LLM EXPLORER 60,384 MODELS INDEXED

Granite 3B Code Base 128K by ibm-granite

By ibm-granite · 100 downloads
HF Repo ↗ Copy URL

Granite 3B Code Base 128K is an open-source language model by ibm-granite. Features: 3b LLM, VRAM: 7GB, Context: 125K, License: apache-2.0, LLM Explorer Score: 0.13.

  Arxiv:2405.04324   Code   Dataset:bigcode/starcoderdata Dataset:codeparrot/github-code...   Dataset:math-ai/stackmathqa Dataset:open-web-math/open-web...   Granite   Llama   Model-index   Region:us   Safetensors   Sharded   Tensorflow
Model Card on HF 🤗: https://huggingface.co/ibm-granite/granite-3b-code-base-128k 

Granite 3B Code Base 128K Parameters and Internals

Model Type 
text generation, code generation
Use Cases 
Areas:
enterprise, software engineering productivity
Applications:
code generation, code explanation, code fixing, generating unit tests, generating documentation, addressing technical debt issues, vulnerability detection, code translation
Limitations:
Not undergone any safety alignment, Potential susceptibility to hallucination in generation scenarios
Considerations:
Not fully reliable for crucial decisions; ethical and responsible usage recommended.
Additional Notes 
Trained on extensive code datasets across 116 languages.
Supported Languages 
Python (high), C (high), C++ (high), Go (high), Java (high), JavaScript (high), TypeScript (high)
Training Details 
Data Sources:
codeparrot/github-code-clean, bigcode/starcoderdata, open-web-math/open-web-math, math-ai/StackMathQA, bigcode/humanevalpack, repoqa, lcc, repobench
Data Volume:
4B tokens
Methodology:
Continual pretraining with repository-level file packing and per-language length upsampling
Context Length:
128000
Hardware Used:
NVIDIA A100 GPUs, NVIDIA H100 GPUs
Responsible Ai Considerations 
Mitigation Strategies:
Not undergone any safety alignment; active area of research to handle risks of problematic outputs and malicious utilization.
Input Output 
Input Format:
text input (e.g., code snippets)
Accepted Modalities:
text
Output Format:
generated code
Release Notes 
Date:
July 18th, 2024
Notes:
Extended context length from 2K to 128K with continual pretraining.
LLM NameGranite 3B Code Base 128K
Repository 🤗https://huggingface.co/ibm-granite/granite-3b-code-base-128k 
Model Size3b
Required VRAM7 GB
Updated2026-06-25
Maintaineribm-granite
Model Typellama
Model Files  5.0 GB: 1-of-2   2.0 GB: 2-of-2
Model ArchitectureLlamaForCausalLM
Licenseapache-2.0
Context Length128000
Model Max Length128000
Transformers Version4.41.0
Tokenizer ClassGPT2Tokenizer
Padding Token<|endoftext|>
Vocabulary Size49152
Torch Data Typebfloat16

Best Alternatives to Granite 3B Code Base 128K

Best Alternatives
Context / RAM
Downloads
Likes
Nanbeige4.1 3B256K / 7.9 GB208751161
Nanbeige4.1 3B Heretic256K / 7.9 GB14950
Nanbeige4.1 3B Heretic Mxfp4256K / 7.9 GB342
Nanbeige4.1 3B Bf16256K / 7.9 GB473
...anbeige4.1 3B Heretic Bfloat16256K / 7.9 GB331
Nanbeige4.1 3B Heretic256K / 7.9 GB141
Nanbeige4.1 3B Heretic256K / 7.9 GB170
ISA 03 Mini 3B Hybrid Preview256K / 6.5 GB44
ToolMind Web 3B192K / 7.9 GB15628
Llama 3.2 3B Instruct128K / 6.5 GB18295722368
Note: green Score (e.g. "73.2") means that the model is better than ibm-granite/granite-3b-code-base-128k.