id: qwen3-4b-instruct-2507
name: Qwen3 4B Instruct 2507
type: base-model
task:
  - text
  - chat
  - summarization
  - code
  - tool-use
  - 1c-rag
language:
  - ru
  - en
source: huggingface
upstream_id: Qwen/Qwen3-4B-Instruct-2507
upstream_url: https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507
license: apache-2.0
status: staging
created_at: 2026-06-18
storage_path: /models/base/qwen3-4b-instruct-2507
format: safetensors
quantization: none
context_length: 262144
deployment_context_length: 32768
vram_required_gb: null
parameters: 4B
base_model: null
adapter_for: null
datasets: []
eval_suites:
  - evals/text/smoke.yaml
deployment:
  target: docker-gpu.cin.su
  runtime: vllm
  served_model_name: qwen3-4b-instruct
  compose: core/deploy/docker-gpu/vllm/compose.yaml
notes: "First base text model for local inference. Start with 32K context to reduce OOM risk; increase after GPU VRAM is confirmed."