a real loader + engine land.