Qwen2.5-VL-3B-Instruct VRAM & Memory Requirements
Specific checkpoints with sourced architecture and weight files. GGUF Q4_K_M (Unsloth): 2.20 GiB–2.63 GiB at 8,192 tokens, 1 sequence, FP16 KV, llama.cpp.
This result is an estimate, not a measured trace.Not verifiedFile-backed weights
Publisher and provenance
- PublisherQwen
- Hugging Face
Qwen/Qwen2.5-VL-3B-Instruct - Revision
66285546d2b821cf421d4f5eb2576359d3770cd3 - Total parameters3,754,622,976
- Layers36
- Query heads16
- KV heads2 (gqa)
- Head dimension128
- Recorded context capability128,000
- Completenesscomplete · full
Memory layers in this estimate
Artifact bytes, raw KV, engine extras, and total demand are separate. This is not a runtime trace.
- Weight artifact
- 1.80 GiB (1,929,901,408 bytes)
- Raw KV cache
- 0.28 GiB (301,989,888 bytes)
- Engine extras / workspace
- 0.13 GiB–0.56 GiB
- Total estimated demand
- 2.20 GiB–2.63 GiB
Published artifacts
| Quantization | Selected files | Bytes | Engine | Evidence |
|---|---|---|---|---|
| BF16 safetensors bf16 | model-00001-of-00002.safetensors, model-00002-of-00002.safetensors | 7,509,245,952 (6.99 GiB / 7.51 GB) | vllm, sglang, unspecified | file-metadata |
| GGUF Q4_K_M (Unsloth) gguf-q4_k_m | Qwen2.5-VL-3B-Instruct-Q4_K_M.gguf | 1,929,901,408 (1.80 GiB / 1.93 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q5_K_M (Unsloth) gguf-q5_k_m | Qwen2.5-VL-3B-Instruct-Q5_K_M.gguf | 2,224,813,408 (2.07 GiB / 2.22 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q6_K (Unsloth) gguf-q6_k | Qwen2.5-VL-3B-Instruct-Q6_K.gguf | 2,538,157,408 (2.36 GiB / 2.54 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q8_0 (Unsloth) gguf-q8_0 | Qwen2.5-VL-3B-Instruct-Q8_0.gguf | 3,285,474,656 (3.06 GiB / 3.29 GB) | llama.cpp, ollama, unspecified | file-metadata |
| INT4 theoretical | No published file | 1,877,311,488 (theoretical) | — | theoretical |
Memory by quantization and context
Concurrent sequences = 1. KV defaults to FP16. The range includes documented runtime extras for the listed engine. These figures use the same core as the interactive calculator.
| Quantization | 2,048 tokens | 8,192 tokens | 32,768 tokens |
|---|---|---|---|
| BF16 safetensors | 7.69 GiB–9.57 GiB | 7.90 GiB–9.78 GiB | 8.74 GiB–10.65 GiB |
| GGUF Q4_K_M (Unsloth) | 1.99 GiB–2.41 GiB | 2.20 GiB–2.63 GiB | 3.05 GiB–3.54 GiB |
| GGUF Q5_K_M (Unsloth) | 2.27 GiB–2.69 GiB | 2.48 GiB–2.91 GiB | 3.32 GiB–3.82 GiB |
| GGUF Q6_K (Unsloth) | 2.56 GiB–2.98 GiB | 2.77 GiB–3.21 GiB | 3.61 GiB–4.11 GiB |
| GGUF Q8_0 (Unsloth) | 3.26 GiB–3.69 GiB | 3.47 GiB–3.92 GiB | 4.31 GiB–4.82 GiB |
| INT4 theoretical Theoretical rows are not downloadable artifacts. | 1.94 GiB–2.91 GiB | 2.15 GiB–3.17 GiB | 3.00 GiB–4.21 GiB |
Raw KV contribution
Standard GQA/MHA raw KV is 2 × layers × kvHeads × headDim × tokens × sequences × bytes per element. This page uses KV heads (2), not query heads (16). Using query heads would overstate this cache by 8×.
- 8,192 tokens, 1 sequence, FP16: 301,989,888 bytes (0.2813 GiB)
- 32,768 tokens: 1,207,959,552 bytes (1.1250 GiB)
Related pages
Sources
- Qwen2.5-VL-3B-Instruct model card
- config.json — revision 66285546d2b821cf421d4f5eb2576359d3770cd3
- safetensors index — metadata.total_size=7509245952
- Artifact repo unsloth/Qwen2.5-VL-3B-Instruct-GGUF
Assumptions that affect this result
- Requested context is within the recorded native window of 128000 tokens.
- Full-resident weight estimates use total parameters, including inactive MoE experts.
- Weight bytes come from the selected weight files only, not from every file in the repository.
- Download file size is not peak GPU memory. Runtime layout, allocator padding, and KV are extra.
- Counted files: Qwen2.5-VL-3B-Instruct-Q4_K_M.gguf.
- This checkpoint has a vision component. Text-only inference does not include vision encoder or projector bytes.
- Context budget is the number of cached tokens per sequence (prompt plus reserved generation).
- Concurrent sequences share one model replica; weights are not multiplied by sequence count.
Calculator
Loading catalog snapshot…
Search or pick a catalog checkpoint to see memory demand.
Search or pick a catalog checkpoint to see memory demand.