File size: 20,231 Bytes
d1f32a5 424862d d1f32a5 424862d d1f32a5 424862d c1d110c d007dbb 424862d d007dbb 424862d d007dbb 424862d aa59634 424862d d007dbb 424862d aa59634 424862d aa59634 424862d d007dbb 424862d 4b0ecf6 424862d 4b0ecf6 424862d d007dbb 424862d d007dbb 424862d d007dbb 424862d d007dbb 424862d d007dbb 424862d d007dbb 373bda9 d007dbb 424862d d007dbb 424862d d007dbb 424862d 373bda9 424862d c991207 424862d c991207 424862d c991207 424862d c991207 424862d c991207 424862d c991207 424862d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 | ---
library_name: transformers
license: apache-2.0
pipeline_tag: image-text-to-text
language:
- en
- de
- fr
- es
- it
- pt
- hi
- ja
- ko
- zh
- ar
tags:
- vision
- multimodal
- conversational
- multilingual
- native-resolution
---
# North Micro Vision Instruct
North Micro Vision Instruct is a 2.4B-parameter open-weight vision-language model with native-resolution image support, released under the Apache 2.0 license. It is designed as a compact foundation for prototyping, task-specific fine-tuning, and specialized multimodal applications.
Developed by [Cohere](https://cohere.com/).
> **Technical deep dive:** Read the [North Micro Vision technical blog post](https://huggingface.co/blog/CohereLabs/meet-north-micro-vision-instruct) for architecture, training, and evaluation details.
## Highlights
- Native-resolution image processing that preserves aspect ratios and fine visual detail.
- Broad image-understanding capabilities across VQA, captioning, grounding, OCR, charts, and documents.
- Multilingual and multi-image support.
- Compact 2.4B-parameter scale suited to customization and deployment experimentation.
- Apache 2.0-licensed model weights.
## Model Details
| Property | Value |
| --- | --- |
| Model ID | `CohereLabs/North-Micro-Vision-Instruct` |
| Total parameters | 2.4B |
| Language model | 2B parameters |
| Vision encoder | 400M parameters; custom-trained starting from [SigLIP 2 SO400M](https://huggingface.co/google/siglip2-so400m-patch16-384) |
| Inputs | Interleaved text and images |
| Output | Text |
| Languages | English, German, French, Spanish, Italian, Portuguese, Hindi, Japanese, Korean, Chinese, Arabic, and more |
| Tokenizer vocabulary size | 262,144 |
| LM Backbone context window | 128K tokens |
| Multimodal training context | 8K tokens |
| Checkpoint precision | bfloat16 |
| License | Apache 2.0 |
The language backbone supports a 128K-token context window, but the validated operating range for multimodal prompts is up to 8K tokens. Longer multimodal contexts may rely on extrapolation and have not been benchmarked.
## Quickstart
### Installation
Install [PyTorch](https://pytorch.org/get-started/locally/) for your platform first. North Micro Vision requires Transformers 5.16.0, together with `accelerate` for automatic device placement and Pillow for image loading. Until Transformers 5.16.0 is released, install the runtime dependencies and Transformers from source:
```bash
uv pip install accelerate pillow
uv pip install "git+https://github.com/huggingface/transformers.git"
```
Once Transformers 5.16.0 is available on PyPI, install the released package with:
```bash
uv pip install accelerate pillow "transformers==5.16.0"
```
Flash Attention 2 is optional. On supported CUDA systems, install it with:
```bash
uv pip install flash-attn --no-build-isolation
```
If you do not use `uv`, replace `uv pip` with `pip` in the commands above.
### Transformers
The following example loads an image from a URL and asks the model to describe it. Prompts can interleave text with one or more images; for text-only prompts, omit the image entries.
```python
import torch
from transformers import AutoModelForImageTextToText, AutoProcessor
model_id = "CohereLabs/North-Micro-Vision-Instruct"
processor = AutoProcessor.from_pretrained(
model_id,
)
model = AutoModelForImageTextToText.from_pretrained(
model_id,
dtype="auto",
device_map="auto",
)
# To enable Flash Attention 2, load the model with the following settings:
# model = AutoModelForImageTextToText.from_pretrained(
# model_id,
# dtype=torch.bfloat16,
# attn_implementation="flash_attention_2",
# device_map="auto",
# )
image_url = "https://cdn-uploads.huggingface.co/production/uploads/66d732effe6684fc16b12c28/Io_5OCmftsmH-n158ZtPs.png"
messages = [
{
"role": "user",
"content": [
{"type": "image", "url": image_url},
{"type": "text", "text": "What do you see?"},
],
}
]
inputs = processor.apply_chat_template(
messages,
tokenize=True,
add_generation_prompt=True,
return_tensors="pt",
return_dict=True,
).to(model.device)
outputs = model.generate(
**inputs,
max_new_tokens=128,
do_sample=True,
temperature=0.7,
top_p=0.8,
top_k=20,
)
generated_ids = [
output_ids[len(input_ids) :]
for input_ids, output_ids in zip(inputs.input_ids, outputs)
]
response = processor.batch_decode(
generated_ids,
skip_special_tokens=True,
clean_up_tokenization_spaces=False,
)[0]
print(response)
```
The example uses the recommended Transformers sampling settings. For deterministic output, set `do_sample=False` and omit `temperature`, `top_p`, and `top_k`.
## Architecture
North Micro Vision combines a custom-trained 400M-parameter native-resolution vision encoder with an in-house 2B-parameter language model North Micro LLM. The language model follows our Command A+ architecture, interleaving three sliding-window attention layers that use rotary positional embeddings with one global attention layer without positional embeddings. The vision encoder combines 2D RoPE with learned 1D positional embeddings to preserve spatial structure across native-resolution inputs.
The projector maps visual features into the language model's embedding space. Following DeepStack, patch embeddings from multiple vision-encoder layers are injected into corresponding early LLM layers, giving the language model access to visual representations at different levels of abstraction.
*High-level North Micro Vision architecture, consisting of a native-resolution vision encoder, a projector, and a language model.*
### Grounding Coordinates
Bounding boxes are returned as `[x1, y1, x2, y2]` on a normalized 0–1000 scale. Map them back to the original image by scaling each axis:
```python
x1_px = x1 / 1000 * image_width
y1_px = y1 / 1000 * image_height
x2_px = x2 / 1000 * image_width
y2_px = y2 / 1000 * image_height
```
### vLLM
Public vLLM support is coming soon. Until it is available, use Transformers as shown above. The recommended vLLM settings will be:
```python
temperature = 0.7
top_p = 0.8
top_k = 20
min_p = 0.0
presence_penalty = 1.5
repetition_penalty = 1.0
```
## Intended Use
North Micro Vision Instruct is intended for research and development use cases such as:
- Prototyping and task-specific fine-tuning.
- General visual question answering and image captioning.
- Multilingual and multi-image understanding.
- Visual grounding and spatial understanding.
- OCR, chart and document understanding, and structured information extraction.
## Limitations
- The model is intended as a compact foundation for customization rather than a replacement for larger general-purpose chat assistants.
- It is not a reasoning model and has limited math and code-generation capabilities.
- Tool calling and agentic workflows are not supported.
- System prompts are not recommended because the model was not trained with them, although the chat template accepts the `system` role.
- Multimodal training used an 8K-token context; longer contexts have not been validated.
- Native-resolution inputs can increase memory use and latency as image dimensions grow.
## Ecosystem Support
### Fast Inference 🚀
- [MLX-VLM model weights](https://huggingface.co/collections/mlx-community/north-micro-vision) - Community-contributed by Prince Canuma and Neywa.
### Fine-tuning
In partnership with NVIDIA, we're also shipping an AutoModel recipe for North Micro Vision, so developers can fine-tune and deploy it on NVIDIA GPUs right out of the box.
- [NVIDIA AutoModel recipe](https://github.com/NVIDIA-NeMo/Automodel/tree/main/examples/vlm_finetune/cohere_micro_vision) - Fine-tune and deploy North Micro Vision on NVIDIA GPUs.
- [Axolotl fine-tuning support](https://docs.axolotl.ai/docs/models/cohere-north-micro-vision-instruct.html) - Community-supported fine-tuning using the Axolotl framework.
## Benchmark Results
The complete comparison is provided below. We ran vision-language and text-only evaluations with [VLMEvalKit](https://github.com/open-compass/vlmevalkit), capping generation at 1,024 tokens; see the [technical blog post](https://huggingface.co/blog/CohereLabs/meet-north-micro-vision-instruct) for the full methodology.
<table>
<thead>
<tr>
<th></th>
<th style="font-weight: bold; background-color: rgba(127, 127, 127, 0.08);">North-Micro-Vision-Instruct</th>
<th>Ministral-3-3B-Instruct</th>
<th>LFM2.5-VL-1.6B</th>
<th>Phi-3.5-vision-instruct</th>
<th>Gemma-4-E2B-it</th>
<th>Qwen3-VL-2B-Instruct</th>
<th>Qwen3.5-2B-Instruct</th>
<th>SmolVLM2.2B</th>
</tr>
</thead>
<tbody>
<tr>
<th style="text-align: left; font-weight: normal;">Size</th>
<td style="background-color: rgba(127, 127, 127, 0.08);">2.4B</td>
<td>3.8B</td>
<td>1.6B</td>
<td>4.2B</td>
<td>5.1B</td>
<td>2.2B</td>
<td>2.1B</td>
<td>2.2B</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">License</th>
<td style="background-color: rgba(127, 127, 127, 0.08);">Apache 2.0</td>
<td>Apache 2.0</td>
<td>LFM v1.0</td>
<td>MIT</td>
<td>Apache 2.0</td>
<td>Apache 2.0</td>
<td>Apache 2.0</td>
<td>Apache 2.0</td>
</tr>
<tr>
<th colspan="9" style="text-align: left; background-color: rgba(127, 127, 127, 0.16);">General VQA</th>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">MMBench<sub>DEV_EN_V11</sub></th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.687</td>
<td>0.692</td>
<td>0.696</td>
<td>0.731</td>
<td>0.693</td>
<td>0.744</td>
<td>0.760</td>
<td>0.674</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">MMStar</th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.518</td>
<td>0.531</td>
<td>0.508</td>
<td>0.495</td>
<td>0.529</td>
<td>0.506</td>
<td>0.614</td>
<td>0.460</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">RealWorldQA</th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.622</td>
<td>0.583</td>
<td>0.642</td>
<td>0.580</td>
<td>0.507</td>
<td>0.646</td>
<td>0.693</td>
<td>0.567</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">GQA<sub>TestDev_Balanced</sub></th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.574</td>
<td>0.544</td>
<td>0.395</td>
<td>0.650</td>
<td>0.387</td>
<td>0.572</td>
<td>0.539</td>
<td>0.000<sup>‡</sup></td>
</tr>
<tr>
<th colspan="9" style="text-align: left; background-color: rgba(127, 127, 127, 0.16);">Multilingual</th>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">MTL<sub>MMBench_DEV</sub></th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.636</td>
<td>0.674</td>
<td>0.623</td>
<td>0.619</td>
<td>0.648</td>
<td>0.664</td>
<td>0.669</td>
<td>0.454</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">MMMB</th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.728</td>
<td>0.734</td>
<td>0.717</td>
<td>0.686</td>
<td>0.743</td>
<td>0.723</td>
<td>0.745</td>
<td>0.577</td>
</tr>
<tr>
<th colspan="9" style="text-align: left; background-color: rgba(127, 127, 127, 0.16);">Multi-image</th>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">BLINK</th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.527</td>
<td>0.471</td>
<td>0.484</td>
<td>0.561</td>
<td>0.468</td>
<td>0.514</td>
<td>0.563</td>
<td>0.420</td>
</tr>
<tr>
<th colspan="9" style="text-align: left; background-color: rgba(127, 127, 127, 0.16);">Chart / Document / OCR</th>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">ChartQA<sub>Test</sub></th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.808</td>
<td>0.791</td>
<td>0.739</td>
<td>0.821</td>
<td>0.422</td>
<td>0.693</td>
<td>0.775</td>
<td>0.682</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">DocVQA<sub>VAL</sub></th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.921</td>
<td>0.896</td>
<td>0.877</td>
<td>0.860</td>
<td>0.732</td>
<td>0.825</td>
<td>0.926</td>
<td>0.799</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">InfoVQA<sub>VAL</sub></th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.652</td>
<td>0.589</td>
<td>0.627</td>
<td>0.561</td>
<td>0.380</td>
<td>0.622</td>
<td>0.731</td>
<td>0.383</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">OCRBench<sub>v2_en</sub></th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.367</td>
<td>0.414</td>
<td>0.415</td>
<td>0.339</td>
<td>0.435</td>
<td>0.417</td>
<td>0.481</td>
<td>0.304</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">OCRBench</th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.792</td>
<td>0.735</td>
<td>0.802</td>
<td>0.642</td>
<td>0.719</td>
<td>0.751</td>
<td>0.861</td>
<td>0.727</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">AI2D_TEST</th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.775</td>
<td>0.741</td>
<td>0.728</td>
<td>0.790</td>
<td>0.712</td>
<td>0.713</td>
<td>0.752</td>
<td>0.697</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">CharXiv<sub>DQ</sub></th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.600</td>
<td>0.766</td>
<td>0.516</td>
<td>0.637</td>
<td>0.751</td>
<td>0.595</td>
<td>0.761</td>
<td>0.482</td>
</tr>
<tr>
<th colspan="9" style="text-align: left; background-color: rgba(127, 127, 127, 0.16);">STEM</th>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">MMMU<sub>DEV_VAL</sub></th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.329</td>
<td>0.508</td>
<td>0.380</td>
<td>0.432</td>
<td>0.477</td>
<td>0.379</td>
<td>0.474</td>
<td>0.399</td>
</tr>
<tr>
<th colspan="9" style="text-align: left; background-color: rgba(127, 127, 127, 0.16);">Grounding / Counting</th>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">RefCOCO<sub>avg</sub><sup>†</sup></th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.732</td>
<td>0.317</td>
<td>0.581</td>
<td>0.451</td>
<td>0.084</td>
<td>0.304</td>
<td>0.785</td>
<td>0.018</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">CountBench</th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.725</td>
<td>0.737</td>
<td>0.910</td>
<td>0.645</td>
<td>0.534</td>
<td>0.848</td>
<td>0.805</td>
<td>0.764</td>
</tr>
<tr>
<th colspan="9" style="text-align: left; background-color: rgba(127, 127, 127, 0.16);">Robustness / Hallucination</th>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">HallusionBench</th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.615</td>
<td>0.652</td>
<td>0.601</td>
<td>0.585</td>
<td>0.598</td>
<td>0.673</td>
<td>0.655</td>
<td>0.600</td>
</tr>
<tr>
<th colspan="9" style="text-align: left; background-color: rgba(127, 127, 127, 0.16);">Text</th>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">MMLU<sub>test</sub></th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.504</td>
<td>0.660</td>
<td>0.464</td>
<td>0.355</td>
<td>0.692</td>
<td>0.630</td>
<td>0.543</td>
<td>0.084</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">MMLU-Pro<sub>test</sub></th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.307</td>
<td>0.475</td>
<td>0.199</td>
<td>0.286</td>
<td>0.441</td>
<td>0.428</td>
<td>0.298</td>
<td>0.099</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">Multi-If</th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.373</td>
<td>0.470</td>
<td>0.443</td>
<td>0.304</td>
<td>0.687</td>
<td>0.523</td>
<td>0.464</td>
<td>0.236</td>
</tr>
<tr>
<th style="text-align: left; font-weight: normal;">IFEval</th>
<td style="background-color: rgba(127, 127, 127, 0.08);">0.749</td>
<td>0.725</td>
<td>0.776</td>
<td>0.543</td>
<td>0.869</td>
<td>0.734</td>
<td>0.679</td>
<td>0.501</td>
</tr>
</tbody>
</table>
<p><small><sup>†</sup> Averaged over RefCOCO_val, RefCOCO_testA, RefCOCO_testB, RefCOCO+_val, RefCOCO+_testA, RefCOCO+_testB, RefCOCOg_val, RefCOCOg_test.</small></p>
<p><small><sup>‡</sup> SmolVLM2.2B's GQA output was scored as 0.000 under VLMEvalKit's answer-extraction rules.</small></p>
## Citation
```bibtex
@misc{cohere_north_micro_vision_instruct,
title = {{North Micro Vision}: A 2.4B Native-Resolution Vision-Language Model},
url = {https://huggingface.co/blog/CohereLabs/meet-north-micro-vision-instruct},
author = {{Team Cohere}},
month = {August},
year = {2026}
}
```
## Contact
For errors or questions about this model card, contact [Cohere Labs](mailto:labs@cohere.com).
|