github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 08:30:05 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #6440] Model architecture Gemma2ForCausalLm #50561

New Issue
Closed
opened 2026-04-28 16:23:41 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-04-28 16:23:41 -05:00
Owner
Copy Link

Originally created by @luisgg98 on GitHub (Aug 20, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/6440

What is the issue?

Good afternoon, I would like to start marking I am not 100% sure whether this is an issue or maybe I am misunderstanding the concept of architecture.
I tried to create a model on ollama by using a Modelfile at version 0.3.0.

imagen
I go the previous error, so I decided to upgrade ollama to version 0.3.6
Which pops off error "unsupported architecture":
imagen

Configuration file of the model I have finetuned:

{
  "_name_or_path": "google/gemma-2-9b",
  "architectures": [
    "Gemma2ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "attn_logit_softcapping": 50.0,
  "bos_token_id": 2,
  "cache_implementation": "hybrid",
  "eos_token_id": 1,
  "final_logit_softcapping": 30.0,
  "head_dim": 256,
  "hidden_act": "gelu_pytorch_tanh",
  "hidden_activation": "gelu_pytorch_tanh",
  "hidden_size": 3584,
  "initializer_range": 0.02,
  "intermediate_size": 14336,
  "max_position_embeddings": 8192,
  "model_type": "gemma2",
  "num_attention_heads": 16,
  "num_hidden_layers": 42,
  "num_key_value_heads": 8,
  "pad_token_id": 0,
  "query_pre_attn_scalar": 256,
  "rms_norm_eps": 1e-06,
  "rope_theta": 10000.0,
  "sliding_window": 4096,
  "sliding_window_size": 4096,
  "torch_dtype": "float16",
  "transformers_version": "4.43.3",
  "use_cache": false,
  "vocab_size": 256000
}

It's a model based on Gemma2 9b. Gemma2 works smoothly.
imagen

Modelfile:

FROM /path/model/
TEMPLATE """<start_of_turn>user/
{{ if .System }}{{ .System }} {{ end }}{{ .Prompt }}<end_of_turn>
<start_of_turn>model
{{ .Response }}<end_of_turn>"""
PARAMETER stop "<start_of_turn>"
PARAMETER stop "<end_of_turn>"
SYSTEM """Below is an instruction that describes a task. Write a response that appropriately completes the request.
Generate a concise summary in Spanish the following input. Here is the input"""

I don't understand by Gemma2 is supported but models based on it are not.
Thank you for reaching this point. Please if you have any clue about what's the issue please write it down bellow.

OS

Linux

GPU

Nvidia

CPU

No response

Ollama version

0.3.6

Originally created by @luisgg98 on GitHub (Aug 20, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/6440 ### What is the issue? Good afternoon, I would like to start marking I am not 100% sure whether this is an issue or maybe I am misunderstanding the concept of architecture. I tried to create a model on ollama by using a Modelfile at version 0.3.0. ![imagen](https://github.com/user-attachments/assets/dfdd4b08-5eb8-4939-a536-fd739dd2e784) I go the previous error, so I decided to upgrade ollama to version 0.3.6 Which pops off error "unsupported architecture": ![imagen](https://github.com/user-attachments/assets/b51c374e-dec6-4e81-acf5-b292ce3af43d) Configuration file of the model I have finetuned: ```json { "_name_or_path": "google/gemma-2-9b", "architectures": [ "Gemma2ForCausalLM" ], "attention_bias": false, "attention_dropout": 0.0, "attn_logit_softcapping": 50.0, "bos_token_id": 2, "cache_implementation": "hybrid", "eos_token_id": 1, "final_logit_softcapping": 30.0, "head_dim": 256, "hidden_act": "gelu_pytorch_tanh", "hidden_activation": "gelu_pytorch_tanh", "hidden_size": 3584, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "model_type": "gemma2", "num_attention_heads": 16, "num_hidden_layers": 42, "num_key_value_heads": 8, "pad_token_id": 0, "query_pre_attn_scalar": 256, "rms_norm_eps": 1e-06, "rope_theta": 10000.0, "sliding_window": 4096, "sliding_window_size": 4096, "torch_dtype": "float16", "transformers_version": "4.43.3", "use_cache": false, "vocab_size": 256000 } ``` It's a model based on Gemma2 9b. Gemma2 works smoothly. ![imagen](https://github.com/user-attachments/assets/965205ba-03df-4273-bb3e-2ad9ce34b807) Modelfile: ```Dockerfile FROM /path/model/ TEMPLATE """<start_of_turn>user/ {{ if .System }}{{ .System }} {{ end }}{{ .Prompt }}<end_of_turn> <start_of_turn>model {{ .Response }}<end_of_turn>""" PARAMETER stop "<start_of_turn>" PARAMETER stop "<end_of_turn>" SYSTEM """Below is an instruction that describes a task. Write a response that appropriately completes the request. Generate a concise summary in Spanish the following input. Here is the input""" ``` I don't understand by Gemma2 is supported but models based on it are not. Thank you for reaching this point. Please if you have any clue about what's the issue please write it down bellow. ### OS Linux ### GPU Nvidia ### CPU _No response_ ### Ollama version 0.3.6
GiteaMirror added the bug label 2026-04-28 16:23:41 -05:00
GiteaMirror commented 2026-04-28 16:23:45 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Aug 20, 2024):

The create function of ollama primarily works with GGUF files. The ability to import safetensors (I'm assuming that's what's in /path/model) is a convenience function but it's limited in the architectures it can process. If you want to import a model that is not in the supported architecture, you need to use the convert_hf_to_gguf.py script in llama.cpp

<!-- gh-comment-id:2298609585 --> @rick-github commented on GitHub (Aug 20, 2024): The `create` function of ollama primarily works with GGUF files. The ability to import safetensors (I'm assuming that's what's in /path/model) is a convenience function but it's limited in the architectures it can process. If you want to import a model that is not in the supported architecture, you need to use the `convert_hf_to_gguf.py` script in [llama.cpp](https://github.com/ggerganov/llama.cpp)
GiteaMirror commented 2026-04-28 16:23:45 -05:00
Author
Owner
Copy Link

@mxyng commented on GitHub (Aug 21, 2024):

Gemma2 conversion is implemented by https://github.com/ollama/ollama/pull/5365 which has been merged and will be in the next release

<!-- gh-comment-id:2302936214 --> @mxyng commented on GitHub (Aug 21, 2024): Gemma2 conversion is implemented by https://github.com/ollama/ollama/pull/5365 which has been merged and will be in the next release
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#50561
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?