From 6b38b321170e1b1a3653fdbab9d1586dc8668e48 Mon Sep 17 00:00:00 2001
From: Meng Zhang <meng@tabbyml.com>
Date: Wed, 4 Oct 2023 15:04:09 -0700
Subject: [PATCH] docs: add chat_template to model spec

---
 MODEL_SPEC.md | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/MODEL_SPEC.md b/MODEL_SPEC.md
index 32cf371..59a387a 100644
--- a/MODEL_SPEC.md
+++ b/MODEL_SPEC.md
@@ -15,10 +15,11 @@ tokenizer.json
 
 This file provides meta information about the model. An example file appears as follows:
 
-```js
+```json
 {
     "auto_model": "AutoModelForCausalLM",
-    "prompt_template": "<PRE>{prefix}<SUF>{suffix}<MID>"
+    "prompt_template": "<PRE>{prefix}<SUF>{suffix}<MID>",
+    "chat_template":  "<s>{% for message in messages %}{% if message['role'] == 'user' %}{{ '[INST] ' + message['content'] + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ message['content'] + '</s> ' }}{% endif %}{% endfor %}",
 }
 ```
 
@@ -30,6 +31,8 @@ The **prompt_template** field is optional. When present, it is assumed that the
 
 One example for the **prompt_template** is `<PRE>{prefix}<SUF>{suffix}<MID>`. In this format, `{prefix}` and `{suffix}` will be replaced with their corresponding values, and the entire prompt will be fed into the LLM.
 
+The **chat_template** field is optional. When it is present, it is assumed that the model supports an instruct/chat-style interaction, and can be passed to `--chat-model`.
+
 ### tokenizer.json
 This is the standard fast tokenizer file created using [Hugging Face Tokenizers](https://github.com/huggingface/tokenizers). Most Hugging Face models already come with it in repository.