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

Qwen · qwen2 · based on Qwen/Qwen2.5-3B

Qwen2.5-3B-Instruct

Qwen2.5-3B-Instruct by Qwen: a 3.1B-parameter dense open-weight model under the other license with tool calling.

otherDense33K contextTool callingAuto-imported · not yet reviewed

5,090,920 downloads · 572 likes · Model card · synced 2026-09-19

Specifications

Parameters
3.1B
Architecture
dense · qwen2
Layers
36
Hidden size
2,048
KV heads · head dim
2 · 128
Vocabulary
151,936
Native dtype
bfloat16
Context window
32,768 tokens
Released
2024-09-17
Last modified on Hub
2024-09-25

Features & licensing

License
other
Commercial use
No
Modalities
text
Tool / function calling
Yes
Reasoning mode
No
Languages
EN
Quantised variants on Hub
AWQ, FP8, GGUF, GPTQ, bitsandbytes
Pipeline
text-generation
transformerssafetensorsqwen2text-generationchatconversationalentext-generation-inferenceendpoints_compatible

Task fit

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

  • Chat70
  • RAG / Q&A70
  • Code70
  • Summarisation70
  • Extraction70
  • Agents70

Hardware to run Qwen2.5-3B-Instruct

Weights need about 7 GB at FP16, 4 GB at INT8 and 2 GB at INT4. The KV cache adds roughly 37 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 2.6 GB, which fits on 1 × NVIDIA A100 (80 GB).

GPU configurations at the reference workload
ConfigurationVRAMUtilisationEst. first tokenCloud / month
1 × NVIDIA L4 (24 GB)24 GB11 %~0.5 s$584
1 × NVIDIA L40S (48 GB)48 GB5 %~0.2 s$1,387
1 × NVIDIA A100 (80 GB)recommended80 GB3 %~0.1 s$2,336
1 × NVIDIA H100 (80 GB)80 GB3 %~0.0 s$3,285

Estimates only. Run an assessment for your own traffic.

Auto-imported from Hugging Face. Architecture, license and feature data are synced from the model repository; the task-fit scores below are provisional defaults until a curator reviews them.