Alibaba
Qwen3 2507 4B
4B
Parameters
2,3 Go
Q4_K_M file
≥ 3,8 Go
Recommended VRAM
262 144
Max context
Speed by graphics card
Simulate in the “Configurator” →Standard llama-bench run (prompt 512 / generation 128 tokens), model fully in VRAM. Click a card for details. Electricity cost at the default price (0,25 €/kWh).
| Card · Q4_K_M | VRAM | Generation | Prompt | VRAM used | Power | € elec. / M tok | Price |
|---|---|---|---|---|---|---|---|
| RTX 5070 Ti | 16 Go | 224 tok/s | 11 567 | 3,8 Go | 112 W | 0,03 €/M tok | — |
Not measured on these cards yet — click for an estimate:
Technical sheet
- Published
- 2025-08-05
- Architecture
- 36 layers · 8 KV heads · dimension 128
- Attention
- Standard: every layer keeps the whole context in memory.
- Context cache (f16)
- ≈ 141 MB per 1,000 tokens · 4,5 Go for 32,768 tokens
- GGUF file
- bartowski/Qwen_Qwen3-4B-Instruct-2507-GGUF · Qwen_Qwen3-4B-Instruct-2507-Q4_K_M.gguf Download (2,3 Go)
- Official repo
- Qwen/Qwen3-4B-Instruct-2507
Public Hugging Face data (official config.json, GGUF repo). The context cache only counts full-attention layers.