Add bos_token and add_generation_prompt to the alpaca chat template

[minpeter]

Description

Fixes and resolves issues with the alpaca template used when using the axolotl inference feature.

AS-IS

### Instruction: Describe the structure of an atom.

<Completion starts here>

TO-BE

<bos_token>### Instruction:
Describe the structure of an atom.

### Response:
<Completion starts here>

Motivation and Context

How has this been tested?

Screenshots (if appropriate)

Types of changes

Social Handles (Optional)

discord: minpeter

[NanoCode012]

Thanks for catching that!

[xzuyn]

That template would still have issues. It can't handle system turns, and also leaves \n\n after the last turn.

<bos_token>### Instruction:
<user turn>

### Response:
<assistant turn><eos_token>

### Instruction:
<user turn>

### Response:
<assistant turn><eos_token>

---

This template would fix those. Due to the way the original alpaca format handles system prompts, I've limited it to only having a system prompt on the first turn.

{{ bos_token }}{% for message in messages %}{% if message['role'] == 'system' and loop.first %}{{ message['content'] }}{% elif message['role'] == 'user' %}{{ '### Instruction:\n' + message['content'] }}{% elif message['role'] == 'assistant' %}{{ '### Response:\n' + message['content'] + eos_token }}{% endif %}{% if not loop.last %}{{ '\n\n' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '\n\n### Response:\n' }}{% endif %}

<bos_token><optional system turn>

### Instruction:
<user turn>

### Response:
<assistant turn><eos_token>

### Instruction:
<user turn>

### Response:
<assistant turn><eos_token>

[minpeter]

Please review the changes and proceed with merging..! .cc @NanoCode012

[NanoCode012]

Since we're adding an eos_token as well to separate turns (along others), let me think this through as we're modifying a common template.