github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #11615] GEMMA3N Using matformer, must set intermediate_size of type array #33430

New Issue
Open
opened 2026-04-22 16:05:06 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-22 16:05:06 -05:00
Owner
Copy Link

Originally created by @xValentim on GitHub (Jul 31, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/11615

What is the issue?

If you customize the model that use matformer architecture, ollama import models crashes. It happens because in gemma3n model architecture we recieve this:

type gemma3nModel struct {
	ModelParameters

	TextModel struct {
		ActivationSparsityPattern []float32 `json:"activation_sparsity_pattern"`
		AltupActiveIdx            uint32    `json:"altup_active_idx"`
		AltupCoefClip             float32   `json:"altup_coef_clip"`
		AltupCorrectScale         bool      `json:"altup_correct_scale"`
		AltupLRMultiplier         float32   `json:"altup_lr_multiplier"`
		AltupNumInputs            uint32    `json:"altup_num_inputs"`
		HeadDim                   uint32    `json:"head_dim"`
		HiddenSize                uint32    `json:"hidden_size"`
		HiddenSizePerLayerInput   uint32    `json:"hidden_size_per_layer_input"`
		IntermediateSize          uint32    `json:"intermediate_size"`
		MaxPositionEmbeddings     uint32    `json:"max_position_embeddings"`
		NumAttentionHeads         uint32    `json:"num_attention_heads"`
		NumHiddenLayers           uint32    `json:"num_hidden_layers"`
		NumKeyValueHeads          uint32    `json:"num_key_value_heads"`
		NumKVSharedLayers         uint32    `json:"num_kv_shared_layers"`
		RMSNormEPS                float32   `json:"rms_norm_eps"`
		RopeLocalBaseFreq         float32   `json:"rope_local_base_freq"`
		RopeTheta                 float32   `json:"rope_theta"`
		SlidingWindow             uint32    `json:"sliding_window"`
		LayerTypes                []string  `json:"layer_types"`
	} `json:"text_config"`
	VisionModel struct{} `json:"vision_config"`
}

And, if you customize your matformer model, your config.json will be like this:

...
"hidden_activation": "gelu_pytorch_tanh",
    "hidden_size": 2048,
    "hidden_size_per_layer_input": 256,
    "initializer_range": 0.02,
    "intermediate_size": [
      8192,
      8192,
      8192,
      8192,
      8192,
      8192,
      8192,
      8192,
      8192,
      8192,
      16384,
      16384,
      16384,
      16384,
      16384,
      16384,
      16384,
      16384,
      16384,
      16384,
      16384,
      16384,
      16384,
      16384,
      16384,
      8192,
      8192,
      8192,
      8192,
      8192,
      8192,
      8192,
      8192,
      8192,
      8192
    ],
    "laurel_rank": 64,
    "layer_types": [
      "sliding_attention",
      "sliding_attention",
      "sliding_attention",
...

Someone can help me? I tried to run dev mode using go run . serve, but I have many errors.

Relevant log output

ollama create gemma3n-sliced -f Modelfile
gathering model components
copying file sha256:28db4662fe863787861c87bc89ed212b552062819ca0e6fe1723af84de1417ac 100%
copying file sha256:b6c35ee648c07754b44cd9e371c75d4caa05c4504910b7ad29b1847ee9d8ba5d 100%
copying file sha256:9e5d11c5125889bdeca6b91a95359f2dfb578bd147198f3c5ed08870e6155cba 100%
copying file sha256:7d792604c9b2e9796f432bcbb073c4798d5fd0d6dbc69dc3d3eeee955225d4db 100%
copying file sha256:5a630a1f95ef9f8e49ebc8ed5b7aeb1bbd86c695ca5af439ae07187ee5b52fc8 100%
copying file sha256:648419887a29e74bcc9eb6b111734e1bfb569fe4df6ee81c95ee7d6c7864d448 100%
copying file sha256:8521a807b78a86c80140be868c8fab67584c13b7070bdae2e4dd475cfbdc105e 100%
copying file sha256:8d52220a251a5ae826d89741237bd30ede0bd3df398b1f1d7939ad8ac3e39c08 100%
converting model
Error: json: cannot unmarshal array into Go struct field .text_config.intermediate_size of type uint32

OS

Windows

GPU

Nvidia

CPU

Intel

Ollama version

No response

Originally created by @xValentim on GitHub (Jul 31, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/11615 ### What is the issue? If you customize the model that use matformer architecture, ollama import models crashes. It happens because in gemma3n model architecture we recieve this: ```go type gemma3nModel struct { ModelParameters TextModel struct { ActivationSparsityPattern []float32 `json:"activation_sparsity_pattern"` AltupActiveIdx uint32 `json:"altup_active_idx"` AltupCoefClip float32 `json:"altup_coef_clip"` AltupCorrectScale bool `json:"altup_correct_scale"` AltupLRMultiplier float32 `json:"altup_lr_multiplier"` AltupNumInputs uint32 `json:"altup_num_inputs"` HeadDim uint32 `json:"head_dim"` HiddenSize uint32 `json:"hidden_size"` HiddenSizePerLayerInput uint32 `json:"hidden_size_per_layer_input"` IntermediateSize uint32 `json:"intermediate_size"` MaxPositionEmbeddings uint32 `json:"max_position_embeddings"` NumAttentionHeads uint32 `json:"num_attention_heads"` NumHiddenLayers uint32 `json:"num_hidden_layers"` NumKeyValueHeads uint32 `json:"num_key_value_heads"` NumKVSharedLayers uint32 `json:"num_kv_shared_layers"` RMSNormEPS float32 `json:"rms_norm_eps"` RopeLocalBaseFreq float32 `json:"rope_local_base_freq"` RopeTheta float32 `json:"rope_theta"` SlidingWindow uint32 `json:"sliding_window"` LayerTypes []string `json:"layer_types"` } `json:"text_config"` VisionModel struct{} `json:"vision_config"` } ``` And, if you customize your matformer model, your config.json will be like this: ```json ... "hidden_activation": "gelu_pytorch_tanh", "hidden_size": 2048, "hidden_size_per_layer_input": 256, "initializer_range": 0.02, "intermediate_size": [ 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192 ], "laurel_rank": 64, "layer_types": [ "sliding_attention", "sliding_attention", "sliding_attention", ... ``` Someone can help me? I tried to run dev mode using `go run . serve`, but I have many errors. ### Relevant log output ```shell ollama create gemma3n-sliced -f Modelfile gathering model components copying file sha256:28db4662fe863787861c87bc89ed212b552062819ca0e6fe1723af84de1417ac 100% copying file sha256:b6c35ee648c07754b44cd9e371c75d4caa05c4504910b7ad29b1847ee9d8ba5d 100% copying file sha256:9e5d11c5125889bdeca6b91a95359f2dfb578bd147198f3c5ed08870e6155cba 100% copying file sha256:7d792604c9b2e9796f432bcbb073c4798d5fd0d6dbc69dc3d3eeee955225d4db 100% copying file sha256:5a630a1f95ef9f8e49ebc8ed5b7aeb1bbd86c695ca5af439ae07187ee5b52fc8 100% copying file sha256:648419887a29e74bcc9eb6b111734e1bfb569fe4df6ee81c95ee7d6c7864d448 100% copying file sha256:8521a807b78a86c80140be868c8fab67584c13b7070bdae2e4dd475cfbdc105e 100% copying file sha256:8d52220a251a5ae826d89741237bd30ede0bd3df398b1f1d7939ad8ac3e39c08 100% converting model Error: json: cannot unmarshal array into Go struct field .text_config.intermediate_size of type uint32 ``` ### OS Windows ### GPU Nvidia ### CPU Intel ### Ollama version _No response_
GiteaMirror added the bug label 2026-04-22 16:05:06 -05:00
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#33430
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?