Qwen2-VL-72B-Instruct
- Open weights
Prices
0 offerings · prices as listedNo priced offering is listed for this model yet.
Details
- Family
- qwen-vl
Not stated on Alibaba Qwen’s pages: max input, input modalities, output modalities, reasoning, reasoning efforts, reasoning can be off, max thinking budget, tool use, structured output, built-in tools, knowledge cutoff, licence, aliases.
Run locally
Weights: Qwen/Qwen2-VL-72B-Instruct- Parameters 73.4B
- Dense
- Native precision BF16
- Layers 80
- KV cache per token 320 KB
- Config positions 32K (config.json max_position_embeddings)
- Needs
- ≈ 40.6 GB (weights 36.3 + KV cache 2.5 + overhead 1.8)
- Usable
- 21.6 GB (90% of its memory)
- In system memory
- 19.1 GB · 38/80 layers on the GPU (--n-gpu-layers 38)
- Generation speed
- 2–3 tok/s
- First token
- about 2.37–3.55 s for a 2K prompt
| Precision | Weights | Needs at 8K | On your hardware | Longest context |
|---|---|---|---|---|
| BF16 (native)maker’s file | 136.7 GB | ≈ 146.1 | ✕Doesn’t fit | — |
| Q8_0 | 72.6 GB | ≈ 78.8 | ✕Doesn’t fit | — |
| INT8 (GPTQ)maker’s file | 73.1 GB | ≈ 79.3 | ✕Doesn’t fit | — |
| FP8 | 68.4 GB | ≈ 74.3 | ✕Doesn’t fit | — |
| Q6_K | 56.1 GB | ≈ 61.5 | ✕Doesn’t fit | — |
| Q5_K_M | 48.6 GB | ≈ 53.6 | ✕Doesn’t fit | — |
| Q5_0 | 47.0 GB | ≈ 51.9 | ✕Doesn’t fit | — |
| Q4_K_M | 41.5 GB | ≈ 46.1 | ✕Doesn’t fit | — |
| Q4_0 | 38.5 GB | ≈ 42.9 | ○Needs system memory 2–3 tok/s | — |
| INT4 (AWQ / GPTQ)maker’s file | 40.0 GB | ≈ 44.5 | ○Needs system memory 2–3 tok/s | — |
| IQ4_XS | 36.3 GB | ≈ 40.6 | ○Needs system memory 2–3 tok/s | — |
| Q3_K_Mquality drops | 33.4 GB | ≈ 37.6 | ○Needs system memory 2–4 tok/s | — |
| IQ3_XXSquality drops | 26.1 GB | ≈ 30.0 | ○Needs system memory 4–6 tok/s | — |
| Q2_Kquality drops | 25.6 GB | ≈ 29.4 | ○Needs system memory 4–7 tok/s | — |
| IQ2_XXSquality drops | 17.6 GB | ≈ 21.0 | ●Runs well 23–35 tok/s | 10K |
Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.
Compared with other hardware
Each at the highest precision it runs. Click one for details.
Smallest that runs it (Q4_K_M, 8K): GPU NVIDIA H100 SXM; unified memory Apple M5 Max (40-core GPU) · 64 GB.
Run commands
The maker publishes no files at this precision. Choose its native precision or one of its own builds, or quantise it yourself with llama.cpp.
Memory to fine-tune
Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.
Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated
Cited scores
From third-party leaderboards, not measured by usThis model isn’t on any leaderboard we cite yet.
Changes
All changes →- New language model
Sources
- Qwen/Qwen2-VL-72B-Instruct config.json (opens in a new tab) huggingface.co · checked Oct 31, 2026
- Qwen/Qwen2-VL-72B-Instruct model card (opens in a new tab) huggingface.co · checked Oct 31, 2026
- Qwen/Qwen2-VL-72B-Instruct-AWQ files (opens in a new tab) huggingface.co · checked Oct 31, 2026
- Qwen/Qwen2-VL-72B-Instruct-GPTQ-Int4 files (opens in a new tab) huggingface.co · checked Oct 31, 2026
- Qwen/Qwen2-VL-72B-Instruct-GPTQ-Int8 files (opens in a new tab) huggingface.co · checked Oct 31, 2026
- Qwen/Qwen2-VL-72B-Instruct files (opens in a new tab) huggingface.co · checked Oct 31, 2026
Alibaba Qwen’s official policies: data use, privacy, terms →
Some values on this page are still waiting to be checked against official pages: a filled dot is verified, a hollow one is not yet (with or without a source on file). How we verify data
Alibaba Qwen · China · qwen.ai
Related models
Snapshots, the same family and other models from this provider