Llama3.2-3B-Esper2-GGUF

QuantFactory/Llama3.2-3B-Esper2-GGUF

This is quantized version of ValiantLabs/Llama3.2-3B-Esper2 created using llama.cpp

Original Model Card

Esper 2 is a DevOps and cloud architecture code specialist built on Llama 3.2 3b.

• Expertise-driven, an AI assistant focused on AWS, Azure, GCP, Terraform, Dockerfiles, pipelines, shell scripts and more!
• Real world problem solving and high quality code instruct performance within the Llama 3.2 Instruct chat format
• Finetuned on synthetic DevOps-instruct and code-instruct data generated with Llama 3.1 405b.
• Overall chat performance supplemented with generalist chat data.

Try our code-instruct AI assistant Enigma!

Version

This is the 2024-10-03 release of Esper 2 for Llama 3.2 3b.

Esper 2 is also available for Llama 3.1 8b!

Esper 2 will be coming to more model sizes soon :)

Prompting Guide

Esper 2 uses the Llama 3.2 Instruct prompt format. The example script below can be used as a starting point for general chat:

import transformers
import torch

model_id = "ValiantLabs/Llama3.2-3B-Esper2"

pipeline = transformers.pipeline(
    "text-generation",
    model=model_id,
    model_kwargs={"torch_dtype": torch.bfloat16},
    device_map="auto",
)

messages = [
    {"role": "system", "content": "You are an AI assistant."},
    {"role": "user", "content": "Hi, how do I optimize the size of a Docker image?"}
]

outputs = pipeline(
    messages,
    max_new_tokens=2048,
)

print(outputs[0]["generated_text"][-1])

The Model

Esper 2 is built on top of Llama 3.2 3b Instruct, improving performance through high quality DevOps, code, and chat data in Llama 3.2 Instruct prompt style.

Our current version of Esper 2 is trained on DevOps data from sequelbox/Titanium, supplemented by code-instruct data from sequelbox/Tachibana and general chat data from sequelbox/Supernova.

Esper 2 is created by Valiant Labs.

Check out our HuggingFace page for Shining Valiant 2, Enigma, and our other Build Tools models for creators!

Follow us on X for updates on our models!

We care about open source. For everyone to use.

We encourage others to finetune further from our models.