ollama - 💡(How to fix) Fix llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'gemma4' llama_model_load_from_file_impl: failed to load model [1 participants]

What is the issue?

➜ ~ ollama --version Warning: could not connect to a running Ollama instance Warning: client version is 0.20.6 ➜ ~ ollama start time=2026-04-13T23:24:23.633+08:00 level=INFO source=routes.go:1752 msg="server config" env="map[HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:0 OLLAMA_DEBUG:INFO OLLAMA_DEBUG_LOG_REQUESTS:false OLLAMA_EDITOR: OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://127.0.0.1:11434 OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:/Users/ander/.ollama/models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NO_CLOUD:false OLLAMA_NUM_PARALLEL:1 OLLAMA_ORIGINS:[http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://* vscode-file://*] OLLAMA_REMOTES:[ollama.com] OLLAMA_SCHED_SPREAD:false http_proxy: https_proxy: no_proxy:]" time=2026-04-13T23:24:23.634+08:00 level=INFO source=routes.go:1754 msg="Ollama cloud disabled: false" time=2026-04-13T23:24:23.634+08:00 level=INFO source=images.go:499 msg="total blobs: 6" time=2026-04-13T23:24:23.634+08:00 level=INFO source=images.go:506 msg="total unused blobs removed: 0" time=2026-04-13T23:24:23.635+08:00 level=INFO source=routes.go:1810 msg="Listening on 127.0.0.1:11434 (version 0.20.6)" time=2026-04-13T23:24:23.635+08:00 level=INFO source=runner.go:67 msg="discovering available GPUs..." time=2026-04-13T23:24:23.636+08:00 level=INFO source=server.go:444 msg="starting runner" cmd="/Applications/Ollama.app/Contents/Resources/ollama runner --ollama-engine --port 51319" time=2026-04-13T23:24:23.706+08:00 level=INFO source=types.go:42 msg="inference compute" id=0 filter_id=0 library=Metal compute=0.0 name=Metal description="Apple M4" libdirs="" driver=0.0 pci_id="" type=discrete total="17.8 GiB" available="17.8 GiB" time=2026-04-13T23:24:23.706+08:00 level=INFO source=routes.go:1860 msg="vram-based default context" total_vram="17.8 GiB" default_num_ctx=4096 time=2026-04-13T23:24:34.184+08:00 level=WARN source=types.go:977 msg="invalid option provided" option=penalize_newline ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices ggml_metal_library_init: using embedded metal library ggml_metal_library_init: loaded in 0.006 sec ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s) ggml_metal_device_init: GPU name: Apple M4 ggml_metal_device_init: GPU family: MTLGPUFamilyApple9 (1009) ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4 (5002) ggml_metal_device_init: simdgroup reduction = true ggml_metal_device_init: simdgroup matrix mul. = true ggml_metal_device_init: has unified memory = true ggml_metal_device_init: has bfloat = true ggml_metal_device_init: has tensor = false ggml_metal_device_init: use residency sets = true ggml_metal_device_init: use shared buffers = true ggml_metal_device_init: recommendedMaxWorkingSetSize = 19069.67 MB llama_model_load_from_file_impl: using device Metal (Apple M4) (unknown id) - 18185 MiB free llama_model_loader: loaded meta data with 60 key-value pairs and 658 tensors from /Users/ander/.ollama/models/blobs/sha256-b8707e57f676d8dd1b80f623b45200cc92e6966b0e95275e606f412095a49fde (version GGUF V3 (latest)) llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output. llama_model_loader: - kv 0: general.architecture str = gemma4 llama_model_loader: - kv 1: general.type str = model llama_model_loader: - kv 2: general.sampling.top_k i32 = 64 llama_model_loader: - kv 3: general.sampling.top_p f32 = 0.950000 llama_model_loader: - kv 4: general.sampling.temp f32 = 1.000000 llama_model_loader: - kv 5: general.name str = Gemma-4-26B-A4B-It llama_model_loader: - kv 6: general.finetune str = it llama_model_loader: - kv 7: general.basename str = Gemma-4-26B-A4B-It llama_model_loader: - kv 8: general.quantized_by str = Unsloth llama_model_loader: - kv 9: general.size_label str = 26B-A4B llama_model_loader: - kv 10: general.license str = apache-2.0 llama_model_loader: - kv 11: general.license.link str = https://ai.google.dev/gemma/docs/gemm... llama_model_loader: - kv 12: general.repo_url str = https://huggingface.co/unsloth llama_model_loader: - kv 13: general.base_model.count u32 = 1 llama_model_loader: - kv 14: general.base_model.0.name str = Gemma 4 26B A4B It llama_model_loader: - kv 15: general.base_model.0.organization str = Google llama_model_loader: - kv 16: general.base_model.0.repo_url str = https://huggingface.co/google/gemma-4... llama_model_loader: - kv 17: general.tags arr[str,2] = ["unsloth", "image-text-to-text"] llama_model_loader: - kv 18: gemma4.block_count u32 = 30 llama_model_loader: - kv 19: gemma4.context_length u32 = 262144 llama_model_loader: - kv 20: gemma4.embedding_length u32 = 2816 llama_model_loader: - kv 21: gemma4.feed_forward_length u32 = 2112 llama_model_loader: - kv 22: gemma4.attention.head_count u32 = 16 llama_model_loader: - kv 23: gemma4.attention.head_count_kv arr[i32,30] = [8, 8, 8, 8, 8, 2, 8, 8, 8, 8, 8, 2, ... llama_model_loader: - kv 24: gemma4.rope.freq_base f32 = 1000000.000000 llama_model_loader: - kv 25: gemma4.rope.freq_base_swa f32 = 10000.000000 llama_model_loader: - kv 26: gemma4.attention.layer_norm_rms_epsilon f32 = 0.000001 llama_model_loader: - kv 27: gemma4.expert_count u32 = 128 llama_model_loader: - kv 28: gemma4.expert_used_count u32 = 8 llama_model_loader: - kv 29: gemma4.attention.key_length u32 = 512 llama_model_loader: - kv 30: gemma4.attention.value_length u32 = 512 llama_model_loader: - kv 31: gemma4.final_logit_softcapping f32 = 30.000000 llama_model_loader: - kv 32: gemma4.attention.sliding_window u32 = 1024 llama_model_loader: - kv 33: gemma4.attention.shared_kv_layers u32 = 0 llama_model_loader: - kv 34: gemma4.embedding_length_per_layer_input u32 = 0 llama_model_loader: - kv 35: gemma4.attention.sliding_window_pattern arr[bool,30] = [true, true, true, true, true, false,... llama_model_loader: - kv 36: gemma4.attention.key_length_swa u32 = 256 llama_model_loader: - kv 37: gemma4.attention.value_length_swa u32 = 256 llama_model_loader: - kv 38: gemma4.expert_feed_forward_length u32 = 704 llama_model_loader: - kv 39: gemma4.rope.dimension_count u32 = 512 llama_model_loader: - kv 40: gemma4.rope.dimension_count_swa u32 = 256 llama_model_loader: - kv 41: tokenizer.ggml.model str = gemma4 llama_model_loader: - kv 42: tokenizer.ggml.tokens arr[str,262144] = ["<pad>", "<eos>", "<bos>", "<unk>", ... llama_model_loader: - kv 43: tokenizer.ggml.scores arr[f32,262144] = [-1000.000000, -1000.000000, -1000.00... llama_model_loader: - kv 44: tokenizer.ggml.token_type arr[i32,262144] = [3, 1, 3, 3, 3, 1, 1, 1, 1, 1, 1, 1, ... llama_model_loader: - kv 45: tokenizer.ggml.merges arr[str,514906] = ["\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n \n", ... llama_model_loader: - kv 46: tokenizer.ggml.bos_token_id u32 = 2 llama_model_loader: - kv 47: tokenizer.ggml.eos_token_id u32 = 106 llama_model_loader: - kv 48: tokenizer.ggml.unknown_token_id u32 = 3 llama_model_loader: - kv 49: tokenizer.ggml.padding_token_id u32 = 0 llama_model_loader: - kv 50: tokenizer.ggml.mask_token_id u32 = 4 llama_model_loader: - kv 51: tokenizer.chat_template str = {%- macro format_parameters(propertie... llama_model_loader: - kv 52: tokenizer.ggml.add_space_prefix bool = false llama_model_loader: - kv 53: tokenizer.ggml.add_bos_token bool = true llama_model_loader: - kv 54: general.quantization_version u32 = 2 llama_model_loader: - kv 55: general.file_type u32 = 15 llama_model_loader: - kv 56: quantize.imatrix.file str = gemma-4-26B-A4B-it-GGUF/imatrix_unslo... llama_model_loader: - kv 57: quantize.imatrix.dataset str = unsloth_calibration_gemma-4-26B-A4B-i... llama_model_loader: - kv 58: quantize.imatrix.entries_count u32 = 295 llama_model_loader: - kv 59: quantize.imatrix.chunks_count u32 = 141 llama_model_loader: - type f32: 392 tensors llama_model_loader: - type q5_1: 30 tensors llama_model_loader: - type q8_0: 206 tensors llama_model_loader: - type q4_K: 30 tensors print_info: file format = GGUF V3 (latest) print_info: file type = Q4_K - Medium print_info: file size = 15.70 GiB (5.34 BPW) llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'gemma4' llama_model_load_from_file_impl: failed to load model time=2026-04-13T23:24:34.427+08:00 level=INFO source=sched.go:462 msg="failed to create server" model=modelscope.cn/unsloth/gemma-4-26B-A4B-it-GGUF:latest error="unable to load model: /Users/ander/.ollama/models/blobs/sha256-b8707e57f676d8dd1b80f623b45200cc92e6966b0e95275e606f412095a49fde" [GIN] 2026/04/13 - 23:24:34 | 500 | 352.501291ms | 127.0.0.1 | POST "/api/chat" time=2026-04-13T23:24:36.541+08:00 level=WARN source=types.go:977 msg="invalid option provided" option=penalize_newline

Relevant log output

OS

macOS

GPU

Apple

CPU

Apple

Ollama version

0.20.6