Qwen2-VL-2B-Instruct

2.21B parameters, images and text in, text out, 32K tokens of context.

Verified 2026-10-25 Rank #125 in LLMs · score 94.9 method v1 Sources 3 Downloads 30d 1.1M · #102 of 619 tracked

Can I run it?1.3GB at INT4Estimatefits a 16 GB cardmethod

Can I use it?Apache 2.0Permissive; patent grantlicence file2026-10-24

Do people use it?1.1Mdownloads, 30 days#102 of 619 we track · top 16%+3 likes, downloads −4% this week528 likesHugging Face2026-10-24

Worth reading

Picked by rule from 10 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Most discussed · 155 HN pts

Llama.cpp Now Supports Qwen2-VL (Vision Language Model)

A pull request adds support for Qwen2VL to llama.cpp, introducing a new command-line tool called llama-qwen2vl-cli for inference.

The vision model's computation graph runs on a single backend using the unsupported RoPE Vulkan OP.
The IM2COL operation is only implemented for f32 and f16 data types on Metal.

github.com 2024-12-14 155 HN pts · 50 comments 954 words about the Qwen2 family family

Latest · 2 years ago

Qwen2-Math

Qwen2-Math is a series of specialized math language models built upon the Qwen2 LLMs that significantly outperforms the mathematical capabilities of open-source models and even closed-source models.

The model series primarily supports English, with bilingual English and Chinese math models scheduled for release soon.
Evaluation of multi-choice benchmarks including MMLU STEM and GaoKao problems uses a 5-shot setting instead of zero-shot.

qwenlm.github.io 2024-08-08 128 HN pts · 38 comments 5.8k words about the Qwen2 family family

10 best of 10 tracked

The 2 picks above, plus these 8. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleQwen2 LLM Releasedqwenlm.github.io · 2024-06-06 · 261 HN pts · official
PaperQwen2-VL: Enhancing Vision-Language Model's Perception of ...arxiv.org · undated
ArticleAn Analysis of Chinese LLM Censorship and Bias with Qwen 2 Instructhuggingface.co · 2024-06-09 · 16 HN pts
ArticleAlibaba Qwen2.5-Omni-7B: Open Source End-to-End Multimodal AI Modelalizila.com · 2025-03-27 · 29 HN pts
ArticleQwen2.5-Max: The Next Leap in AI Innovationmedium.com · 2025-01-29 · 17 HN pts
ArticleRun Qwen2-VL on Your Computer with Text, Images, and ...kaitchup.substack.com · undated
ArticleEvaluating Qwen2-VL-2B-Instruct on Multimodal Reasoningwandb.ai · undated
ArticleQwen2-VL: A Vision Language Model That Runs Locallymedium.com · undated

Show all 8

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

No API provider serves it, so its downloads are people running it themselves: 59 quantized builds and 100 Spaces.

As a service0no API provider listed
On own hardware59quantized builds
Built on571378 fine-tunes · 192 adapters · 1 merges
In apps100+Hugging Face Spaces

Top Spaces running it: gokaygokay/FLUX-Prompt-Generator ♥ 1,400 · showlab/ShowUI ♥ 240 · embedl/hfviewer ♥ 181 · artificialguybr/qwen-vl ♥ 126

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 16 GB card at INT4

About 1.3 GB at INT4, 2.7 GB at FP8, 5.3 GB at BF16: weights only, so leave room for the KV cache.

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

Most coverage came in the first two weeks

Teal is the weights (2024-08-28), amber triangles are 6 dated articles (each links to it), grey ticks are repo commits.

What people built on it

630 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

630 derivatives on the Hub, most of them fine-tunes

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

GGUF bartowski/Qwen2-VL-2B-Instruct-GGUF for llama.cpp, Ollama, LM Studio 208.3k downloads GPTQ Qwen/Qwen2-VL-2B-Instruct-GPTQ-Int4 for vLLM, SGLang 761.3k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

6 rows

Benchmarks 6 rows

measured third-party leaderboard rowauthor-reported number from the model card or eval files (6)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.

Benchmark	Value	Setting / notes	Reported by	Date
author-reported MMStar · Understand images and video	48 percent	—	model author	2026-10-07
author-reported MTVQA	20 percent	—	model author	2026-10-07
author-reported MVBench	63.2 percent	—	model author	2026-10-07
author-reported MathVision · Understand images and video	12.4 percent	—	model author	2026-10-07
author-reported OCRBench · Extract text from documents	794	—	model author	2026-10-07
author-reported RealWorldQA · Understand images and video	62.9 percent	—	model author	2026-10-07

Use-case evidence

4 use cases

What it is good at 4 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Facts

22 fields

Listed in LLMs

Facts

Qwen2-VL-2B-Instruct has 2.21B parameters. (safetensors metadata, observed 2026-10-24)
Context length is 32,768 tokens (32K). (config.json max_position_embeddings, observed 2026-10-24)
License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-24)
Created 2024-08-28, last modified 2025-01-12. (HF API, observed 2026-10-24)

Pipeline tag	image-text-to-text
Library	transformers
Architectures	Qwen2VLForConditionalGeneration
Model type	qwen2_vl
Dtype	bfloat16
Params by dtype	BF16: 2,208,985,600
Size band	M
MoE	no
Context length	32,768 (config.json max_position_embeddings)
Languages	en
Papers	arXiv:2409.12191, arXiv:2308.12966
License	Apache 2.0 (apache-2.0)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · Qwen2VLForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen2VLForConditionalGeneration · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for Qwen2-VL-2B-Instruct: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	5.3 GB · Single GPU (<24 GB)
FP8 (1 byte/param)	2.7 GB · Edge (<4 GB)
INT4 (0.5 bytes/param)	1.3 GB · Edge (<4 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

40 variants

Lineage

Fine-tuned from Qwen/Qwen2-VL-2B (same organization) Family: Qwen/Qwen2 630 derivatives on the Hub

Base model: Qwen/Qwen2-VL-2B

finetune: 378quantized: 59adapter: 192merge: 1

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
jinaai/jina-reranker-m0	finetune	—	2.44B	M	cc-by-nc-4.0	jinaai	120	2.4M
Alibaba-NLP/gme-Qwen2-VL-2B-Instruct	finetune	—	2.21B	M	apache-2.0	Alibaba-NLP	135	758.8k
Qwen/Qwen2-VL-2B-Instruct-GPTQ-Int4	quantized	GPTQ	2.21B	M	apache-2.0	Qwen	34	761.3k
MrLight/dse-qwen2-2b-mrl-v1	finetune	—	—	—	apache-2.0	MrLight	68	654.7k
Qwen/Qwen2-VL-2B-Instruct-AWQ	quantized	AWQ	2.44B	M	apache-2.0	Qwen	35	424.4k
prithivMLmods/Qwen2-VL-OCR-2B-Instruct	finetune	—	2.21B	M	apache-2.0	prithivMLmods	104	353.4k
showlab/ShowUI-2B	finetune	—	—	—	mit	showlab	281	298.4k
numind/NuExtract-2.0-2B	finetune	—	2.21B	M	mit	numind	48	243.6k
unsloth/Qwen2-VL-2B-Instruct-unsloth-bnb-4bit	quantized	-bnb	2.25B	M	apache-2.0	unsloth	7	226.5k
bartowski/Qwen2-VL-2B-Instruct-GGUF	quantized	GGUF	—	—	apache-2.0	bartowski	37	208.3k
lightonai/MonoQwen2-VL-v0.1	adapter	—	—	—	apache-2.0	lightonai	48	175.7k
medieval-data/qwen2-vl-2b-catmus	finetune	—	2.21B	M	—	medieval-data	6	169.1k
unsloth/Qwen2-VL-2B-Instruct-bnb-4bit	quantized	-bnb	2.27B	M	apache-2.0	unsloth	1	91.7k
unsloth/Qwen2-VL-2B-Instruct	finetune	—	2.21B	M	apache-2.0	unsloth	5	89.4k
ggml-org/Qwen2-VL-2B-Instruct-GGUF	quantized	GGUF	—	—	apache-2.0	ggml-org	4	58.6k
JZPeterPan/MedVLM-R1	finetune	—	2.21B	M	apache-2.0	JZPeterPan	19	49.3k
raghavlite/B3_Qwen2_2B	adapter	—	—	—	—	raghavlite	1	43.3k
nanonets/Nanonets-OCR2-1.5B-exp	finetune	—	1.65B	S	apache-2.0	nanonets	51	36.9k
Qwen/Qwen2-VL-2B-Instruct-GPTQ-Int8	quantized	GPTQ	2.21B	M	apache-2.0	Qwen	25	34.8k
mobilint/Qwen2-VL-2B-Instruct	quantized	—	233M	S	apache-2.0	mobilint	0	33k
erax-ai/EraX-VL-2B-V1.5	finetune	—	2.21B	M	apache-2.0	erax-ai	10	29.8k
h2oai/Qwen2-VL-2B-Instruct-GPTQ-Int4	quantized	GPTQ	2.21B	M	apache-2.0	h2oai	0	28.8k
prithivMLmods/Qwen2-VL-OCR2-2B-Instruct	finetune	—	2.21B	M	apache-2.0	prithivMLmods	6	20.7k
OpenSearch-AI/Ops-MM-embedding-v1-2B	finetune	—	2.21B	M	apache-2.0	OpenSearch-AI	17	19.6k
second-state/Qwen2-VL-2B-Instruct-GGUF	quantized	GGUF	—	—	apache-2.0	second-state	2	19.3k
friedrichor/Unite-Base-Qwen2-VL-2B	finetune	—	2.21B	M	apache-2.0	friedrichor	0	14.8k
prithivMLmods/Radiology-Infer-Mini	finetune	—	2.21B	M	apache-2.0	prithivMLmods	14	12.5k
microsoft/GUI-Actor-2B-Qwen2-VL	finetune	—	2.23B	M	mit	microsoft	27	8.9k
litert-community/Qwen2-VL-2B	quantized	—	—	—	apache-2.0	litert-community	3	9.7k
AdaptLLM/biomed-Qwen2-VL-2B-Instruct	finetune	—	2.44B	M	apache-2.0	AdaptLLM	2	8.8k
mradermacher/Qwen2-VL-2B-Instruct-GGUF	quantized	GGUF	—	—	apache-2.0	mradermacher	1	8.2k
mradermacher/Qwen2-VL-2B-Instruct-i1-GGUF	quantized	GGUF	—	—	apache-2.0	mradermacher	0	7.5k
marsh123/UniSE-MLLM	finetune	—	2.21B	M	mit	marsh123	2	6.2k
budaya-lab/Jawi-OCR-Qwen-v1	finetune	—	2.21B	M	—	budaya-lab	1	6.2k
prithivMLmods/Callisto-OCR3-2B-Instruct	finetune	—	2.21B	M	apache-2.0	prithivMLmods	7	5.6k
onnx-community/Qwen2-VL-2B-Instruct	quantized	—	—	—	apache-2.0	onnx-community	11	4.9k
prithivMLmods/LatexMind-2B-Codec	finetune	—	2.21B	M	apache-2.0	prithivMLmods	4	4.7k
prithivMLmods/Qwen2-VL-Ocrtest-2B-Instruct	finetune	—	2.21B	M	apache-2.0	prithivMLmods	5	3.9k
tensorblock/Qwen2-VL-2B-Instruct-GGUF	quantized	GGUF	—	—	apache-2.0	tensorblock	0	4.2k
prithivMLmods/Imgscope-OCR-2B-0527	finetune	—	2.21B	M	apache-2.0	prithivMLmods	5	3.8k

Repository changelog

17 commits

Repository changelog 17 commits

Date	Commit	SHA
2025-01-12	Update README.md	895c3a4
2024-12-06	Update README.md	4759251
2024-09-21	Update README.md	aca7837
2024-09-06	Update generation_config.json	3c86da4
2024-09-03	fix(ckpt)	18a49a7
2024-09-03	add vcr results	77c8a84
2024-09-02	Update README.md	33cac37
2024-08-31	add correct pipeline tag (#2)	f8732c7
2024-08-30	Update README.md (#1)	2aa4870
2024-08-29	Update README.md	096da3b
2024-08-29	Update README.md	1074ef9
2024-08-29	Update README.md	5e2938b
2024-08-29	Update README.md	a623fe7
2024-08-29	Create LICENSE	274514b
2024-08-29	Create README.md	136e0f8
2024-08-28	Initial commit	e28f5d3
2024-08-28	initial commit	567aa36

Articles and coverage

10 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official3

Qwen2 LLM Released

qwenlm.github.io2024-06-06officialauthorityHN front pageabout the Qwen2 familyHN 261 pts · 130 comments4,713 words

Alibaba releases Qwen2, a new series of open-source models featuring five sizes ranging from 0.5B to 72B parameters. The update applies Group Query Attention to all sizes and trains on data in 27 additional languages besides English and Chinese. Specific instruction-tuned versions like Qwen2-7B-Instruct and Qwen2-72B-Instruct support extended context lengths up to 128K tokens, while smaller models adopt the Apache 2.0 license and the largest retains the Qianwen License.written, reviewed, cited

Qwen2-57B-A14B disables tie embedding because large sparse embeddings consume a large proportion of total model parameters.
Instruction-tuned models require the Needle in a Haystack task to verify correct long-context understanding beyond perplexity evaluation.
The Qwen2-72B-Instruct model performs comparably to GPT-4 in safety tests across four categories of multilingual unsafe queries.

Qwen2-Math

qwenlm.github.io2024-08-08officialauthorityHN front pageabout the Qwen2 familyHN 128 pts · 38 comments5,833 words

Qwen2-Math is a series of specialized math language models built upon the Qwen2 LLMs that significantly outperforms the mathematical capabilities of open-source models and even closed-source models. The largest model, Qwen2-Math-72B-Instruct, is evaluated on benchmarks including GSM8K, Math, and MMLU-STEM using few-shot chain-of-thought prompting. Instruction tuning combines a dense reward signal with a binary signal via Rejection Sampling and reinforcement learning with Group Relative Policy Optimization (GRPO) after SFT.written, reviewed, cited

The model series primarily supports English, with bilingual English and Chinese math models scheduled for release soon.
Evaluation of multi-choice benchmarks including MMLU STEM and GaoKao problems uses a 5-shot setting instead of zero-shot.
Performance reporting for complex competition evaluations includes Greedy, Maj@64, RM@64, and RM@256 metrics.

An Analysis of Chinese LLM Censorship and Bias with Qwen 2 Instruct

huggingface.co2024-06-09authorityabout the Qwen2 familyHN 16 pts · 6 comments3,368 words

Qwen 2 Instruct models operate under an Apache 2.0 license but exhibit unique government-mandated alignment with significantly (>80%) fewer refusals in Chinese than English. The deccp dataset documents these behaviors, showing that responses to sensitive topics often include defensive nationalist propaganda or warnings rather than standard refusals. While the models are competitive on benchmarks, their output varies wildly in tone depending on the language used.written, reviewed, cited

Qwen2-72B-Instruct ranks just below Llama-3-70b-Instruct on the ChatArena Leaderboard several days after release.
The deccp dataset contains approximately 95 refusal questions used to evaluate Qwen2 Instructs censorship behavior.

Analysis6

Qwen2-VL: Enhancing Vision-Language Model's Perception of ...

arxiv.orgauthorityabout the Qwen2 family12,379 words

The Qwen2-VL-72B model achieves results comparable to leading models such as GPT-4o and Claude3.5-Sonnet across various multimodal benchmarks. Qwen2-VL is capable of understanding videos over 20 minutes in length, enhancing its ability to perform high-quality video-based question answering.generated, cited

Alibaba Qwen2.5-Omni-7B: Open Source End-to-End Multimodal AI Model

alizila.com2025-03-27about the Qwen2 familyHN 29 pts · 3 comments

Qwen2.5-Max: The Next Leap in AI Innovation

medium.com2025-01-29about the Qwen2 familyHN 17 pts · 4 comments

Run Qwen2-VL on Your Computer with Text, Images, and ...

kaitchup.substack.comabout the Qwen2 family144 words

Evaluating Qwen2-VL-2B-Instruct on Multimodal Reasoning

wandb.ai969 words

Qwen2-VL: A Vision Language Model That Runs Locally

medium.comabout the Qwen2 family

Community1

Llama.cpp Now Supports Qwen2-VL (Vision Language Model)

github.com2024-12-14authorityHN front pageabout the Qwen2 familyHN 155 pts · 50 comments954 words

A pull request adds support for Qwen2VL to llama.cpp, introducing a new command-line tool called llama-qwen2vl-cli for inference. The update includes vision model architecture and specific rope types like GGML_ROPE_TYPE_VISION for multi-resolution processing. Commenters report that llama-server does not yet support this model and users on Mac encounter unsupported IM2COL errors with Q4_k_m quantization.written, reviewed, cited

The vision model's computation graph runs on a single backend using the unsupported RoPE Vulkan OP.
The IM2COL operation is only implemented for f32 and f16 data types on Metal.

Papers

2 papers

Papers 2

arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).

Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution this model

Peng Wang, Shuai Bai, Sinan Tan +16 · 2024-09-18 · PDF · HF paper page (80 upvotes)
Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond this model

Jinze Bai, Shuai Bai, Shusheng Yang +6 · 2023-08-24 · PDF · HF paper page (12 upvotes)

Similar models

3 models

Similar models

Qwen3-VL-2B-Instruct · 2.13B
same size band (M), same task (image-text-to-text), more providers, permissive license
Qwen3.5-2B · 2.27B
same size band (M), same task (image-text-to-text), more providers, permissive license
SmolVLM2-2.2B-Instruct · 2.25B
same size band (M), same task (image-text-to-text), permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:35 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:02 UTC · Last verified 2026-10-25

Follow Qwen2-VL-2B-Instruct

Watch this model

/feeds/models/Qwen--Qwen2-VL-2B-Instruct.xml(RSS 2.0) · every change to Qwen2-VL-2B-Instruct
/feeds/families/Qwen--Qwen2.xml(RSS 2.0) · every change across the Qwen2 family

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.