~/tools / Model files and chat templates / Ollama Template Tester

Ollama Template Tester — Run an Ollama TEMPLATE and see the exact prompt

Paste the TEMPLATE of an Ollama model (from ollama show --template or your Modelfile), add a conversation, tools or a prompt, and see the exact text Ollama hands to the model: special tokens, line breaks and all. The tool runs the template the way Ollama does, with Go's template rules, Ollama's data (messages, tool calls, tools, think) and its three ways of running a template. It also tells you whether Ollama will use your TEMPLATE at all or the chat template inside the GGUF file, and when Ollama will refuse a request for tools or thinking. Nothing leaves your browser.

<|im_start|>system
You are a helpful assistant.

# Tools

You may call one or more functions to assist with the user query.

You are provided with function signatures within <tools></tools> XML tags:
<tools>
{"type": "function", "function": {get_weather Get the current weather for a city {object <nil> <nil> [city] {"city":{"type":"string","description":"City name"}}}}}
</tools>

For each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:
<tool_call>
{"name": <function-name>, "arguments": <args-json-object>}
</tool_call><|im_end|>
<|im_start|>user
What is the weather in Tokyo?<|im_end|>
<|im_start|>assistant
<tool_call>
{"name": "get_weather", "arguments": {"city":"Tokyo"}}
</tool_call><|im_end|>
<|im_start|>user
<tool_response>
{"temperature": 20}
</tool_response><|im_end|>
<|im_start|>assistant
It is 20°C in Tokyo.<|im_end|>
<|im_start|>user
Thanks!<|im_end|>
<|im_start|>assistant

curl -s http://localhost:11434/api/chat -d '{"model":"MODEL","messages":[{"role":"system","content":"You are a helpful assistant."},{"role":"user","content":"What is the weather in Tokyo?"},{"role":"assistant","content":"","tool_calls":[{"function":{"name":"get_weather","arguments":{"city":"Tokyo"}}}]},{"role":"tool","content":"{\"temperature\": 20}","tool_name":"get_weather"},{"role":"assistant","content":"It is 20°C in Tokyo."},{"role":"user","content":"Thanks!"}],"tools":[{"type":"function","function":{"name":"get_weather","description":"Get the current weather for a city","parameters":{"type":"object","properties":{"city":{"type":"string","description":"City name"}},"required":["city"]}}}],"stream":false,"_debug_render_only":true}'

Ollama turns a conversation into the text a model reads with the TEMPLATE of the model, written in Go's template language: {{ range .Messages }}, {{ .Role }}, {{ .Content }}, {{ if .Tools }} and so on. A small mistake, such as a missing line break, a {{- that trims too much or a field name that does not exist, changes what the model sees or makes every request fail. This tool runs your template the way Ollama does. It follows Go's text/template rules (trim markers, variables, range, with, comparisons, printf, the error messages), passes the same data Ollama passes (System, Messages with Role, Content, Thinking, ToolCalls, ToolName and ToolCallID, Tools, Think, ThinkLevel, IsThinkSet, Prompt, Suffix, Response) and has Ollama's extra functions (json, currentDate, yesterdayDate, toTypeScriptType). It also copies how Ollama prepares the data: system messages are joined into .System, consecutive messages from the same role are joined with a blank line, the model's SYSTEM is added when the first message is not a system message, and images become [img-0], [img-1] and so on. A template that never uses .Messages runs the older way, once per turn with .Prompt and .Response, and Ollama adds {{ .Response }} to a template that uses neither. Since Ollama 0.40, a model can also carry a chat template inside its GGUF file (tokenizer.chat_template), and Ollama picks one of the two by what each can do: tools, thinking and fill-in-the-middle. Paste the GGUF chat template under Advanced and the tool tells you which one Ollama will use. We checked the tool in three ways: against Go's own text/template (Go 1.26.6) running Ollama's template code on 243 cases, including Ollama's 20 built-in templates, whose test outputs it matches; against Ollama 0.40.11 on this computer with 330 /api/generate requests; and with 49 /api/chat requests to qwen2.5:7b and to models we made with our own TEMPLATEs. The output and the error messages matched.

The core logic of this tool is open source under the MIT License. github.com/seigo5555/toolkitbox-lib/src/ollama-template.ts