granite-34b-code-instruct-llamafile

Granite 34B Code Instruct - llamafile

This repository contains executable weights (which we call llamafiles) that run on Linux, MacOS, Windows, FreeBSD, OpenBSD, and NetBSD for AMD64 and ARM64.

• Model creator: IBM
• Original model: ibm-granite/granite-34b-code-instruct
• Base model: ibm-granite/granite-34b-code-base

Granite 34B is a coding model released by IBM in April of 2024.

Quickstart

Assuming your system has at least 64GB of RAM, you can try running the following command which download, concatenate, and execute the model.

wget https://huggingface.co/jartine/granite-34b-code-instruct-llamafile/resolve/main/granite-34b-code-instruct.Q5_0.llamafile
chmod +x granite-34b-code-instruct.Q5_0.llamafile
./granite-34b-code-instruct.Q5_0.llamafile --help   # view manual
./granite-34b-code-instruct.Q5_0.llamafile          # launch web gui + oai api
./granite-34b-code-instruct.Q5_0.llamafile -p ...   # cli interface (scriptable)

Alternatively, you may download an official llamafile executable from Mozilla Ocho on GitHub, in which case you can use the Granite llamafiles as a simple weights data file.

llamafile -m granite-34b-code-instruct.Q5_0.llamafile ...

For further information, please see the llamafile README.

Having trouble? See the "Gotchas" section of the README.

Prompting

The chat template is stored in the GGUF files. From the CLI interface, Mistral style prompts seem to work with this model too:

[INST] {{prompt}} [/INST]

Command template:

./granite-34b-code-instruct.Q5_0.llamafile -p "[INST]{{prompt}}[/INST]"

The maximum context size of this model is 8192 tokens. These llamafiles use a default context size of 512 tokens. Whenever you need the maximum context size to be available with llamafile for any given model, you can pass the -c 0 flag. The default temperature for these llamafiles is 0. It can be changed, e.g. --temp 0.8.

Benchmarks

About Quantization

Our own evaluation of this model leads us to believe that it works best with the Q5_0 and Q8_0 quants. We tried other quantization formats such as Q6_K but it didn't seem to be a good of a fit for this model.

About llamafile

llamafile is a new format introduced by Mozilla Ocho on Nov 20th 2023. It uses Cosmopolitan Libc to turn LLM weights into runnable llama.cpp binaries that run on the stock installs of six OSes for both ARM64 and AMD64.

In addition to being executables, llamafiles are also zip archives. Each llamafile contains a GGUF file, which you can extract using the unzip command. If you want to change or add files to your llamafiles, then the zipalign command (distributed on the llamafile github) should be used instead of the traditional zip command.

---

Granite-34B-Code-Instruct

Model Summary

Granite-34B-Code-Instruct is a 34B parameter model fine tuned from Granite-34B-Code-Base on a combination of permissively licensed instruction data to enhance instruction following capabilities including logical reasoning and problem-solving skills.

• Developers: IBM Research
• GitHub Repository: ibm-granite/granite-code-models
• Paper: Granite Code Models: A Family of Open Foundation Models for Code Intelligence
• Release Date: May 6th, 2024
• License: Apache 2.0.

Usage

Intended use

The model is designed to respond to coding related instructions and can be used to build coding assistants.

<!-- TO DO: Check starcoder2 instruct code example that includes the template https://huggingface.co/bigcode/starcoder2-15b-instruct-v0.1 -->

Generation

This is a simple example of how to use Granite-34B-Code-Instruct model.

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda" # or "cpu"
model_path = "ibm-granite/granite-34b-code-instruct"
tokenizer = AutoTokenizer.from_pretrained(model_path)
# drop device_map if running on CPU
model = AutoModelForCausalLM.from_pretrained(model_path, device_map=device)
model.eval()
# change input text as desired
chat = [
    { "role": "user", "content": "Write a code to find the maximum value in a list of numbers." },
]
chat = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
# tokenize the text
input_tokens = tokenizer(chat, return_tensors="pt")
# transfer tokenized inputs to the device
for i in input_tokens:
    input_tokens[i] = input_tokens[i].to(device)
# generate output tokens
output = model.generate(**input_tokens, max_new_tokens=100)
# decode output tokens into text
output = tokenizer.batch_decode(output)
# loop over the batch to print, in this example the batch size is 1
for i in output:
    print(i)

<!-- TO DO: Check this part -->

Training Data

Granite Code Instruct models are trained on the following types of data.

• Code Commits Datasets: we sourced code commits data from the CommitPackFT dataset, a filtered version of the full CommitPack dataset. From CommitPackFT dataset, we only consider data for 92 programming languages. Our inclusion criteria boils down to selecting programming languages common across CommitPackFT and the 116 languages that we considered to pretrain the code-base model (Granite-34B-Code-Base).
• Math Datasets: We consider two high-quality math datasets, MathInstruct and MetaMathQA. Due to license issues, we filtered out GSM8K-RFT and Camel-Math from MathInstruct dataset.
• Code Instruction Datasets: We use Glaive-Code-Assistant-v3, Glaive-Function-Calling-v2, NL2SQL11 and a small collection of synthetic API calling datasets.
• Language Instruction Datasets: We include high-quality datasets such as HelpSteer and an open license-filtered version of Platypus. We also include a collection of hardcoded prompts to ensure our model generates correct outputs given inquiries about its name or developers.

Infrastructure

We train the Granite Code models using two of IBM's super computing clusters, namely Vela and Blue Vela, both outfitted with NVIDIA A100 and H100 GPUs respectively. These clusters provide a scalable and efficient infrastructure for training our models over thousands of GPUs.

Ethical Considerations and Limitations

Granite code instruct models are primarily finetuned using instruction-response pairs across a specific set of programming languages. Thus, their performance may be limited with out-of-domain programming languages. In this situation, it is beneficial providing few-shot examples to steer the model's output. Moreover, developers should perform safety testing and target-specific tuning before deploying these models on critical applications. The model also inherits ethical considerations and limitations from its base model. For more information, please refer to Granite-34B-Code-Base model card.