IBM Granite 8B Open-Source Code Instruction Model - Free Support for Multilingual Code Generation and Explanation

Granite 8b Code Instruct 4k GGUF

Developed by ibm-granite

The IBM Granite 8B Code Instruct model is an AI model focused on code generation and explanation, supporting multiple programming languages and excelling in benchmarks like HumanEval.

Large Language Model OtherOpen Source License:Apache-2.0 #Multilingual Code Generation #Programming Task Optimization #Code Explanation and Repair

Downloads 979

Release Time : 5/30/2024

Model Overview

This model is a fine-tuned code generation model based on granite-8b-code-base-4k, excelling in tasks such as code generation, explanation, and repair across multiple programming languages.

Model Features

Multilingual Code Generation

Supports code generation in multiple programming languages including Python, JavaScript, Java, Go, C++, and Rust.

High Performance

Excels in the HumanEval benchmark, with pass@1 metrics exceeding 50% for multiple languages.

Code Explanation Capability

Can explain the functionality and logic of code, aiding developers in understanding code.

Code Repair Capability

Can identify and fix errors in code.

Model Capabilities

Code Generation

Code Explanation

Code Repair

Multilingual Support

Text Generation

Use Cases

Software Development

Code Autocompletion

Automatically generates code implementations based on function signatures or comments.

Achieved 57.9% pass@1 in HumanEvalSynthesis test for Python.

Code Documentation Generation

Generates explanatory documentation for existing code.

Achieved 53.0% pass@1 in HumanEvalExplain test for Python.

Code Error Fixing

Identifies and fixes errors in code.

Achieved 48.2% pass@1 in HumanEvalFix test for Java.

Programming Education

Programming Learning Assistance

Helps students understand programming concepts and code logic.

🚀 ibm-granite/granite-8b-code-instruct-4k-GGUF

This is the Q4_K_M converted version of the original model, offering efficient text - generation capabilities for various programming languages.

image/png

🚀 Quick Start

This is the Q4_K_M converted version of the original ibm-granite/granite-8b-code-instruct-4k. Refer to the original model card for more details.

📦 Installation

Use with llama.cpp

git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp

# install
make

# run generation
./main -m granite-8b-code-instruct-4k-GGUF/granite-8b-code-instruct.Q4_K_M.gguf -n 128 -p "def generate_random(x: int):" --color

📚 Documentation

Model Information

Property	Details
Pipeline Tag	text-generation
Base Model	ibm-granite/granite-8b-code-base-4k
Inference	false
License	apache-2.0
Library Name	transformers
Tags	code, granite

Datasets

bigcode/commitpackft
TIGER-Lab/MathInstruct
meta-math/MetaMathQA
glaiveai/glaive-code-assistant-v3
glaive-function-calling-v2
bugdaryan/sql-create-context-instruction
garage-bAInd/Open-Platypus
nvidia/HelpSteer

Metrics

code_eval

Model Index

The model granite-8b-code-instruct-4k has the following results on different tasks and datasets:

Task Type	Dataset Name	Metric (pass@1)	Value
text-generation	HumanEvalSynthesis(Python)	pass@1	57.9
text-generation	HumanEvalSynthesis(JavaScript)	pass@1	52.4
text-generation	HumanEvalSynthesis(Java)	pass@1	58.5
text-generation	HumanEvalSynthesis(Go)	pass@1	43.3
text-generation	HumanEvalSynthesis(C++)	pass@1	48.2
text-generation	HumanEvalSynthesis(Rust)	pass@1	37.2
text-generation	HumanEvalExplain(Python)	pass@1	53.0
text-generation	HumanEvalExplain(JavaScript)	pass@1	42.7
text-generation	HumanEvalExplain(Java)	pass@1	52.4
text-generation	HumanEvalExplain(Go)	pass@1	36.6
text-generation	HumanEvalExplain(C++)	pass@1	43.9
text-generation	HumanEvalExplain(Rust)	pass@1	16.5
text-generation	HumanEvalFix(Python)	pass@1	39.6
text-generation	HumanEvalFix(JavaScript)	pass@1	40.9
text-generation	HumanEvalFix(Java)	pass@1	48.2
text-generation	HumanEvalFix(Go)	pass@1	41.5
text-generation	HumanEvalFix(C++)	pass@1	39.0
text-generation	HumanEvalFix(Rust)	pass@1	32.9

📄 License

This project is licensed under the apache-2.0 license.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご