ApX logoApX logo

Qwen2.5-7B

Parameters

7B

Context Length

131K

Modality

Text

Architecture

Dense

KV Cache Rate

64 KB/tok

License

Apache 2.0

Release Date

19 Sept 2024

Knowledge Cutoff

Jun 2024

API Pricing (per 1M)

Self-hosted only

Hardware & System Requirements

Precision

Native

Context

Workload Memory: 6.26 GB

INT4 · 8K context

Base Weights: 3.26 GB

KV Cache: 0.50 GB

Runtime & Graphs: 2.50 GB

Minimum Required Hardware

1x NVIDIA L4 24GB

6.3 / 21.6 GB usable

1x RTX 3060 12GB

6.3 / 10.8 GB usable

Apple M4 24GB

6.3 / 18.0 GB usable

Determine requirements for your specific workload:

Open in VRAM Calculator

Runtime requirements

Checkpoint format

FP16

CUDA

>= 11.8

PyTorch

>= 2.1

Transformers

>= 4.40.1

ArchitecturePrecisionAttention kernel

Blackwell, Hopper

Native FP16

FlashAttention-3

Ada Lovelace, Ampere

Native FP16

FlashAttention-2

Architecture Diagram

Input TokensToken EmbeddingPosition: RoPEHidden: 4.1k · Context: 131K · Vocab: 152.1kx 32 layersRMSNormPre-AttentionGrouped-Query Attention64Q / 8KV headsHead dim: 64+RMSNormPre-FFNFeed-Forward NetworkSwiGLUIntermediate: 18.9k+Final RMSNormOutput Logits

Evaluation Benchmarks

Rank

#224

BenchmarkScoreRank

Professional Knowledge

MMLU Pro

0.563

56

Graduate-Level QA

GPQA

0.364

117

Rankings

Overall Rank

#224

Coding Rank

-

About Qwen2.5-7B

Qwen2.5-7B is a popular open-weights dense model by Alibaba Cloud engineered for advanced reasoning, code synthesis, and structured data extraction. It supports over 29 languages and long-sequence processing up to 131K tokens.

Specifications

Attention

Attention Structure

Grouped-Query Attention

Attention Heads

64

Key-Value Heads

8

Position Embedding

ROPE

RoPE Theta

1,000,000

Normalization

RMS Normalization

Activation Function

SwigLU

Dimensions

Base Precision

FP16

Hidden Dimension Size

4,096

Number of Layers

32

FFN Intermediate Size (Dense)

18,944

Tokenizer

Vocabulary Size

152,064

About Qwen2.5

Qwen2.5 by Alibaba is a family of dense, decoder-only language models available in various sizes, with some variants utilizing Mixture-of-Experts. These models are pretrained on large-scale datasets, supporting extended context lengths and multilingual communication. The family includes specialized models for coding, mathematics, and multimodal tasks, such as vision and audio processing.


Other Qwen2.5 Models