github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 16:40:08 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #15747] Error loading model architecture - unknown model architecture: 'qwen35moe' #72097

New Issue
Open
opened 2026-05-05 03:28:19 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-05-05 03:28:19 -05:00
Owner
Copy Link

Originally created by @myocrude on GitHub (Apr 22, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15747

What is the issue?

Environment:

  • Ollama version: 0.21.0
  • OS: CachyOS / Arch-based Linux
  • GPU: NVIDIA (attached nvidia-smi output)
  • Open WebUI is not required to reproduce

Model:
hf.co/Youssofal/Qwen3.6-35B-A3B-Abliterated-Heretic-GGUF:Q4_K_M

Steps to reproduce:

  1. Install/update Ollama to 0.21.0
  2. Run:
    ollama pull hf.co/Youssofal/Qwen3.6-35B-A3B-Abliterated-Heretic-GGUF:Q4_K_M
  3. Run:
    ollama run hf.co/Youssofal/Qwen3.6-35B-A3B-Abliterated-Heretic-GGUF:Q4_K_M

Expected behavior:
Model should load or fail with a clearer compatibility message.

Actual behavior:
Returns 500 error and journal shows:
unknown model architecture: 'qwen35moe'

Notes:

  • Download succeeds
  • Blob is read successfully
  • Failure happens during model architecture loading
  • Reproduced outside Open WebUI via local API / CLI

After Restart - Log.txt
GPU Info.txt
Ollama Service Status.txt
ollama-journal-boot.txt
OS Info.txt

Relevant log output

description="NVIDIA GeForce RTX 4070 SUPER" libdirs=ollama,cuda_v13 driver=13.2 pci_id=0000:01:00.0 type=discrete total="12.0 GiB" available="10.6 GiB"
abr 22 10:44:10 CachyOS ollama[51669]: time=2026-04-22T10:44:10.720-03:00 level=INFO source=routes.go:1860 msg="vram-based default context" total_vram="12.0 GiB" default_num_ctx=4096
abr 22 10:44:26 CachyOS ollama[51669]: [GIN] 2026/04/22 - 10:44:26 | 200 |      30.759µs |       127.0.0.1 | HEAD     "/"
abr 22 10:44:26 CachyOS ollama[51669]: [GIN] 2026/04/22 - 10:44:26 | 200 |  114.907452ms |       127.0.0.1 | POST     "/api/show"
abr 22 10:44:27 CachyOS ollama[51669]: [GIN] 2026/04/22 - 10:44:27 | 200 |  112.800755ms |       127.0.0.1 | POST     "/api/show"
abr 22 10:44:27 CachyOS ollama[51669]: time=2026-04-22T10:44:27.136-03:00 level=INFO source=server.go:444 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --port 37437"
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: loaded meta data with 47 key-value pairs and 733 tensors from /usr/share/ollama/.ollama/models/blobs/sha256-ae2fb73ac0da875640269f1e65e9c7fb415b066c6d544c3eef9adb0d03f04792 (version GGUF V3 (latest))
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv   0:                       general.architecture str              = qwen35moe
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv   1:                               general.type str              = model
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv   2:                     general.sampling.top_k i32              = 20
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv   3:                     general.sampling.top_p f32              = 0.950000
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv   4:                      general.sampling.temp f32              = 1.000000
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv   5:                               general.name str              = Qwen3.6 35B A3B Abliterated Heretic BF16
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv   6:                           general.finetune str              = Abliterated-Heretic
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv   7:                           general.basename str              = Qwen3.6
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv   8:                         general.size_label str              = 35B-A3B
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv   9:                   general.base_model.count u32              = 1
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  10:                  general.base_model.0.name str              = Qwen3.6 35B A3B
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  11:          general.base_model.0.organization str              = Qwen
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  12:              general.base_model.0.repo_url str              = https://huggingface.co/Qwen/Qwen3.6-3...
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  13:                               general.tags arr[str,15]      = ["qwen", "qwen3.6", "qwen3_5_moe", "m...
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  14:                      qwen35moe.block_count u32              = 40
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  15:                   qwen35moe.context_length u32              = 262144
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  16:                 qwen35moe.embedding_length u32              = 2048
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  17:             qwen35moe.attention.head_count u32              = 16
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  18:          qwen35moe.attention.head_count_kv u32              = 2
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  19:          qwen35moe.rope.dimension_sections arr[i32,4]       = [11, 11, 10, 0]
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  20:                   qwen35moe.rope.freq_base f32              = 10000000.000000
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  21: qwen35moe.attention.layer_norm_rms_epsilon f32              = 0.000001
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  22:                     qwen35moe.expert_count u32              = 256
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  23:                qwen35moe.expert_used_count u32              = 8
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  24:             qwen35moe.attention.key_length u32              = 256
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  25:           qwen35moe.attention.value_length u32              = 256
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  26:       qwen35moe.expert_feed_forward_length u32              = 512
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  27: qwen35moe.expert_shared_feed_forward_length u32              = 512
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  28:                  qwen35moe.ssm.conv_kernel u32              = 4
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  29:                   qwen35moe.ssm.state_size u32              = 128
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  30:                  qwen35moe.ssm.group_count u32              = 16
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  31:               qwen35moe.ssm.time_step_rank u32              = 32
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  32:                   qwen35moe.ssm.inner_size u32              = 4096
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  33:          qwen35moe.full_attention_interval u32              = 4
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  34:             qwen35moe.rope.dimension_count u32              = 64
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  35:                       tokenizer.ggml.model str              = gpt2
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  36:                         tokenizer.ggml.pre str              = qwen35
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  37:                      tokenizer.ggml.tokens arr[str,248320]  = ["!", "\"", "#", "$", "%", "&", "'", ...
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  38:                  tokenizer.ggml.token_type arr[i32,248320]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  39:                      tokenizer.ggml.merges arr[str,247587]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  40:                tokenizer.ggml.eos_token_id u32              = 248046
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  41:            tokenizer.ggml.padding_token_id u32              = 248044
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  42:                tokenizer.ggml.bos_token_id u32              = 248044
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  43:                    tokenizer.chat_template str              = {%- set image_count = namespace(value...
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  44:               general.quantization_version u32              = 2
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  45:                          general.file_type u32              = 15
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv  46:                            general.license str              = apache-2.0
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - type  f32:  301 tensors
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - type q4_K:  371 tensors
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - type q6_K:   61 tensors
abr 22 10:44:27 CachyOS ollama[51669]: print_info: file format = GGUF V3 (latest)
abr 22 10:44:27 CachyOS ollama[51669]: print_info: file type   = Q4_K - Medium
abr 22 10:44:27 CachyOS ollama[51669]: print_info: file size   = 19.70 GiB (4.88 BPW)
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'qwen35moe'
abr 22 10:44:27 CachyOS ollama[51669]: llama_model_load_from_file_impl: failed to load model
abr 22 10:44:27 CachyOS ollama[51669]: time=2026-04-22T10:44:27.355-03:00 level=INFO source=sched.go:462 msg="failed to create server" model=hf.co/Youssofal/Qwen3.6-35B-A3B-Abliterated-Heretic-GGUF:Q4_K_M error="unable to load model: /usr/share/ollama/.ollama/models/blobs/sha256-ae2fb73ac0da875640269f1e65e9c7fb415b066c6d544c3eef9adb0d03f04792"
abr 22 10:44:27 CachyOS ollama[51669]: [GIN] 2026/04/22 - 10:44:27 | 500 |   352.19334ms |       127.0.0.1 | POST     "/api/generate"

OS

Linux

GPU

Nvidia

CPU

AMD

Ollama version

ollama version is 0.21.0

Originally created by @myocrude on GitHub (Apr 22, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15747 ### What is the issue? Environment: - Ollama version: 0.21.0 - OS: CachyOS / Arch-based Linux - GPU: NVIDIA (attached nvidia-smi output) - Open WebUI is not required to reproduce Model: hf.co/Youssofal/Qwen3.6-35B-A3B-Abliterated-Heretic-GGUF:Q4_K_M Steps to reproduce: 1. Install/update Ollama to 0.21.0 2. Run: ollama pull hf.co/Youssofal/Qwen3.6-35B-A3B-Abliterated-Heretic-GGUF:Q4_K_M 3. Run: ollama run hf.co/Youssofal/Qwen3.6-35B-A3B-Abliterated-Heretic-GGUF:Q4_K_M Expected behavior: Model should load or fail with a clearer compatibility message. Actual behavior: Returns 500 error and journal shows: unknown model architecture: 'qwen35moe' Notes: - Download succeeds - Blob is read successfully - Failure happens during model architecture loading - Reproduced outside Open WebUI via local API / CLI [After Restart - Log.txt](https://github.com/user-attachments/files/26971983/After.Restart.-.Log.txt) [GPU Info.txt](https://github.com/user-attachments/files/26971984/GPU.Info.txt) [Ollama Service Status.txt](https://github.com/user-attachments/files/26971985/Ollama.Service.Status.txt) [ollama-journal-boot.txt](https://github.com/user-attachments/files/26971986/ollama-journal-boot.txt) [OS Info.txt](https://github.com/user-attachments/files/26971987/OS.Info.txt) ### Relevant log output ```shell description="NVIDIA GeForce RTX 4070 SUPER" libdirs=ollama,cuda_v13 driver=13.2 pci_id=0000:01:00.0 type=discrete total="12.0 GiB" available="10.6 GiB" abr 22 10:44:10 CachyOS ollama[51669]: time=2026-04-22T10:44:10.720-03:00 level=INFO source=routes.go:1860 msg="vram-based default context" total_vram="12.0 GiB" default_num_ctx=4096 abr 22 10:44:26 CachyOS ollama[51669]: [GIN] 2026/04/22 - 10:44:26 | 200 | 30.759µs | 127.0.0.1 | HEAD "/" abr 22 10:44:26 CachyOS ollama[51669]: [GIN] 2026/04/22 - 10:44:26 | 200 | 114.907452ms | 127.0.0.1 | POST "/api/show" abr 22 10:44:27 CachyOS ollama[51669]: [GIN] 2026/04/22 - 10:44:27 | 200 | 112.800755ms | 127.0.0.1 | POST "/api/show" abr 22 10:44:27 CachyOS ollama[51669]: time=2026-04-22T10:44:27.136-03:00 level=INFO source=server.go:444 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --port 37437" abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: loaded meta data with 47 key-value pairs and 733 tensors from /usr/share/ollama/.ollama/models/blobs/sha256-ae2fb73ac0da875640269f1e65e9c7fb415b066c6d544c3eef9adb0d03f04792 (version GGUF V3 (latest)) abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output. abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 0: general.architecture str = qwen35moe abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 1: general.type str = model abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 2: general.sampling.top_k i32 = 20 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 3: general.sampling.top_p f32 = 0.950000 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 4: general.sampling.temp f32 = 1.000000 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 5: general.name str = Qwen3.6 35B A3B Abliterated Heretic BF16 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 6: general.finetune str = Abliterated-Heretic abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 7: general.basename str = Qwen3.6 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 8: general.size_label str = 35B-A3B abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 9: general.base_model.count u32 = 1 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 10: general.base_model.0.name str = Qwen3.6 35B A3B abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 11: general.base_model.0.organization str = Qwen abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 12: general.base_model.0.repo_url str = https://huggingface.co/Qwen/Qwen3.6-3... abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 13: general.tags arr[str,15] = ["qwen", "qwen3.6", "qwen3_5_moe", "m... abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 14: qwen35moe.block_count u32 = 40 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 15: qwen35moe.context_length u32 = 262144 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 16: qwen35moe.embedding_length u32 = 2048 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 17: qwen35moe.attention.head_count u32 = 16 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 18: qwen35moe.attention.head_count_kv u32 = 2 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 19: qwen35moe.rope.dimension_sections arr[i32,4] = [11, 11, 10, 0] abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 20: qwen35moe.rope.freq_base f32 = 10000000.000000 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 21: qwen35moe.attention.layer_norm_rms_epsilon f32 = 0.000001 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 22: qwen35moe.expert_count u32 = 256 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 23: qwen35moe.expert_used_count u32 = 8 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 24: qwen35moe.attention.key_length u32 = 256 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 25: qwen35moe.attention.value_length u32 = 256 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 26: qwen35moe.expert_feed_forward_length u32 = 512 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 27: qwen35moe.expert_shared_feed_forward_length u32 = 512 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 28: qwen35moe.ssm.conv_kernel u32 = 4 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 29: qwen35moe.ssm.state_size u32 = 128 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 30: qwen35moe.ssm.group_count u32 = 16 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 31: qwen35moe.ssm.time_step_rank u32 = 32 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 32: qwen35moe.ssm.inner_size u32 = 4096 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 33: qwen35moe.full_attention_interval u32 = 4 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 34: qwen35moe.rope.dimension_count u32 = 64 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 35: tokenizer.ggml.model str = gpt2 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 36: tokenizer.ggml.pre str = qwen35 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 37: tokenizer.ggml.tokens arr[str,248320] = ["!", "\"", "#", "$", "%", "&", "'", ... abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 38: tokenizer.ggml.token_type arr[i32,248320] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ... abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 39: tokenizer.ggml.merges arr[str,247587] = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",... abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 40: tokenizer.ggml.eos_token_id u32 = 248046 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 41: tokenizer.ggml.padding_token_id u32 = 248044 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 42: tokenizer.ggml.bos_token_id u32 = 248044 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 43: tokenizer.chat_template str = {%- set image_count = namespace(value... abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 44: general.quantization_version u32 = 2 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 45: general.file_type u32 = 15 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - kv 46: general.license str = apache-2.0 abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - type f32: 301 tensors abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - type q4_K: 371 tensors abr 22 10:44:27 CachyOS ollama[51669]: llama_model_loader: - type q6_K: 61 tensors abr 22 10:44:27 CachyOS ollama[51669]: print_info: file format = GGUF V3 (latest) abr 22 10:44:27 CachyOS ollama[51669]: print_info: file type = Q4_K - Medium abr 22 10:44:27 CachyOS ollama[51669]: print_info: file size = 19.70 GiB (4.88 BPW) abr 22 10:44:27 CachyOS ollama[51669]: llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'qwen35moe' abr 22 10:44:27 CachyOS ollama[51669]: llama_model_load_from_file_impl: failed to load model abr 22 10:44:27 CachyOS ollama[51669]: time=2026-04-22T10:44:27.355-03:00 level=INFO source=sched.go:462 msg="failed to create server" model=hf.co/Youssofal/Qwen3.6-35B-A3B-Abliterated-Heretic-GGUF:Q4_K_M error="unable to load model: /usr/share/ollama/.ollama/models/blobs/sha256-ae2fb73ac0da875640269f1e65e9c7fb415b066c6d544c3eef9adb0d03f04792" abr 22 10:44:27 CachyOS ollama[51669]: [GIN] 2026/04/22 - 10:44:27 | 500 | 352.19334ms | 127.0.0.1 | POST "/api/generate" ``` ### OS Linux ### GPU Nvidia ### CPU AMD ### Ollama version ollama version is 0.21.0
GiteaMirror added the bug label 2026-05-05 03:28:19 -05:00
GiteaMirror commented 2026-05-05 03:28:22 -05:00
Author
Owner
Copy Link

@AASOxRevolution commented on GitHub (Apr 28, 2026):

Same issue on Ollama 0.21.2, two "different" Qwen3.6-35B models both fail identically.

Hardware: RTX 5090m · 24 GB VRAM · 96 GB RAM
Platform: Olares One (self-hosted, Kubernetes deployment)
Ollama: 0.21.2

Both of these models fail:
qwen3-6-35b-hauhau-uncensored:latest
qwen3-6-35b-q8:latest

time=2026-04-28T10:06:11.250Z level=INFO source=server.go:444 msg="starting runner" cmd="/usr/bin/ollama runner --ollama-engine --port 42447"
llama_model_loader: loaded meta data with 41 key-value pairs and 733 tensors from /root/.ollama/models/blobs/sha256-e56476632cf65ab1a32c135aead9eba4278f3bd3eb0499aa74304566d6e27d5b (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              = qwen35moe
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                     general.sampling.top_k i32              = 20
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              = Qwen_Qwen3.6 35B A3B
llama_model_loader: - kv   6:                           general.basename str              = Qwen_Qwen3.6
llama_model_loader: - kv   7:                         general.size_label str              = 35B-A3B
llama_model_loader: - kv   8:                      qwen35moe.block_count u32              = 40
llama_model_loader: - kv   9:                   qwen35moe.context_length u32              = 262144
llama_model_loader: - kv  10:                 qwen35moe.embedding_length u32              = 2048
llama_model_loader: - kv  11:             qwen35moe.attention.head_count u32              = 16
llama_model_loader: - kv  12:          qwen35moe.attention.head_count_kv u32              = 2
llama_model_loader: - kv  13:          qwen35moe.rope.dimension_sections arr[i32,4]       = [11, 11, 10, 0]
llama_model_loader: - kv  14:                   qwen35moe.rope.freq_base f32              = 10000000.000000
llama_model_loader: - kv  15: qwen35moe.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  16:                     qwen35moe.expert_count u32              = 256
llama_model_loader: - kv  17:                qwen35moe.expert_used_count u32              = 8
llama_model_loader: - kv  18:             qwen35moe.attention.key_length u32              = 256
llama_model_loader: - kv  19:           qwen35moe.attention.value_length u32              = 256
llama_model_loader: - kv  20:       qwen35moe.expert_feed_forward_length u32              = 512
llama_model_loader: - kv  21: qwen35moe.expert_shared_feed_forward_length u32              = 512
llama_model_loader: - kv  22:                  qwen35moe.ssm.conv_kernel u32              = 4
llama_model_loader: - kv  23:                   qwen35moe.ssm.state_size u32              = 128
llama_model_loader: - kv  24:                  qwen35moe.ssm.group_count u32              = 16
llama_model_loader: - kv  25:               qwen35moe.ssm.time_step_rank u32              = 32
llama_model_loader: - kv  26:                   qwen35moe.ssm.inner_size u32              = 4096
llama_model_loader: - kv  27:          qwen35moe.full_attention_interval u32              = 4
llama_model_loader: - kv  28:             qwen35moe.rope.dimension_count u32              = 64
llama_model_loader: - kv  29:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  30:                         tokenizer.ggml.pre str              = qwen35
llama_model_loader: - kv  31:                      tokenizer.ggml.tokens arr[str,248320]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  32:                  tokenizer.ggml.token_type arr[i32,248320]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  33:                      tokenizer.ggml.merges arr[str,247587]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  34:                tokenizer.ggml.eos_token_id u32              = 248046
llama_model_loader: - kv  35:            tokenizer.ggml.padding_token_id u32              = 248044
llama_model_loader: - kv  36:                tokenizer.ggml.bos_token_id u32              = 248044
llama_model_loader: - kv  37:               tokenizer.ggml.add_bos_token bool             = false
llama_model_loader: - kv  38:                    tokenizer.chat_template str              = {%- set image_count = namespace(value...
llama_model_loader: - kv  39:               general.quantization_version u32              = 2
llama_model_loader: - kv  40:                          general.file_type u32              = 7
llama_model_loader: - type  f32:  301 tensors
llama_model_loader: - type q8_0:  432 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q8_0
print_info: file size   = 34.36 GiB (8.52 BPW) 
[GIN] 2026/04/28 - 10:06:11 | 500 |  575.813739ms |  192.168.50.234 | POST     "/api/generate"

llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'qwen35moe'
llama_model_load_from_file_impl: failed to load model
time=2026-04-28T10:06:11.623Z level=INFO source=sched.go:462 msg="failed to create server" model=qwen3-6-35b-q8:latest error="unable to load model: /root/.ollama/models/blobs/sha256-e56476632cf65ab1a32c135aead9eba4278f3bd3eb0499aa74304566d6e27d5b"

The GGUF loads fine, metadata reads correctly, but the architecture handler for qwen35moe is missing entirely.

<!-- gh-comment-id:4334302242 --> @AASOxRevolution commented on GitHub (Apr 28, 2026): Same issue on Ollama 0.21.2, two "different" Qwen3.6-35B models both fail identically. Hardware: RTX 5090m · 24 GB VRAM · 96 GB RAM Platform: Olares One (self-hosted, Kubernetes deployment) Ollama: 0.21.2 Both of these models fail: qwen3-6-35b-hauhau-uncensored:latest qwen3-6-35b-q8:latest ```text time=2026-04-28T10:06:11.250Z level=INFO source=server.go:444 msg="starting runner" cmd="/usr/bin/ollama runner --ollama-engine --port 42447" llama_model_loader: loaded meta data with 41 key-value pairs and 733 tensors from /root/.ollama/models/blobs/sha256-e56476632cf65ab1a32c135aead9eba4278f3bd3eb0499aa74304566d6e27d5b (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 = qwen35moe llama_model_loader: - kv 1: general.type str = model llama_model_loader: - kv 2: general.sampling.top_k i32 = 20 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 = Qwen_Qwen3.6 35B A3B llama_model_loader: - kv 6: general.basename str = Qwen_Qwen3.6 llama_model_loader: - kv 7: general.size_label str = 35B-A3B llama_model_loader: - kv 8: qwen35moe.block_count u32 = 40 llama_model_loader: - kv 9: qwen35moe.context_length u32 = 262144 llama_model_loader: - kv 10: qwen35moe.embedding_length u32 = 2048 llama_model_loader: - kv 11: qwen35moe.attention.head_count u32 = 16 llama_model_loader: - kv 12: qwen35moe.attention.head_count_kv u32 = 2 llama_model_loader: - kv 13: qwen35moe.rope.dimension_sections arr[i32,4] = [11, 11, 10, 0] llama_model_loader: - kv 14: qwen35moe.rope.freq_base f32 = 10000000.000000 llama_model_loader: - kv 15: qwen35moe.attention.layer_norm_rms_epsilon f32 = 0.000001 llama_model_loader: - kv 16: qwen35moe.expert_count u32 = 256 llama_model_loader: - kv 17: qwen35moe.expert_used_count u32 = 8 llama_model_loader: - kv 18: qwen35moe.attention.key_length u32 = 256 llama_model_loader: - kv 19: qwen35moe.attention.value_length u32 = 256 llama_model_loader: - kv 20: qwen35moe.expert_feed_forward_length u32 = 512 llama_model_loader: - kv 21: qwen35moe.expert_shared_feed_forward_length u32 = 512 llama_model_loader: - kv 22: qwen35moe.ssm.conv_kernel u32 = 4 llama_model_loader: - kv 23: qwen35moe.ssm.state_size u32 = 128 llama_model_loader: - kv 24: qwen35moe.ssm.group_count u32 = 16 llama_model_loader: - kv 25: qwen35moe.ssm.time_step_rank u32 = 32 llama_model_loader: - kv 26: qwen35moe.ssm.inner_size u32 = 4096 llama_model_loader: - kv 27: qwen35moe.full_attention_interval u32 = 4 llama_model_loader: - kv 28: qwen35moe.rope.dimension_count u32 = 64 llama_model_loader: - kv 29: tokenizer.ggml.model str = gpt2 llama_model_loader: - kv 30: tokenizer.ggml.pre str = qwen35 llama_model_loader: - kv 31: tokenizer.ggml.tokens arr[str,248320] = ["!", "\"", "#", "$", "%", "&", "'", ... llama_model_loader: - kv 32: tokenizer.ggml.token_type arr[i32,248320] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ... llama_model_loader: - kv 33: tokenizer.ggml.merges arr[str,247587] = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",... llama_model_loader: - kv 34: tokenizer.ggml.eos_token_id u32 = 248046 llama_model_loader: - kv 35: tokenizer.ggml.padding_token_id u32 = 248044 llama_model_loader: - kv 36: tokenizer.ggml.bos_token_id u32 = 248044 llama_model_loader: - kv 37: tokenizer.ggml.add_bos_token bool = false llama_model_loader: - kv 38: tokenizer.chat_template str = {%- set image_count = namespace(value... llama_model_loader: - kv 39: general.quantization_version u32 = 2 llama_model_loader: - kv 40: general.file_type u32 = 7 llama_model_loader: - type f32: 301 tensors llama_model_loader: - type q8_0: 432 tensors print_info: file format = GGUF V3 (latest) print_info: file type = Q8_0 print_info: file size = 34.36 GiB (8.52 BPW) [GIN] 2026/04/28 - 10:06:11 | 500 | 575.813739ms | 192.168.50.234 | POST "/api/generate" ``` llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'qwen35moe' llama_model_load_from_file_impl: failed to load model time=2026-04-28T10:06:11.623Z level=INFO source=sched.go:462 msg="failed to create server" model=qwen3-6-35b-q8:latest error="unable to load model: /root/.ollama/models/blobs/sha256-e56476632cf65ab1a32c135aead9eba4278f3bd3eb0499aa74304566d6e27d5b" The GGUF loads fine, metadata reads correctly, but the architecture handler for qwen35moe is missing entirely.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#72097
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?