What are the hardware requirements for CodeLlama 70B Hf 2.4bpw H6 EXL2?

CodeLlama 70B Hf 2.4bpw H6 EXL2 requires approximately 21.3 GB of VRAM and supports a context window of 16K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed CodeLlama 70B Hf 2.4bpw H6 EXL2 and how large is it?

CodeLlama 70B Hf 2.4bpw H6 EXL2 is developed by LoneStriker, a model with 70b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

Where can I download or evaluate CodeLlama 70B Hf 2.4bpw H6 EXL2?

CodeLlama 70B Hf 2.4bpw H6 EXL2 is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

CodeLlama 70B Hf 2.4bpw H6 EXL2 by LoneStriker — VRAM 21.3GB, 16K context

Name: CodeLlama 70B Hf 2.4bpw H6 EXL2
Author: LoneStriker

CodeLlama 70B Hf 2.4bpw H6 EXL2 is an open-source language model by LoneStriker. Features: 70b LLM, VRAM: 21.3GB, Context: 16K, License: llama2, Quantized, Code Generating, LLM Explorer Score: 0.11.

Arxiv:2308.12950 Code Codegen Endpoints compatible Exl2 Llama Llama2 Pytorch Quantized Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/LoneStriker/CodeLlama-70b-hf-2.4bpw-h6-exl2

CodeLlama 70B Hf 2.4bpw H6 EXL2 Parameters and Internals

Model Type

text generation, code synthesis

Use Cases

Areas:

commercial, research

Applications:

code synthesis, code understanding, Python specific tasks, instruction following

Primary Use Cases:

code assistant, general code generation

Limitations:

Tests have not covered all scenarios, May produce inaccurate or objectionable responses

Considerations:

See Responsible Use Guide

Supported Languages

programming_languages (English, Python)

Training Details

Data Volume:

16k tokens fine-tuning, supports up to 100k tokens at inference

Methodology:

Fine-tuning with transformers

Context Length:

100000

Hardware Used:

A100-80GB GPUs

Model Architecture:

Auto-regressive language model with optimized transformer architecture

Responsible Ai Considerations

Fairness:

Testing conducted in English

Mitigation Strategies:

Perform safety testing and tuning tailored to specific applications

Input Output

Input Format:

Text input only

Output Format:

Text generation only

LLM Name	CodeLlama 70B Hf 2.4bpw H6 EXL2
Repository 🤗	https://huggingface.co/LoneStriker/CodeLlama-70b-hf-2.4bpw-h6-exl2
Model Size	70b
Required VRAM	21.3 GB
Updated	2026-08-03
Maintainer	LoneStriker
Model Type	llama
Model Files	8.5 GB: 1-of-3 8.6 GB: 2-of-3 4.2 GB: 3-of-3
Supported Languages	code
Quantization Type	exl2
Generates Code	Yes
Model Architecture	LlamaForCausalLM
License	llama2
Context Length	16384
Model Max Length	16384
Transformers Version	4.37.1
Tokenizer Class	LlamaTokenizer
Vocabulary Size	32016
Torch Data Type	bfloat16

Rank the CodeLlama 70B Hf 2.4bpw H6 EXL2 capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to CodeLlama 70B Hf 2.4bpw H6 EXL2

Best Alternatives	Context / RAM	Downloads	Likes
...rpreter CL 70B 4.65bpw H6 EXL2	16K / 40.6 GB	5	0
CodeLlama 70B Hf 4bit MLX	16K / 39.1 GB	112	12
...odeLlama 70B Hf 4.0bpw H6 EXL2	16K / 35 GB	4	1
CodeLLaMA 70B Hf Fp16	16K / 138.7 GB	8	2
...Llama 70B Instruct Hf 4bit MLX	4K / 39.1 GB	417	25
...deLlama 70B Python Hf 4bit MLX	4K / 39.1 GB	108	11
...70B Instruct Nf4 Fp16 Upscaled	4K / 138.7 GB	10	1
... 70B Python Hf 2.65bpw H6 EXL2	2K / 23.5 GB	5	3
...70B Instruct Hf 5.0bpw H6 EXL2	2K / 43.6 GB	5	6
...70B Instruct Hf 2.4bpw H6 EXL2	2K / 21.3 GB	6	1

Note: green Score (e.g. "73.2") means that the model is better than LoneStriker/CodeLlama-70b-hf-2.4bpw-h6-exl2.