Skip to content
NEWLive updates feed from Hugging Face, arXiv and GitHub  See what changed →

Alibaba · Qwen2.5 · based on Qwen/Qwen2.5-32B

Qwen2.5-32B-Instruct

A 32B dense open-weight model under Apache-2.0 with 128K context and broad multilingual coverage; fits on two 48 GB GPUs at INT4.

Apache-2.0Dense128K contextTool calling

2,234,705 downloads · 362 likes · 27,641 GitHub stars · Model card · GitHub · synced 2026-09-19

Specifications

Parameters
32.8B
Architecture
dense · qwen2
Layers
64
Hidden size
5,120
KV heads · head dim
8 · 128
Vocabulary
152,064
Native dtype
bfloat16
Context window
128,000 tokens
Released
2024-09-17
Last modified on Hub
2024-09-25

Features & licensing

License
Apache-2.0
Commercial use
Yes
Modalities
text
Tool / function calling
Yes
Reasoning mode
No
Languages
EN, ZH, UR, AR, FR, DE, ES, JA
Quantised variants on Hub
AWQ, EXL2, FP8, GGUF, GPTQ, bitsandbytes
Pipeline
text-generation
transformerssafetensorsqwen2text-generationchatconversationalentext-generation-inferenceendpoints_compatible

Task fit

Editorial scores (0–100) used by the recommendation engine.

  • Chat85
  • RAG / Q&A86
  • Code80
  • Summarisation84
  • Extraction85
  • Agents78

Hardware to run Qwen2.5-32B-Instruct

Weights need about 72 GB at FP16, 38 GB at INT8 and 22 GB at INT4. The KV cache adds roughly 262 MB per 1,000 tokens per request. For a reference workload of 5 requests per second with 1,500 input and 300 output tokens, total GPU memory is around 35.6 GB, which fits on 1 × NVIDIA A100 (80 GB).

GPU configurations at the reference workload
ConfigurationVRAMUtilisationEst. first tokenCloud / month
2 × NVIDIA L4 (24 GB)48 GB74 %~0.5 s$1,168
1 × NVIDIA L40S (48 GB)48 GB74 %~0.2 s$1,387
1 × NVIDIA A100 (80 GB)recommended80 GB45 %~0.1 s$2,336
1 × NVIDIA H100 (80 GB)80 GB45 %~0.0 s$3,285

Estimates only. Run an assessment for your own traffic.

Qwen2.5-32B-Instruct balances quality and cost for retrieval and chat workloads. Its permissive license makes it a common choice for regulated deployments.