Qwen2.5-Coder (7B)

AI model Released Sep 18, 2024

Context window: 32,768 tokens^*
Total parameters: 7.62 B^*
Open weights: Yes^*

Capability & preference

General capability · ECI

A cross-benchmark capability estimate. Index points, not a percentage; intervals show uncertainty.

Qwen2.5-Coder (7B) 107.9^* Qwen2.5-Coder (7B) · 95% interval 98.4–116.3

Run it locally

Inference · memory estimate

Qwen2.5-Coder (7B)

7.62 B^* total parameters · qwen2^* · BF16^*

Published checkpoint tensor payload: 14.2^* GiB. This includes stored copies and buffers; it is separate from the runtime estimates below.

Weight precision Total context Check a GPU Check a Local AI system

Estimated accelerator memory5.1^* GiB

Weights: 3.9^* GiB
KV cache: 0.2^* GiB
Runtime allowance: 1.0^* GiB

One user, full model in accelerator memory.

GPUs with memory headroom

Local AI systems with memory headroom

NVIDIA Jetson Orin Nano Super Developer Kit (8 GB) Planning budget: 6.0 GiB
NVIDIA Jetson AGX Orin Developer Kit (64 GB) Planning budget: 48.0 GiB
AMD Ryzen AI Halo (Ryzen AI Max+ 395, 128 GB) Planning budget: 96.0 GiB
ASUS Ascent GX10 (128 GB) Planning budget: 96.0 GiB
Apple Mac Studio (M5 Max 18-core CPU, 40-core GPU, 128 GB) Planning budget: 96.0 GiB
Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB) Planning budget: 96.0 GiB

Shared-memory systems use a planning budget of 75% of installed memory, leaving 25% for the OS and other allocations. This is an assumption, not a measured GPU allocation. Firmware and driver limits may be lower and may require configuration. Memory matches do not verify a runtime, quantized model artifact or inference speed.

Includes FP16 KV cache, prompt plus generated tokens, quantization metadata and a runtime allowance. GPU matches reserve 10% of VRAM. Memory fit is not a speed or software-compatibility guarantee; quantized weights and runtime support must be available. Cache uses a conservative full-attention upper bound.

Capabilities

Context window: 32,768 tokens^*
Open weights: Yes^*
Model license: apache-2.0^*
Total parameters: 7.62 B^*
Checkpoint tensor payload: 14.19 GiB^*

Input modalities

text^*

Output modalities

text^*

Processing & platform

Architecture: qwen2^*

System configuration

Detailed specifications

Specifications 1 · 4 fields

Access model: Open weights (unrestricted)^*
Open weights: Yes^*
Organization: Alibaba^*
Organization country: China^*

Specifications 2 · 1 fields

Checkpoint memory architecture

Unique parameters: 7,615,616,512^*
Stored tensor elements: 7,615,616,512^*
Checkpoint tensor bytes: 15,231,233,024^*
Checkpoint data type: BF16^*
Architecture: qwen2^*
Context limit: 32,768^*
Transformer layers: 28^*
Key/value heads: 4^*
Attention head dimension: 128^*
Cache layout: full-attention-upper-bound^*

Specifications 3 · 24 fields

Architecture: qwen2^*
Attention dropout: 0.0^*
Activation function: silu^*
Hidden dimension: 3,584^*
Feed-forward dimension: 18,944^*
Maximum window layers: 28^*
Attention heads: 28^*
Transformer layers: 28^*
Key/value heads: 4^*
RMS normalization epsilon: 1.0e-6^*
RoPE base: 1,000,000.0^*
Sliding attention window: 131,072^*
Tied word embeddings: No^*
Checkpoint data type: bfloat16^*
Sliding attention enabled: No^*
Vocabulary size: 152,064^*
Context window: 32,768^* tokens
Checkpoint input modalities: text^*
Checkpoint licence: apache-2.0^*
Checkpoint output modalities: text^*
Total parameters: 7.615616512^* B
Checkpoint tensor bytes: 15,231,233,024^* bytes
Saved tensor elements: 7,615,616,512^* elements
Checkpoint tensor payload: 14.185191^* GiB

* Provisional data