github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-21 13:11:39 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #12896] HF Qwen3VL (&MoE) didn't work on Ollama #8546

New Issue
Open
opened 2026-04-12 21:15:32 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-12 21:15:32 -05:00
Owner
Copy Link

Originally created by @GlisseManTV on GitHub (Nov 1, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/12896

What is the issue?

ollama run hf.co/unsloth/Qwen3-VL-30B-A3B-Instruct-GGUF:IQ4_NL
llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'qwen3vlmoe'
ollama run hf.co/unsloth/Qwen3-VL-8B-Instruct-GGUF:IQ4_NL
llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'qwen3vl'

Relevant log output

time=2025-11-01T01:45:39.140+01:00 level=INFO source=cpu_windows.go:186 msg="" package=0 cores=16 efficiency=0 threads=32
time=2025-11-01T01:45:39.167+01:00 level=INFO source=sched.go:559 msg="updated VRAM based on existing loaded models" gpu=GPU-b0e2e04f-ff12-3126-25de-b4526643f498 library=CUDA total="12.0 GiB" available="2.9 GiB"
time=2025-11-01T01:45:39.168+01:00 level=INFO source=sched.go:559 msg="updated VRAM based on existing loaded models" gpu=GPU-74ab4ebe-fc23-4f91-8e84-e912f7b61f0a library=CUDA total="12.0 GiB" available="1.7 GiB"
llama_model_loader: loaded meta data with 42 key-value pairs and 399 tensors from C:\Users\appuser\.ollama\models\blobs\sha256-31fabbd15be7617110ee6bfdc9e1a7db9c052e1e5a6e59f54446d9125e36a7a7 (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              = qwen3vl
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Qwen3-Vl-8B-Instruct
llama_model_loader: - kv   3:                           general.finetune str              = Instruct
llama_model_loader: - kv   4:                           general.basename str              = Qwen3-Vl-8B-Instruct
llama_model_loader: - kv   5:                       general.quantized_by str              = Unsloth
llama_model_loader: - kv   6:                         general.size_label str              = 8B
llama_model_loader: - kv   7:                            general.license str              = apache-2.0
llama_model_loader: - kv   8:                           general.repo_url str              = https://huggingface.co/unsloth
llama_model_loader: - kv   9:                   general.base_model.count u32              = 1
llama_model_loader: - kv  10:                  general.base_model.0.name str              = Qwen3 VL 8B Instruct
llama_model_loader: - kv  11:          general.base_model.0.organization str              = Qwen
llama_model_loader: - kv  12:              general.base_model.0.repo_url str              = https://huggingface.co/Qwen/Qwen3-VL-...
llama_model_loader: - kv  13:                               general.tags arr[str,2]       = ["unsloth", "image-text-to-text"]
llama_model_loader: - kv  14:                        qwen3vl.block_count u32              = 36
llama_model_loader: - kv  15:                     qwen3vl.context_length u32              = 262144
llama_model_loader: - kv  16:                   qwen3vl.embedding_length u32              = 4096
llama_model_loader: - kv  17:                qwen3vl.feed_forward_length u32              = 12288
llama_model_loader: - kv  18:               qwen3vl.attention.head_count u32              = 32
llama_model_loader: - kv  19:            qwen3vl.attention.head_count_kv u32              = 8
llama_model_loader: - kv  20:                     qwen3vl.rope.freq_base f32              = 5000000.000000
llama_model_loader: - kv  21:   qwen3vl.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  22:               qwen3vl.attention.key_length u32              = 128
llama_model_loader: - kv  23:             qwen3vl.attention.value_length u32              = 128
llama_model_loader: - kv  24:            qwen3vl.rope.dimension_sections arr[i32,4]       = [24, 20, 20, 0]
llama_model_loader: - kv  25:                 qwen3vl.n_deepstack_layers u32              = 3
llama_model_loader: - kv  26:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  27:                         tokenizer.ggml.pre str              = qwen2
llama_model_loader: - kv  28:                      tokenizer.ggml.tokens arr[str,151936]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  29:                  tokenizer.ggml.token_type arr[i32,151936]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  30:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  31:                tokenizer.ggml.eos_token_id u32              = 151645
llama_model_loader: - kv  32:            tokenizer.ggml.padding_token_id u32              = 151654
llama_model_loader: - kv  33:                tokenizer.ggml.bos_token_id u32              = 151643
llama_model_loader: - kv  34:               tokenizer.ggml.add_bos_token bool             = false
llama_model_loader: - kv  35:                    tokenizer.chat_template str              = {%- if tools %}\n    {{- '<|im_start|>...
llama_model_loader: - kv  36:               general.quantization_version u32              = 2
llama_model_loader: - kv  37:                          general.file_type u32              = 25
llama_model_loader: - kv  38:                      quantize.imatrix.file str              = Qwen3-VL-8B-Instruct-GGUF/imatrix_uns...
llama_model_loader: - kv  39:                   quantize.imatrix.dataset str              = unsloth_calibration_Qwen3-VL-8B-Instr...
llama_model_loader: - kv  40:             quantize.imatrix.entries_count u32              = 252
llama_model_loader: - kv  41:              quantize.imatrix.chunks_count u32              = 694
llama_model_loader: - type  f32:  145 tensors
llama_model_loader: - type q4_K:    1 tensors
llama_model_loader: - type q5_K:   36 tensors
llama_model_loader: - type q6_K:    1 tensors
llama_model_loader: - type iq4_nl:  216 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = IQ4_NL - 4.5 bpw
print_info: file size   = 4.46 GiB (4.68 BPW) 
llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'qwen3vl'
llama_model_load_from_file_impl: failed to load model

OS

Windows

GPU

Nvidia

CPU

Intel

Ollama version

0.12.8

Originally created by @GlisseManTV on GitHub (Nov 1, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/12896 ### What is the issue? ollama run hf.co/unsloth/Qwen3-VL-30B-A3B-Instruct-GGUF:IQ4_NL llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'qwen3vlmoe' ollama run hf.co/unsloth/Qwen3-VL-8B-Instruct-GGUF:IQ4_NL llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'qwen3vl' ### Relevant log output ```shell time=2025-11-01T01:45:39.140+01:00 level=INFO source=cpu_windows.go:186 msg="" package=0 cores=16 efficiency=0 threads=32 time=2025-11-01T01:45:39.167+01:00 level=INFO source=sched.go:559 msg="updated VRAM based on existing loaded models" gpu=GPU-b0e2e04f-ff12-3126-25de-b4526643f498 library=CUDA total="12.0 GiB" available="2.9 GiB" time=2025-11-01T01:45:39.168+01:00 level=INFO source=sched.go:559 msg="updated VRAM based on existing loaded models" gpu=GPU-74ab4ebe-fc23-4f91-8e84-e912f7b61f0a library=CUDA total="12.0 GiB" available="1.7 GiB" llama_model_loader: loaded meta data with 42 key-value pairs and 399 tensors from C:\Users\appuser\.ollama\models\blobs\sha256-31fabbd15be7617110ee6bfdc9e1a7db9c052e1e5a6e59f54446d9125e36a7a7 (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 = qwen3vl llama_model_loader: - kv 1: general.type str = model llama_model_loader: - kv 2: general.name str = Qwen3-Vl-8B-Instruct llama_model_loader: - kv 3: general.finetune str = Instruct llama_model_loader: - kv 4: general.basename str = Qwen3-Vl-8B-Instruct llama_model_loader: - kv 5: general.quantized_by str = Unsloth llama_model_loader: - kv 6: general.size_label str = 8B llama_model_loader: - kv 7: general.license str = apache-2.0 llama_model_loader: - kv 8: general.repo_url str = https://huggingface.co/unsloth llama_model_loader: - kv 9: general.base_model.count u32 = 1 llama_model_loader: - kv 10: general.base_model.0.name str = Qwen3 VL 8B Instruct llama_model_loader: - kv 11: general.base_model.0.organization str = Qwen llama_model_loader: - kv 12: general.base_model.0.repo_url str = https://huggingface.co/Qwen/Qwen3-VL-... llama_model_loader: - kv 13: general.tags arr[str,2] = ["unsloth", "image-text-to-text"] llama_model_loader: - kv 14: qwen3vl.block_count u32 = 36 llama_model_loader: - kv 15: qwen3vl.context_length u32 = 262144 llama_model_loader: - kv 16: qwen3vl.embedding_length u32 = 4096 llama_model_loader: - kv 17: qwen3vl.feed_forward_length u32 = 12288 llama_model_loader: - kv 18: qwen3vl.attention.head_count u32 = 32 llama_model_loader: - kv 19: qwen3vl.attention.head_count_kv u32 = 8 llama_model_loader: - kv 20: qwen3vl.rope.freq_base f32 = 5000000.000000 llama_model_loader: - kv 21: qwen3vl.attention.layer_norm_rms_epsilon f32 = 0.000001 llama_model_loader: - kv 22: qwen3vl.attention.key_length u32 = 128 llama_model_loader: - kv 23: qwen3vl.attention.value_length u32 = 128 llama_model_loader: - kv 24: qwen3vl.rope.dimension_sections arr[i32,4] = [24, 20, 20, 0] llama_model_loader: - kv 25: qwen3vl.n_deepstack_layers u32 = 3 llama_model_loader: - kv 26: tokenizer.ggml.model str = gpt2 llama_model_loader: - kv 27: tokenizer.ggml.pre str = qwen2 llama_model_loader: - kv 28: tokenizer.ggml.tokens arr[str,151936] = ["!", "\"", "#", "$", "%", "&", "'", ... llama_model_loader: - kv 29: tokenizer.ggml.token_type arr[i32,151936] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ... llama_model_loader: - kv 30: tokenizer.ggml.merges arr[str,151387] = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",... llama_model_loader: - kv 31: tokenizer.ggml.eos_token_id u32 = 151645 llama_model_loader: - kv 32: tokenizer.ggml.padding_token_id u32 = 151654 llama_model_loader: - kv 33: tokenizer.ggml.bos_token_id u32 = 151643 llama_model_loader: - kv 34: tokenizer.ggml.add_bos_token bool = false llama_model_loader: - kv 35: tokenizer.chat_template str = {%- if tools %}\n {{- '<|im_start|>... llama_model_loader: - kv 36: general.quantization_version u32 = 2 llama_model_loader: - kv 37: general.file_type u32 = 25 llama_model_loader: - kv 38: quantize.imatrix.file str = Qwen3-VL-8B-Instruct-GGUF/imatrix_uns... llama_model_loader: - kv 39: quantize.imatrix.dataset str = unsloth_calibration_Qwen3-VL-8B-Instr... llama_model_loader: - kv 40: quantize.imatrix.entries_count u32 = 252 llama_model_loader: - kv 41: quantize.imatrix.chunks_count u32 = 694 llama_model_loader: - type f32: 145 tensors llama_model_loader: - type q4_K: 1 tensors llama_model_loader: - type q5_K: 36 tensors llama_model_loader: - type q6_K: 1 tensors llama_model_loader: - type iq4_nl: 216 tensors print_info: file format = GGUF V3 (latest) print_info: file type = IQ4_NL - 4.5 bpw print_info: file size = 4.46 GiB (4.68 BPW) llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'qwen3vl' llama_model_load_from_file_impl: failed to load model ``` ### OS Windows ### GPU Nvidia ### CPU Intel ### Ollama version 0.12.8
GiteaMirror added the bug label 2026-04-12 21:15:32 -05:00
GiteaMirror commented 2026-04-12 21:15:34 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Nov 1, 2025):

qwen3vl is supported in the ollama engine. The HF model is a split vision model which is not supported in the ollama engine so the ollama server falls back to the llama.cpp engine, which doesn't support qwen3vl yet.

<!-- gh-comment-id:3475889486 --> @rick-github commented on GitHub (Nov 1, 2025): qwen3vl is supported in the ollama engine. The HF model is a split vision model which is not supported in the ollama engine so the ollama server falls back to the llama.cpp engine, which doesn't support qwen3vl yet.
GiteaMirror commented 2026-04-12 21:15:35 -05:00
Author
Owner
Copy Link

@klimekop6 commented on GitHub (Nov 7, 2025):

Is there any workaround for such split models? I also try to use unsloth qwen3vl quantized model and getting the same error

qwen3vl is supported in the ollama engine. The HF model is a split vision model which is not supported in the ollama engine so the ollama server falls back to the llama.cpp engine, which doesn't support qwen3vl yet.

<!-- gh-comment-id:3503234553 --> @klimekop6 commented on GitHub (Nov 7, 2025): Is there any workaround for such split models? I also try to use unsloth qwen3vl quantized model and getting the same error > qwen3vl is supported in the ollama engine. The HF model is a split vision model which is not supported in the ollama engine so the ollama server falls back to the llama.cpp engine, which doesn't support qwen3vl yet.
GiteaMirror commented 2026-04-12 21:15:36 -05:00
Author
Owner
Copy Link

@Wyldhunt commented on GitHub (Nov 25, 2025):

I also can't seem to find a workaround for this. What all would be required to add support for these models?
They are listed in the Ollama model registry as valid, so I assume it's possible?

<!-- gh-comment-id:3576412564 --> @Wyldhunt commented on GitHub (Nov 25, 2025): I also can't seem to find a workaround for this. What all would be required to add support for these models? They are listed in the Ollama model registry as valid, so I assume it's possible?
GiteaMirror commented 2026-04-12 21:15:37 -05:00
Author
Owner
Copy Link

@OldishCoder commented on GitHub (Dec 12, 2025):

I'm still seeing this error for qwen3-vl:30b:
"llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'qwen3vlmoe'"
With v0.13.3.

Will this be addressed anytime soon?

<!-- gh-comment-id:3646446206 --> @OldishCoder commented on GitHub (Dec 12, 2025): I'm still seeing this error for qwen3-vl:30b: "llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'qwen3vlmoe'" With v0.13.3. Will this be addressed anytime soon?
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#8546
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?