Instella-MoE models in the LocalAI gallery is not supported by any backends

LocalAI version:
docker.io/localai/localai:latest-gpu-hipblas (da7821613d49)

Environment, CPU architecture, OS, and Version:
Linux cachy-server 7.2.0-1-cachyos-server #1 SMP PREEMPT_DYNAMIC Thu, 20 Aug 2026 06:57:13 +0000 x86_64 GNU/Linux

Describe the bug
The new instella-moe model from AMD, added to the gallery in #11308 requires a custom version of llama.cpp. Currently it fails with an error when you try to load it.

To Reproduce

1. On an AMD system, download any variant of instella-moe-16b-a3b-think from the gallery
2. Try to load the model

Expected behavior
Any model in the gallery has a backend in the gallery that can run it

Logs
LocalAI logs

WARN  Backend process exited unexpectedly id="instella-moe-16b-a3b-think-q8" address="127.0.0.1:42565" process="run.sh" exitCode="-1" stderr="0.02.074.860 E srv    load_model: failed to load model, '/models/amd.Instella-MoE-16B-A3B-Think.f16.gguf.Q8_0.gguf'"
ERROR Failed to load model modelID="instella-moe-16b-a3b-think-q8" error=failed to load model with internal loader: could not load model: rpc error: code = Internal desc = Failed to load model: /models/amd.Instella-MoE-16B-A3B-Think.f16.gguf.Q8_0.gguf. Error: llama_model_load: error loading model: unknown model architecture: 'instella-moe'; llama_mode
l_load_from_file_impl: failed to load model; llama_model_load: error loading model: unknown model architecture: 'instella-moe'; llama_model_load_from_file_impl: failed to load model backend="llama-cpp"
ERROR failed to pre-load model model="instella-moe-16b-a3b-think-q8" loaded=[] error=failed to load model with internal loader: could not load model: rpc error: code = Internal desc = Failed to load model: /models/amd.Instella-MoE-16B-A3B-Think.f16.gguf.Q8_0.gguf. Error: llama_model_load: error loading model: unknown model architecture: 'instella-moe'
; llama_model_load_from_file_impl: failed to load model; llama_model_load: error loading model: unknown model architecture: 'instella-moe'; llama_model_load_from_file_impl: failed to load model

Backend logs

+++ realpath run.sh
++ dirname /backends/rocm-llama-cpp/run.sh
+ CURDIR=/backends/rocm-llama-cpp
+ cd /
+ echo 'CPU info:'
CPU info:
+ grep -e 'model\sname' /proc/cpuinfo
+ head -1
model name	: AMD Ryzen 9 7950X3D 16-Core Processor
+ grep -e flags /proc/cpuinfo
+ head -1
flags		: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good amd_lbr_v2 nopl xtopology nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpuid_fault cpb cat_l3 cdp_l3 hw_pstate ssbd mba perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local user_shstk avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic vgif x2avic v_spec_ctrl vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid overflow_recov succor smca fsrm flush_l1d amd_lbr_pmc_freeze
+ BINARY=llama-cpp-fallback
+ '[' -e /backends/rocm-llama-cpp/llama-cpp-cpu-all ']'
+ '[' -n '' ']'
++ uname
+ '[' Linux == Darwin ']'
+ export LD_LIBRARY_PATH=/backends/rocm-llama-cpp/lib:
+ LD_LIBRARY_PATH=/backends/rocm-llama-cpp/lib:
+ '[' -d /backends/rocm-llama-cpp/lib/rocblas/library ']'
+ export ROCBLAS_TENSILE_LIBPATH=/backends/rocm-llama-cpp/lib/rocblas/library
+ ROCBLAS_TENSILE_LIBPATH=/backends/rocm-llama-cpp/lib/rocblas/library
+ '[' -d /backends/rocm-llama-cpp/lib/hipblaslt/library ']'
+ export HIPBLASLT_TENSILE_LIBPATH=/backends/rocm-llama-cpp/lib/hipblaslt/library
+ HIPBLASLT_TENSILE_LIBPATH=/backends/rocm-llama-cpp/lib/hipblaslt/library
+ '[' -e /backends/rocm-llama-cpp/lib/libze_loader.so.1 ']'
+ '[' -f /backends/rocm-llama-cpp/lib/ld.so ']'
+ echo 'Using lib/ld.so'
Using lib/ld.so
Using binary: llama-cpp-fallback
+ echo 'Using binary: llama-cpp-fallback'
+ exec /backends/rocm-llama-cpp/lib/ld.so /backends/rocm-llama-cpp/llama-cpp-fallback --addr 127.0.0.1:36409
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1787506630.819133     666 config.cc:230] gRPC experiments enabled: call_status_override_on_cancellation, event_engine_dns, event_engine_listener, http2_stats_fix, monitoring_experiment, pick_first_new, trace_record_callops, work_serializer_clears_time_cache, work_serializer_dispatch
I0000 00:00:1787506630.819307     666 ev_epoll1_linux.cc:125] grpc epoll fd: 3
I0000 00:00:1787506630.819429     666 server_builder.cc:392] Synchronous server. Num CQs: 1, Min pollers: 1, Max Pollers: 2, CQ timeout (msec): 10000
I0000 00:00:1787506630.820608     666 ev_epoll1_linux.cc:359] grpc epoll fd: 5
I0000 00:00:1787506630.820971     666 tcp_socket_utils.cc:634] TCP_USER_TIMEOUT is available. TCP_USER_TIMEOUT will be used thereafter
Server listening on 127.0.0.1:36409
start_llama_server: starting llama server
start_llama_server: waiting for model to be loaded
0.01.988.564 I system info: n_threads = 16, n_threads_batch = 16, total_threads = 32
0.01.988.567 I 
0.01.988.579 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.01.988.579 I 
0.01.988.618 I srv    load_model: loading model '/models/amd.Instella-MoE-16B-A3B-Think.f16.gguf.Q8_0.gguf'
0.02.035.283 E llama_model_load: error loading model: unknown model architecture: 'instella-moe'
0.02.035.289 E llama_model_load_from_file_impl: failed to load model
0.02.035.303 E common_fit_params: encountered an error while trying to fit params to free device memory: failed to load model
0.02.072.505 E llama_model_load: error loading model: unknown model architecture: 'instella-moe'
0.02.072.507 E llama_model_load_from_file_impl: failed to load model
0.02.072.507 E cmn  common_init_: failed to load model '/models/amd.Instella-MoE-16B-A3B-Think.f16.gguf.Q8_0.gguf'
0.02.072.508 E srv    load_model: failed to load model, '/models/amd.Instella-MoE-16B-A3B-Think.f16.gguf.Q8_0.gguf'

Additional context
The model page on HF links to this fork/branch of llama.cpp with support for instella-moe
https://github.com/csabakecskemeti/llama.cpp/tree/instella-moe

Review the rocm-llama-cpp backend and the gallery addition from #11308; compare the bundled llama.cpp backend with the linked instella-moe branch. Reproduce the failure using an instella-moe-16b-a3b-think variant, then verify that the gallery backend loads it without the unknown architecture error.