~/tools / Model files and chat templates / Ollama Modelfile Checker

Ollama Modelfile Checker — Check a Modelfile the way ollama create reads it

Paste an Ollama Modelfile and see whether ollama create will accept it, before you run it. The tool reads the file the way Ollama does and shows Ollama's own error message with the line, what each instruction ends up as, the parameters with the types Ollama stores, the example messages, and things to watch out for, such as a value that ends early or a TEMPLATE that Ollama may not use. Nothing leaves your browser.

{"num_ctx":8192,"stop":["\u003c|im_end|\u003e"],"temperature":0.7}

FROM qwen2.5:7b
PARAMETER num_ctx 8192
PARAMETER temperature 0.7
PARAMETER stop <|im_end|>
SYSTEM "
You are a concise assistant.
Answer in one or two sentences.
"
MESSAGE user Is Toronto in Canada?
MESSAGE assistant yes

A Modelfile looks simple, but small mistakes make ollama create fail with short messages such as "unexpected EOF" or "command must be one of ...". This tool follows Ollama's source code step by step. First it reads the file with the same state machine as Ollama's parser: instructions in any case, values that end at the end of a line unless they start with " or """, PARAMETER names, MESSAGE roles and comments. Then it does what ollama create does before it contacts the server: it stops at ADAPTER and typical_p, checks CAPABILITY and REQUIRES, and converts each PARAMETER to its type the way Ollama's API does (whole numbers, 32-bit floats, true or false, and stop as a list). Last, it checks the TEMPLATE with Go's template rules, as the server does. It shows the parameters as the JSON the new model stores, the example conversation, and the Modelfile as Ollama writes it back. We checked the tool with 43 Modelfiles on Ollama 0.40.11: for every one, it gave the same error as ollama create, or the same stored parameters, system message, template, messages and licenses as the model ollama create made.