Add missing norm.weight parameters to cohere.json

[jukofyork]

This fixes the missing the self_attn.q_norm.weight and self_attn.k_norm.weight parameters, see:

https://huggingface.co/CohereLabs/c4ai-command-r-plus/blob/main/model.safetensors.index.json

{
  "weight_map": {
    "model.embed_tokens.weight": "model-00001-of-00044.safetensors",
    "model.layers.0.input_layernorm.weight": "model-00002-of-00044.safetensors",
    "model.layers.0.mlp.down_proj.weight": "model-00002-of-00044.safetensors",
    "model.layers.0.mlp.gate_proj.weight": "model-00002-of-00044.safetensors",
    "model.layers.0.mlp.up_proj.weight": "model-00002-of-00044.safetensors",
    "model.layers.0.self_attn.k_norm.weight": "model-00002-of-00044.safetensors",
    "model.layers.0.self_attn.k_proj.weight": "model-00002-of-00044.safetensors",
    "model.layers.0.self_attn.o_proj.weight": "model-00002-of-00044.safetensors",
    "model.layers.0.self_attn.q_norm.weight": "model-00002-of-00044.safetensors",
    "model.layers.0.self_attn.q_proj.weight": "model-00002-of-00044.safetensors",
    "model.layers.0.self_attn.v_proj.weight": "model-00002-of-00044.safetensors",
    "...",
}

---

Note

Low Risk
Single architecture JSON metadata change with no runtime logic; only affects which Cohere tensors mergekit recognizes during merges.

Overview
Updates the Cohere mergekit architecture template so per-layer weight lists match real checkpoints (e.g. Command R+).

Adds model.layers.${layer_index}.self_attn.q_norm.weight and model.layers.${layer_index}.self_attn.k_norm.weight to layer_templates, placed with the other self-attention tensors so merges no longer omit these norms.

^{Reviewed by Cursor Bugbot for commit bb241b6. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[jukofyork]

I have read the CLA Document and I hereby sign the CLA