ollama
Error: invalid file magic when importing Safetensors models
What is the issue?
ollama create test -f Modelfile transferring model data creating model layer Error: invalid file magic
This happens for all the Safetensors models I try to import.
Modelfile content
FROM ./model.safetensors
What did you expect to see?
expect to working :)
Steps to reproduce
No response
Are there any recent changes that introduced the issue?
No response
OS
Windows
Architecture
amd64
Platform
No response
Ollama version
0.1.29
GPU
Nvidia
GPU info
No response
CPU
Intel
Other software
No response
What safetensors model were you trying to import? Right now only Mistral and Mistral fine tunes are supported. More are coming soon though!
Oh, okay, I didn't know that. I was just trying to test random models to see how they're working :D
I tried three and four text models, but none of them worked, anyway none of them were Mistral.
Sorry about that! I have Gemma now working, but haven't yet sent out the PR. I'll add an error message saying that the other models aren't yet supported.
@amnweb can you list which models you tried? I just realized there should be code to catch that.
I think this is the last one I tried
https://huggingface.co/google-bert/bert-base-uncased/tree/main
or this
https://huggingface.co/pysentimiento/robertuito-sentiment-analysis/tree/main
Already delete from disk and can't remember what I was downloaded
Edit: Btw this one also, just found it in history
https://huggingface.co/stabilityai/stable-diffusion-2-1/tree/main
Can you post one of the modelfiles? I'm trying to figure out if you had converted/quantized these yourself or got ollama to convert the safetensors files.
I have downloaded model.safetensors https://huggingface.co/google-bert/bert-base-uncased/resolve/main/model.safetensors
Created Modelfile with content FROM ./model.safetensors and run command from terminal. If I understand Ollama should be convert safetensors or I'm thinking wrong ?
I'm having this same issue with ollama in docker and using https://huggingface.co/TheBloke/CodeLlama-7B-Instruct-GGUF codellama-7b-instruct.Q4_0.gguf file.. here is my modelfile
FROM /models/CodeLlama-7B-Instruct-GGUF/codellama-7b-instruct.Q4_0.gguf
TEMPLATE """[INST] <<SYS>>{{ .System }}<</SYS>>
{{ .Prompt }} [/INST]"""
PARAMETER rope_frequency_base 1000000
PARAMETER stop [INST]
PARAMETER stop [/INST]
PARAMETER stop <<SYS>>
PARAMETER stop <</SYS>>
I'm having this same issue with ollama in docker and using https://huggingface.co/TheBloke/CodeLlama-7B-Instruct-GGUF codellama-7b-instruct.Q4_0.gguf file.. here is my modelfile
I had a corrupt file.
@amnweb sorry for the slow response! I somehow lost track of this. I don't believe any of the models you converted will work inside of the llama.cpp runner unfortunately.
