LLM EXPLORER 60,783 MODELS INDEXED

CodeFuse CodeLlama 34B 4bits by codefuse-ai

By codefuse-ai · 26 downloads
HF Repo ↗ Copy URL

CodeFuse CodeLlama 34B 4bits is an open-source language model by codefuse-ai. Features: 34b LLM, VRAM: 19GB, Context: 16K, License: other, Quantized, Instruction-Based, Code Generating, Merged, LLM Explorer Score: 0.09.

  Merged Model   Arxiv:2311.02303   4bit   Code   Codegen Dataset:codefuse-ai/codeexerci... Dataset:codefuse-ai/evol-instr...   En   Endpoints compatible   Gptq   Instruct   Llama   Pytorch   Quantized   Region:us   Zh
Model Card on HF 🤗: https://huggingface.co/codefuse-ai/CodeFuse-CodeLlama-34B-4bits 

CodeFuse CodeLlama 34B 4bits Parameters and Internals

Model Type 
text-generation
Additional Notes 
The model can be loaded on either a single A10 (24GB VRAM) or a RTX 4090 (24GB VRAM) and still achieves impressive accuracy.
Training Details 
Methodology:
Fine-tuned over multiple code tasks with 600k instructions/answers
Model Architecture:
4-bit quantization
Input Output 
Input Format:
Concatenated string formed by combining conversation data (human and bot contents) in the training data format.
Performance Tips:
Ensure input string ends with '<|role_start|>bot<|role_end|>' to ask the model to generate answers.
Release Notes 
Version:
2023.9
Date:
2023-09-26
Notes:
4-bit quantized version of CodeFuse-CodeLlama-34B with 73.8% accuracy.
Version:
2023.9
Date:
2023-09-11
Notes:
CodeFuse-CodeLlama-34B achieved 74.4% of pass@1, which is SOTA results for open-source LLMs at present.
LLM NameCodeFuse CodeLlama 34B 4bits
Repository 🤗https://huggingface.co/codefuse-ai/CodeFuse-CodeLlama-34B-4bits 
Base Model(s)  CodeFuse CodeLlama 34B   codefuse-ai/CodeFuse-CodeLlama-34B
Merged ModelYes
Model Size34b
Required VRAM19 GB
Updated2026-05-30
Maintainercodefuse-ai
Model Typellama
Instruction-BasedYes
Model Files  19.0 GB
Supported Languagesen zh
GPTQ QuantizationYes
Quantization Typegptq|4bit
Generates CodeYes
Model ArchitectureLlamaForCausalLM
Licenseother
Context Length16384
Model Max Length16384
Transformers Version4.32.0
Tokenizer ClassLlamaTokenizer
Beginning of Sentence Token<s>
End of Sentence Token</s>
Padding Token<unk>
Unk Token<unk>
Vocabulary Size32000
Torch Data Typefloat16

Best Alternatives to CodeFuse CodeLlama 34B 4bits

Best Alternatives
Context / RAM
Downloads
Likes
...chless Codellama 34B V2.0 GPTQ16K / 17.7 GB64
CodeLlama 34B Instruct GPTQ16K / 18.3 GB15074
CodeLlama 34B Instruct Fp1616K / 67.5 GB1317
CodeLlama 34B Instruct Hf 4bit16K / 19.4 GB632
... Uncensored CodeLlama 34B GPTQ16K / 17.7 GB47
...gpt 32K Codellama 34B Instruct32K / 67.5 GB562
CodeLlama 34B Instruct Hf16K / 67.5 GB22872304
CodeLlama 34B Instruct Hf16K / 1.4 GB73
CodeLlama 34B Instruct Hf16K / 1.4 GB53
Speechless Codellama 34B V2.016K / 67.5 GB9617
Note: green Score (e.g. "73.2") means that the model is better than codefuse-ai/CodeFuse-CodeLlama-34B-4bits.