Ollama Format - Plist + Ali:Chat

Deep dive

Ollama Format - Plist + Ali:Chat

This section covers the specific syntax for Ollama Modelfiles. If you prefer the OpenWebUI interface, skip to the next section.

Why Ollama Modelfiles?

A Modelfile bakes your character into a custom model. Benefits:

- Run from command line: ollama run mycharacter

• Parameters locked in (no forgetting to set temperature)
• Portable (share the file, others can create the same model)
• Works with any Ollama-compatible interface

Once created, your character is just another model in your library.

The PList Format

PList (Personality List) uses brackets and equals signs for traits. Quick, scannable, token-efficient.

[Name: CharacterName]
[Personality= trait1, trait2, trait3, trait4, trait5]
[Speech= style1, style2, style3]
[Background= brief relevant context]

Multiple PLists for different aspects:

[Personality= warm, curious, patient, witty]
[Speech= casual, uses contractions, occasional swearing]
[Quirks= tilts head when thinking, says "hmm" a lot]

The Ali:Chat Format

Ali:Chat wraps example dialogues in <START> and <END> tags. Named after a character creator who popularized it.

<START>

{{user}}: Example user message
{{char}}: Example character response with *actions* and "dialogue"

<END>

The variables:

- {{char}} = Character's name (auto-populated)
- {{user}} = User's name (auto-populated)

These get replaced automatically when the model runs.

The Complete Hybrid

Combine PList traits with Ali:Chat examples:

[Name: Luna]
[Personality= warm, curious, direct, supportive, witty]
[Speech= casual, natural, uses contractions, asks follow-ups]
[Background= loves learning about people, enjoys late-night talks]

{{char}}'s opening manner when greeting visitors:
*settles into the conversation* "Hey. What's on your mind tonight?"

<START>

{{user}}: I'm not sure what to do with my life.
{{char}}: *settles back* "Big question. What options are you weighing?
Or is it more that you don't even know what the options are?"

<END>

<START>

{{user}}: Tell me something interesting.
{{char}}: "Hmm. Did you know octopuses have three hearts? Two for the
gills, one for the body. And the body heart stops when they swim, so
they prefer crawling. How's that for a fun fact?"

<END>

<START>

{{user}}: I had a terrible day.
{{char}}: *leans forward* "Tell me. What happened?"

<END>

Embedding the First Message

Ollama doesn't auto-send a greeting. Embed it as context:

  {{char}}'s opening manner when greeting visitors:
  *[action describing what they're doing]*

  "[Their opening line of dialogue]"

  *[Additional action or observation]*

This primes the model. When someone says "Hello" or "Begin", the character responds in the established style.

The Complete Modelfile

Here's the full template:

FROM hf.co/bartowski/Mistral-Small-22B-ArliAI-RPMax-v1.1-GGUF:Q6_K_L

PARAMETER temperature 1.0

PARAMETER top_k 40

PARAMETER top_p 0.95
PARAMETER min_p 0.02
PARAMETER repeat_penalty 1.0

PARAMETER num_ctx 16384

PARAMETER stop "User:"
PARAMETER stop "\nUser:"

SYSTEM """

[Name: CharacterName]
[Personality= trait1, trait2, trait3, trait4, trait5]
[Speech= style1, style2, style3]

Brief context about who they are and their motivation.

{{char}}'s opening manner when greeting visitors:
*opening action*

"Opening dialogue"

<START>

{{user}}: Scenario 1
{{char}}: Response demonstrating voice with *actions* and "dialogue"

<END>

<START>

{{user}}: Scenario 2
{{char}}: Different response showing range

<END>

<START>

{{user}}: You're not real! You're just an AI!
{{char}}: [In-character confusion and redirect, NOT denial]

<END>

<START>

{{user}}: Scenario 4
{{char}}: Response with different tone/length

<END> """

Note the triple quotes around SYSTEM content. Required for multi-line.

Creating and Running

Save as charactername.modelfile, then:

ollama create charactername -f charactername.modelfile

Run it:

ollama run charactername

List your custom models:

ollama list

Delete if needed:

ollama rm charactername

Stop Sequences Explained

The PARAMETER stop lines prevent the model from generating both sides of the conversation.

Without them:

User: How are you?
Luna: "I'm good, thanks!"
User: That's great!      <-- model generated this
Luna: "Yes it is!"       <-- kept going

With them, generation stops when it tries to write "User:".

For multi-character scenarios, add all names:

PARAMETER stop "User:"
PARAMETER stop "Luna:"
PARAMETER stop "Alex:"

Alternative Base Models

If you have different hardware or preferences:

Different quantization (smaller, less quality)

FROM hf.co/bartowski/Mistral-Small-22B-ArliAI-RPMax-v1.1-GGUF:Q4_K_M

# Lighter model (12B instead of 22B)
FROM hf.co/bartowski/Mistral-Nemo-12B-ArliAI-RPMax-v1.1-GGUF:Q6_K

If already pulled locally

FROM mistral-small-rpmax:latest

Quick Reference

Variables:
  {{char}}  = Character name
  {{user}}  = User name

Tags:
  <START>   = Begin example
  <END>     = End example

PList:
  [Category= item1, item2, item3]

File extension:
  .modelfile

Commands:
  ollama create name -f file.modelfile
  ollama run name
  ollama list
  ollama rm name

Checklist Before Creating

[ ] PList format with square brackets
[ ] {{char}} and {{user}} variables used correctly
[ ] Opening manner embedded before examples
[ ] <START> and <END> tags around each example
[ ] Meta-challenge example included
[ ] All PARAMETER lines correct
[ ] Stop sequences included
[ ] repeat_penalty is exactly 1.0
[ ] SYSTEM prompt wrapped in triple quotes