Failed to load Gemma 4

[luiz290788]

I'm trying to load Gemma 4 26B A4B but I'm getting an error:

./SwiftLM --port 8998 --ctx-size 262144 --gpu-layers 99 --model ./unsloth-gemma-4-26b-a4b-it-mlx-8bit --temp 1 --top-k 64 --top-p 0.95 --min-p 0.0

[SwiftLM] Loading model: /Users/luiz/projects/models/unsloth-gemma-4-26b-a4b-it-mlx-8bit
[SwiftLM] Loading from local directory: /Users/luiz/projects/models/unsloth-gemma-4-26b-a4b-it-mlx-8bit
[SwiftLM] ⚠️ Memory strategy: SWAP-ASSISTED (1.5× overcommit, cache limited to 2MB)
[SwiftLM]    Model exceeds RAM by 51%. macOS swap will be used. Expect 2-4× slowdown.
[SwiftLM] --gpu-layers 99 → 99 layers on GPU
[SwiftLM] Loading LLM (large language model)...
Error: Unhandled keys ["experts", "post_feedforward_layernorm_1", "post_feedforward_layernorm_2", "pre_feedforward_layernorm_2", "router"] in language_model.model.layers.0 in Gemma4Model.Gemma4TextModel.Gemma4TextModelInner.Gemma4DecoderLayer

[solderzzc]

Waiting for a tag generated by GitHub action, so we can update for this fix.