unknown model architecture: 'minimax-m3'
Author: B0rnerCreated Jul 10, 2026Updated Jul 10, 2026
Labelsbug
Describe the bug
If I load MiniMax M3 MXFP4 Version I get the error:
unknown model architecture: 'minimax-m3'
I use the current version on linux, pop os.
Is there an existing issue for this?
- I have searched the existing issues
Reproduction
- Download the model, like from here: https://huggingface.co/unsloth/MiniMax-M3-GGUF
- select load model
- -> error
Screenshot
No response
Logs
INFO Loading "MiniMax-M3-MXFP4_MOE-00001-of-00007.gguf"
INFO Using gpu_layers=auto | ctx_size=auto | cache_type=fp16
ggml_cuda_init: found 2 CUDA devices (Total VRAM: 129357 MiB):
Device 0: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes, VRAM: 97249 MiB
Device 1: NVIDIA GeForce RTX 5090, compute capability 12.0, VMM: yes, VRAM: 32107 MiB
load_backend: loaded CUDA backend from /home/app/textgen/installer_files/env/lib/python3.13/site-packages/llama_cpp_binaries/bin/libggml-cuda.so
load_backend: loaded RPC backend from /home/app/textgen/installer_files/env/lib/python3.13/site-packages/llama_cpp_binaries/bin/libggml-rpc.so
load_backend: loaded CPU backend from /home/app/textgen/installer_files/env/lib/python3.13/site-packages/llama_cpp_binaries/bin/libggml-cpu-zen4.so
build_info: b1-8e15485
system_info: n_threads = 24 (n_threads_batch = 24) / 48 | CUDA : ARCHS = 500,610,700,750,800,860,890 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 |
Running without SSL
init: using 47 threads for HTTP server
Web UI is disabled
start: binding port with default address family
main: loading model
common_init_result: fitting params to device memory, for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on
common_params_fit_impl: getting device memory data for initial parameters:
llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'minimax-m3'
llama_model_load_from_file_impl: failed to load model
common_fit_params: encountered an error while trying to fit params to free device memory: failed to load model
common_fit_params: fitting params to free memory took 0.23 seconds
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA RTX PRO 6000 Blackwell Workstation Edition) (0000:11:00.0) - 96676 MiB free
llama_model_load_from_file_impl: using device CUDA1 (NVIDIA GeForce RTX 5090) (0000:e1:00.0) - 30003 MiB free
llama_model_loader: additional 6 GGUFs metadata loaded.
llama_model_loader: loaded meta data with 49 key-value pairs and 948 tensors from user_data/models/MiniMax-M3-MXFP4_MOE-00001-of-00007.gguf (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 = minimax-m3
llama_model_loader: - kv 1: general.type str = model
llama_model_loader: - kv 2: general.name str = Minimax-M3
llama_model_loader: - kv 3: general.basename str = Minimax-M3
llama_model_loader: - kv 4: general.quantized_by str = Unsloth
llama_model_loader: - kv 5: general.size_label str = 128x16B
llama_model_loader: - kv 6: general.license str = other
llama_model_loader: - kv 7: general.license.name str = minimax-community
llama_model_loader: - kv 8: general.license.link str = LICENSE
llama_model_loader: - kv 9: general.repo_url str = https://huggingface.co/unsloth
llama_model_loader: - kv 10: general.tags arr[str,6] = ["multimodal", "moe", "agent", "codin...
llama_model_loader: - kv 11: minimax-m3.block_count u32 = 60
llama_model_loader: - kv 12: minimax-m3.context_length u32 = 1048576
llama_model_loader: - kv 13: minimax-m3.embedding_length u32 = 6144
llama_model_loader: - kv 14: minimax-m3.feed_forward_length u32 = 12288
llama_model_loader: - kv 15: minimax-m3.attention.head_count u32 = 64
llama_model_loader: - kv 16: minimax-m3.attention.head_count_kv u32 = 4
llama_model_loader: - kv 17: minimax-m3.rope.freq_base f32 = 5000000.000000
llama_model_loader: - kv 18: minimax-m3.attention.layer_norm_rms_epsilon f32 = 0.000001
llama_model_loader: - kv 19: minimax-m3.expert_count u32 = 128
llama_model_loader: - kv 20: minimax-m3.expert_used_count u32 = 4
llama_model_loader: - kv 21: minimax-m3.expert_gating_func u32 = 2
llama_model_loader: - kv 22: minimax-m3.attention.key_length u32 = 128
llama_model_loader: - kv 23: minimax-m3.attention.value_length u32 = 128
llama_model_loader: - kv 24: minimax-m3.expert_feed_forward_length u32 = 3072
llama_model_loader: - kv 25: minimax-m3.rope.dimension_count u32 = 64
llama_model_loader: - kv 26: minimax-m3.expert_shared_count u32 = 1
llama_model_loader: - kv 27: minimax-m3.expert_weights_scale f32 = 2.000000
llama_model_loader: - kv 28: minimax-m3.expert_weights_norm bool = true
llama_model_loader: - kv 29: minimax-m3.leading_dense_block_count u32 = 3
llama_model_loader: - kv 30: tokenizer.ggml.model str = gpt2
llama_model_loader: - kv 31: tokenizer.ggml.pre str = minimax-m2
llama_model_loader: - kv 32: tokenizer.ggml.tokens arr[str,200064] = ["Ā", "ā", "Ă", "ă", "Ą", "ą", ...
llama_model_loader: - kv 33: tokenizer.ggml.token_type arr[i32,200064] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv 34: tokenizer.ggml.merges arr[str,199744] = ["Ġ Ġ", "Ġ t", "Ġ a", "i n", "e r...
llama_model_loader: - kv 35: tokenizer.ggml.bos_token_id u32 = 200019
llama_model_loader: - kv 36: tokenizer.ggml.eos_token_id u32 = 200020
llama_model_loader: - kv 37: tokenizer.ggml.unknown_token_id u32 = 16474
llama_model_loader: - kv 38: tokenizer.ggml.padding_token_id u32 = 200000
llama_model_loader: - kv 39: tokenizer.chat_template str = {# ---------- special token variables...
llama_model_loader: - kv 40: general.quantization_version u32 = 2
llama_model_loader: - kv 41: general.file_type u32 = 38
llama_model_loader: - kv 42: quantize.imatrix.file str = MiniMax-M3-GGUF/imatrix_unsloth.gguf
llama_model_loader: - kv 43: quantize.imatrix.dataset str = unsloth_calibration_MiniMax-M3.txt
llama_model_loader: - kv 44: quantize.imatrix.entries_count u32 = 648
llama_model_loader: - kv 45: quantize.imatrix.chunks_count u32 = 182
llama_model_loader: - kv 46: split.no u16 = 0
llama_model_loader: - kv 47: split.tensors.count i32 = 948
llama_model_loader: - kv 48: split.count u16 = 7
llama_model_loader: - type f32: 355 tensors
llama_model_loader: - type q8_0: 422 tensors
llama_model_loader: - type q5_K: 56 tensors
llama_model_loader: - type q6_K: 3 tensors
llama_model_loader: - type mxfp4: 112 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type = MXFP4 MoE
print_info: file size = 238.86 GiB (4.82 BPW)
llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'minimax-m3'
llama_model_load_from_file_impl: failed to load model
common_init_from_params: failed to load model 'user_data/models/MiniMax-M3-MXFP4_MOE-00001-of-00007.gguf'
main: exiting due to model loading error
System Info
using a Linux (pop!os) with an AMD Threadripper, 192 GB RAM, RTX Pro 6000, RTX 5090 - both on current driver with CUDA 13.2
Source: oobabooga/textgen