royleibov/granite-3b-code-base-128k-ZipNN-Compressed

Disclaimer and Requirements

This model is a clone of ibm-granite/granite-3b-code-base-128k compressed using ZipNN. Compressed losslessly to 67% its original size, ZipNN saved ~3GB in storage and potentially ~2TB in data transfer monthly.

Requirement

In order to use the model, ZipNN is necessary:

pip install zipnn

Use This Model

# Use a pipeline as a high-level helper
from transformers import pipeline
from zipnn import zipnn_hf

zipnn_hf()

messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe = pipeline("text-generation", model="royleibov/granite-3b-code-base-128k-ZipNN-Compressed")
pipe(messages)

# Load model directly
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from zipnn import zipnn_hf

zipnn_hf()

model = AutoModelForCausalLM.from_pretrained(
    "royleibov/granite-3b-code-base-128k-ZipNN-Compressed", 
    device_map="cuda", 
    torch_dtype="auto", 
    trust_remote_code=True, 
)
tokenizer = AutoTokenizer.from_pretrained("royleibov/granite-3b-code-base-128k-ZipNN-Compressed")

ZipNN

ZipNN also allows you to seemlessly save local disk space in your cache after the model is downloaded.

To compress the cached model, simply run:

python zipnn_compress_path.py safetensors --model royleibov/granite-3b-code-base-128k-ZipNN-Compressed --hf_cache

The model will be decompressed automatically and safely as long as zipnn_hf() is added at the top of the file like in the example above.

To decompress manualy, simply run:

python zipnn_decompress_path.py --model royleibov/granite-3b-code-base-128k-ZipNN-Compressed --hf_cache

Granite-3B-Code-Base-128K

Model Summary

Granite-3B-Code-Base-128K extends the context length of Granite-3B-Code-Base from 2K to 128K with continual pretraining using the original training data but with repository-level file packing and per-language length upsampling, that we found to be critical for long-context pretraining. We adopt an progressive training strategy where we doubled the context window until it reached the desired length of 128K by appropriately adjusting RoPE theta. We trained on 4B tokens total for all stages, which is only 0.1% of Granite-3B-Code-Base's original pre-training data.

• Developers: IBM Research
• GitHub Repository: ibm-granite/granite-code-models
• Paper: Scaling Granite Code Models to 128K Context
• Release Date: July 18th, 2024
• License: Apache 2.0.

Usage

Intended use

Prominent enterprise use cases of LLMs in software engineering productivity with 128K context length support that includes code generation, code explanation, code fixing, generating unit tests, generating documentation, addressing technical debt issues, vulnerability detection, code translation, and more. All Granite Code Base models, including the 3B parameter model, are able to handle these tasks as they were trained on a large amount of code data from 116 programming languages.

Generation

This is a simple example of how to use Granite-3B-Code-Base-128K model.

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from zipnn import zipnn_hf

zipnn_hf()

device = "cuda" # or "cpu"
model_path = "royleibov/granite-3b-code-base-128k-ZipNN-Compressed"
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
input_text = "def generate():"
# tokenize the text
input_tokens = tokenizer(input_text, 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)
# 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)

Training Data

Starting from the base Granite model, this model was further pretrained on repository-level code data with per-language context-length oversampling, allowing it to effectively utilize up to 128K tokens of context. This continued training stage focused on a curated selection of programming languages, such as Python, C, C++, Go, Java, JavaScript, and TypeScript.

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

The use of Large Language Models involves risks and ethical considerations people must be aware of. Regarding code generation, caution is urged against complete reliance on specific code models for crucial decisions or impactful information as the generated code is not guaranteed to work as intended. Granite-3B-Code-Base-128K model is not the exception in this regard. Even though this model is suited for multiple code-related tasks, it has not undergone any safety alignment, there it may produce problematic outputs. Additionally, it remains uncertain whether smaller models might exhibit increased susceptibility to hallucination in generation scenarios by copying source code verbatim from the training dataset due to their reduced sizes and memorization capacities. This aspect is currently an active area of research, and we anticipate more rigorous exploration, comprehension, and mitigations in this domain. Regarding ethics, a latent risk associated with all Large Language Models is their malicious utilization. We urge the community to use Granite-3B-Code-Base-128K model with ethical intentions and in a responsible way.