README.md

---
tags:
- w8a8
- int8
- vllm
- vision

license: other
license_name: mrl
inference: false
license_link: https://mistral.ai/licenses/MRL-0.1.md
extra_gated_prompt: >-
  # Mistral AI Research License

  If You want to use a Mistral Model, a Derivative or an Output for any purpose
  that is not expressly authorized under this Agreement, You must request a
  license from Mistral AI, which Mistral AI may grant to You in Mistral AI's
  sole discretion. To discuss such a license, please contact Mistral AI via the
  website contact form: https://mistral.ai/contact/

  ## 1. Scope and acceptance

  **1.1. Scope of the Agreement.** This Agreement applies to any use,
  modification, or Distribution of any Mistral Model by You, regardless of the
  source You obtained a copy of such Mistral Model.

  **1.2. Acceptance.** By accessing, using, modifying, Distributing a Mistral
  Model, or by creating, using or distributing a Derivative of the Mistral
  Model, You agree to be bound by this Agreement.

  **1.3. Acceptance on behalf of a third-party.** If You accept this Agreement
  on behalf of Your employer or another person or entity, You warrant and
  represent that You have the authority to act and accept this Agreement on
  their behalf. In such a case, the word "You" in this Agreement will refer to
  Your employer or such other person or entity.

  ## 2. License

  **2.1. Grant of rights**.  Subject to Section 3 below, Mistral AI hereby
  grants You a non-exclusive, royalty-free, worldwide, non-sublicensable,
  non-transferable, limited license to use, copy, modify, and Distribute under
  the conditions provided in Section 2.2 below, the Mistral Model and any
  Derivatives made by or for Mistral AI and to create Derivatives of the Mistral
  Model.

  **2.2. Distribution of Mistral Model and Derivatives made by or for Mistral
  AI.** Subject to Section 3 below, You may Distribute copies of the Mistral
  Model and/or Derivatives made by or for Mistral AI, under the following
  conditions: You must make available a copy of this Agreement to third-party
  recipients of the Mistral Models and/or Derivatives made by or for Mistral AI
  you Distribute, it being specified that any rights to use the Mistral Models
  and/or Derivatives made by or for Mistral AI shall be directly granted by
  Mistral AI to said third-party recipients pursuant to the Mistral AI Research
  License agreement executed between these parties; You must retain in all
  copies of the Mistral Models the following attribution notice within a
  "Notice" text file distributed as part of such copies: "Licensed by Mistral AI
  under the Mistral AI Research License".

  **2.3. Distribution of Derivatives made by or for You.** Subject to Section 3
  below, You may Distribute any Derivatives made by or for You under additional
  or different terms and conditions, provided that: In any event, the use and
  modification of Mistral Model and/or Derivatives made by or for Mistral AI
  shall remain governed by the terms and conditions of this Agreement; You
  include in any such Derivatives made by or for You prominent notices stating
  that You modified the concerned Mistral Model; and Any terms and conditions
  You impose on any third-party recipients relating to Derivatives made by or
  for You shall neither limit such third-party recipients' use of the Mistral
  Model or any Derivatives made by or for Mistral AI in accordance with the
  Mistral AI Research License nor conflict with any of its terms and conditions.

  ## 3. Limitations

  **3.1. Misrepresentation.** You must not misrepresent or imply, through any
  means, that the Derivatives made by or for You and/or any modified version of
  the Mistral Model You Distribute under your name and responsibility is an
  official product of Mistral AI or has been endorsed, approved or validated by
  Mistral AI, unless You are authorized by Us to do so in writing.

  **3.2. Usage Limitation.** You shall only use the Mistral Models, Derivatives
  (whether or not created by Mistral AI) and Outputs for Research Purposes.

  ## 4. Intellectual Property

  **4.1. Trademarks.** No trademark licenses are granted under this Agreement,
  and in connection with the Mistral Models, You may not use any name or mark
  owned by or associated with Mistral AI or any of its affiliates, except (i) as
  required for reasonable and customary use in describing and Distributing the
  Mistral Models and Derivatives made by or for Mistral AI and (ii) for
  attribution purposes as required by this Agreement.

  **4.2. Outputs.** We claim no ownership rights in and to the Outputs. You are
  solely responsible for the Outputs You generate and their subsequent uses in
  accordance with this Agreement. Any Outputs shall be subject to the
  restrictions set out in Section 3 of this Agreement.

  **4.3. Derivatives.** By entering into this Agreement, You accept that any
  Derivatives that You may create or that may be created for You shall be
  subject to the restrictions set out in Section 3 of this Agreement.

  ## 5. Liability

  **5.1. Limitation of liability.** In no event, unless required by applicable
  law (such as deliberate and grossly negligent acts) or agreed to in writing,
  shall Mistral AI be liable to You for damages, including any direct, indirect,
  special, incidental, or consequential damages of any character arising as a
  result of this Agreement or out of the use or inability to use the Mistral
  Models and Derivatives (including but not limited to damages for loss of data,
  loss of goodwill, loss of expected profit or savings, work stoppage, computer
  failure or malfunction, or any damage caused by malware or security breaches),
  even if  Mistral AI has been advised of the possibility of such damages.

  **5.2. Indemnification.** You agree to indemnify and hold harmless Mistral AI
  from and against any claims, damages, or losses arising out of or related to
  Your use or Distribution of the Mistral Models and Derivatives.

  ## 6. Warranty

  **6.1. Disclaimer.** Unless required by applicable law or prior agreed to by
  Mistral AI in writing, Mistral AI provides the Mistral Models and Derivatives
  on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either
  express or implied, including, without limitation, any warranties or
  conditions of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
  PARTICULAR PURPOSE. Mistral AI does not represent nor warrant that the Mistral
  Models and Derivatives will be error-free, meet Your or any third party's
  requirements, be secure or will allow You or any third party to achieve any
  kind of result or generate any kind of content. You are solely responsible for
  determining the appropriateness of using or Distributing the Mistral Models
  and Derivatives and assume any risks associated with Your exercise of rights
  under this Agreement.

  ## 7. Termination

  **7.1. Term.** This Agreement is effective as of the date of your acceptance
  of this Agreement or access to the concerned Mistral Models or Derivatives and
  will continue until terminated in accordance with the following terms.

  **7.2. Termination.** Mistral AI may terminate this Agreement at any time if
  You are in breach of this Agreement. Upon termination of this Agreement, You
  must cease to use all Mistral Models and Derivatives and shall permanently
  delete any copy thereof. The following provisions, in their relevant parts,
  will survive any termination or expiration of this Agreement, each for the
  duration necessary to achieve its own intended purpose (e.g. the liability
  provision will survive until the end of the applicable limitation
  period):Sections 5 (Liability), 6(Warranty), 7 (Termination) and 8 (General
  Provisions).

  **7.3. Litigation.** If You initiate any legal action or proceedings against
  Us or any other entity (including a cross-claim or counterclaim in a lawsuit),
  alleging that the Model or a Derivative, or any part thereof, infringe upon
  intellectual property or other rights owned or licensable by You, then any
  licenses granted to You under this Agreement will immediately terminate as of
  the date such legal action or claim is filed or initiated.

  ## 8. General provisions

  **8.1. Governing laws.** This Agreement will be governed by the laws of
  France, without regard to choice of law principles, and the UN Convention on
  Contracts for the International Sale of Goods does not apply to this
  Agreement.

  **8.2. Competent jurisdiction.** The courts of Paris shall have exclusive
  jurisdiction of any dispute arising out of this Agreement.

  **8.3. Severability.** If any provision of this Agreement is held to be
  invalid, illegal or unenforceable, the remaining provisions shall be
  unaffected thereby and remain valid as if such provision had not been set
  forth herein.

  ## 9. Definitions

  "Agreement": means this Mistral AI Research License agreement governing the
  access, use, and Distribution of the Mistral Models, Derivatives and Outputs.

  "Derivative": means any (i) modified version of the Mistral Model (including
  but not limited to any customized or fine-tuned version thereof), (ii) work
  based on the Mistral Model, or (iii) any other derivative work thereof.

  "Distribution", "Distributing", "Distribute" or "Distributed": means
  supplying, providing or making available, by any means, a copy of the Mistral
  Models and/or the Derivatives as the case may be, subject to Section 3 of this
  Agreement.

  "Mistral AI", "We" or "Us": means Mistral AI, a French société par actions
  simplifiée registered in the Paris commercial registry under the number 952
  418 325, and having its registered seat at 15, rue des Halles, 75001 Paris.

  "Mistral Model": means the foundational large language model(s), and its
  elements which include algorithms, software, instructed checkpoints,
  parameters, source code (inference code, evaluation code and, if applicable,
  fine-tuning code) and any other elements associated thereto made available by
  Mistral AI under this Agreement, including, if any, the technical
  documentation, manuals and instructions for the use and operation thereof.

  "Research Purposes": means any use of a Mistral Model,  Derivative, or Output
  that is solely for (a) personal, scientific or academic research, and (b) for
  non-profit and non-commercial purposes, and not directly or indirectly
  connected to any commercial activities or business operations. For
  illustration purposes, Research Purposes does not include (1) any usage of the
  Mistral Model, Derivative or Output by individuals or contractors employed in
  or engaged by companies in the context of (a) their daily tasks, or (b) any
  activity (including but not limited to any testing or proof-of-concept) that
  is intended to generate revenue, nor (2) any Distribution by a commercial
  entity of the Mistral Model, Derivative or Output whether in return for
  payment or free of charge, in any medium or form, including but not limited to
  through a hosted or managed service (e.g. SaaS, cloud instances, etc.), or
  behind a software layer.

  "Outputs": means any content generated by the operation of the Mistral Models
  or the Derivatives from  a prompt (i.e., text instructions) provided by users.
  For the avoidance of doubt, Outputs do not include any components of a Mistral
  Models, such as any fine-tuned versions of the Mistral Models, the weights, or
  parameters.

  "You": means the individual or entity entering into this Agreement with
  Mistral AI.


  *Mistral AI processes your personal data below to provide the model and
  enforce its license. If you are affiliated with a commercial entity, we may
  also send you communications about our models. For more information on your
  rights and data handling, please see our <a
  href="https://mistral.ai/terms/">privacy policy</a>.*
extra_gated_fields:
  First Name: text
  Last Name: text
  Country: country
  Affiliation: text
  Job title: text
  I understand that I can only use the model, any derivative versions and their outputs for non-commercial research purposes: checkbox
  I understand that if I am a commercial entity, I am not permitted to use or distribute the model internally or externally, or expose it in my own offerings without a commercial license: checkbox
  I understand that if I upload the model, or any derivative version, on any platform, I must include the Mistral Research License: checkbox
  I understand that for commercial use of the model, I can contact Mistral or use the Mistral AI API on la Plateforme or any of our cloud provider partners: checkbox
  By clicking Submit below I accept the terms of the license and acknowledge that the information I provide will be collected stored processed and shared in accordance with the Mistral Privacy Policy: checkbox
  geo: ip_location
extra_gated_description: >-
  Mistral AI processes your personal data below to provide the model and enforce
  its license. If you are affiliated with a commercial entity, we may also send
  you communications about our models. For more information on your rights and
  data handling, please see our <a href="https://mistral.ai/terms/">privacy
  policy</a>.
extra_gated_button_content: Submit
library_name: vllm
pipeline_tag: image-text-to-text
language:
- en
- fr
- de
- es
- it
- pt
- zh
- ja
- ru
- ko
base_model: neuralmagic/Pixtral-Large-Instruct-2411-hf
---

# Pixtral-Large-Instruct-2411-hf-quantized.w8a8

## Model Overview
- **Model Architecture:** neuralmagic/Pixtral-Large-Instruct-2411-hf
  - **Input:** Vision-Text
  - **Output:** Text
- **Model Optimizations:**
  - **Weight quantization:** INT8
  - **Activation quantization:** INT8
- **Release Date:** 2/24/2025
- **Version:** 1.0
- **Model Developers:** Neural Magic

Quantized version of [neuralmagic/Pixtral-Large-Instruct-2411-hf](https://huggingface.co/neuralmagic/Pixtral-Large-Instruct-2411-hf/tree/main).

### Model Optimizations

This model was obtained by quantizing the weights of [neuralmagic/Pixtral-Large-Instruct-2411-hf](https://huggingface.co/neuralmagic/Pixtral-Large-Instruct-2411-hf/tree/main) to INT8 data type, ready for inference with vLLM >= 0.5.2.

## Deployment

### Use with vLLM

This model can be deployed efficiently using the [vLLM](https://docs.vllm.ai/en/latest/) backend, as shown in the example below.

```python
from vllm.assets.image import ImageAsset
from vllm import LLM, SamplingParams

# prepare model
llm = LLM(
    model="neuralmagic/Pixtral-Large-Instruct-2411-hf-quantized.w8a8",
    trust_remote_code=True,
    max_model_len=4096,
    max_num_seqs=2,
)

# prepare inputs
question = "What is the content of this image?"
inputs = {
    "prompt": f"<|user|>\n<|image_1|>\n{question}<|end|>\n<|assistant|>\n",
    "multi_modal_data": {
        "image": ImageAsset("cherry_blossom").pil_image.convert("RGB")
    },
}

# generate response
print("========== SAMPLE GENERATION ==============")
outputs = llm.generate(inputs, SamplingParams(temperature=0.2, max_tokens=64))
print(f"PROMPT  : {outputs[0].prompt}")
print(f"RESPONSE: {outputs[0].outputs[0].text}")
print("==========================================")
```

vLLM also supports OpenAI-compatible serving. See the [documentation](https://docs.vllm.ai/en/latest/) for more details.

## Creation

This model was created with [llm-compressor](https://github.com/vllm-project/llm-compressor) by running the code snippet below as part a multimodal announcement blog.

<details>
  <summary>Model Creation Code</summary>
  
```python
import requests
import torch
from PIL import Image
from transformers import AutoProcessor
from llmcompressor.modifiers.quantization import GPTQModifier
from llmcompressor.transformers import oneshot
from llmcompressor.transformers.tracing import TraceableLlavaForConditionalGeneration
        
# Load model.
model_id = "neuralmagic/Pixtral-Large-Instruct-2411-hf"
model = TraceableLlavaForConditionalGeneration.from_pretrained(
    model_id, device_map="auto", torch_dtype="auto"
)
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)

# Oneshot arguments
DATASET_ID = "flickr30k"
DATASET_SPLIT = {"calibration": "test[:512]"}
NUM_CALIBRATION_SAMPLES = 512
MAX_SEQUENCE_LENGTH = 2048


# Define a oneshot data collator for multimodal inputs.
def data_collator(batch):
    assert len(batch) == 1
    return {
        "input_ids": torch.LongTensor(batch[0]["input_ids"]),
        "attention_mask": torch.tensor(batch[0]["attention_mask"]),
        "pixel_values": torch.tensor(batch[0]["pixel_values"]),
    }


# Recipe
recipe = [
    GPTQModifier(
        targets="Linear",
        scheme="W8A8",
        sequential_targets=["MistralDecoderLayer"],
        ignore=["re:.*lm_head", "re:vision_tower.*", "re:multi_modal_projector.*"],
    ),
]

SAVE_DIR==f"{model_id.split('/')[1]}-quantized.w8a8"

# Perform oneshot
oneshot(
    model=model,
    tokenizer=model_id,
    dataset=DATASET_ID,
    splits=DATASET_SPLIT,
    recipe=recipe,
    max_seq_length=MAX_SEQUENCE_LENGTH,
    num_calibration_samples=NUM_CALIBRATION_SAMPLES,
    trust_remote_code_model=True,
    data_collator=data_collator,
    output_dir=SAVE_DIR
)
```
</details>

## Evaluation

The model was evaluated using [mistral-evals](https://github.com/neuralmagic/mistral-evals) for vision-related tasks and using [lm_evaluation_harness](https://github.com/neuralmagic/lm-evaluation-harness) for select text-based benchmarks. The evaluations were conducted using the following commands:

<details>
<summary>Evaluation Commands</summary>
  
### Vision Tasks
- vqav2
- docvqa
- mathvista
- mmmu
- chartqa

```
vllm serve neuralmagic/pixtral-12b-quantized.w8a8 --tensor_parallel_size 1 --max_model_len 25000 --trust_remote_code --max_num_seqs 8 --gpu_memory_utilization 0.9 --dtype float16 --limit_mm_per_prompt image=7

python -m eval.run eval_vllm \
        --model_name neuralmagic/pixtral-12b-quantized.w8a8 \
        --url http://0.0.0.0:8000 \
        --output_dir ~/tmp \
        --eval_name <vision_task_name>
```

### Text-based Tasks
#### MMLU
  
```
lm_eval \
  --model vllm \
  --model_args pretrained="<model_name>",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=<n>,gpu_memory_utilization=0.8,enable_chunked_prefill=True,trust_remote_code=True \
  --tasks mmlu \
  --num_fewshot 5 \
  --batch_size auto \
  --output_path output_dir

```

#### MGSM

```
lm_eval \
  --model vllm \
  --model_args pretrained="<model_name>",dtype=auto,max_model_len=4096,max_gen_toks=2048,max_num_seqs=128,tensor_parallel_size=<n>,gpu_memory_utilization=0.9 \
  --tasks mgsm_cot_native \
  --apply_chat_template \
  --num_fewshot 0 \
  --batch_size auto \
  --output_path output_dir

```
</details>


### Accuracy

<table>
  <thead>
    <tr>
      <th>Category</th>
      <th>Metric</th>
      <th>neuralmagic/Pixtral-Large-Instruct-2411-hf</th>
      <th>neuralmagic/Pixtral-Large-Instruct-2411-hf-quantized.w8a8</th>
      <th>Recovery (%)</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td rowspan="6"><b>Vision</b></td>
      <td>MMMU (val, CoT)<br><i>explicit_prompt_relaxed_correctness</i></td>
      <td>63.56</td>
      <td>63.89</td>
      <td>100.52%</td>
    </tr>
    <tr>
      <td>VQAv2 (val)<br><i>vqa_match</i></td>
      <td>79.03</td>
      <td>79.12</td>
      <td>100.11%</td>
    </tr>
    <tr>
      <td>DocVQA (val)<br><i>anls</i></td>
      <td>89.55</td>
      <td>89.80</td>
      <td>100.28%</td>
    </tr>
    <tr>
      <td>ChartQA (test, CoT)<br><i>anywhere_in_answer_relaxed_correctness</i></td>
      <td>82.24</td>
      <td>80.44</td>
      <td>97.81%</td>
    </tr>
    <tr>
      <td>Mathvista (testmini, CoT)<br><i>explicit_prompt_relaxed_correctness</i></td>
      <td>67.3</td>
      <td>66.50</td>
      <td>98.81%</td>
    </tr>
    <tr>
      <td><b>Average Score</b></td>
      <td><b>76.34</b></td>
      <td><b>75.95</b></td>
      <td><b>99.49%</b></td>
    </tr>
    <tr>
      <td rowspan="2"><b>Text</b></td>
      <td>MGSM (CoT)</td>
      <td>76.05</td>
      <td>74.76</td>
      <td>98.30%</td>
    </tr>
    <tr>
      <td>MMLU (5-shot)</td>
      <td>82.8</td>
      <td>82.9</td>
      <td>100.12%</td>
    </tr>
  </tbody>
</table>


## Inference Performance


This model achieves up to 1.87x speedup in single-stream deployment and up to 2.0x speedup in multi-stream asynchronous deployment, depending on hardware and use-case scenario.
The following performance benchmarks were conducted with [vLLM](https://docs.vllm.ai/en/latest/) version 0.7.2, and [GuideLLM](https://github.com/neuralmagic/guidellm).

<details>
<summary>Benchmarking Command</summary>
```
  guidellm --model neuralmagic/Pixtral-Large-Instruct-2411-hf-quantized.w8a8 --target "http://localhost:8000/v1" --data-type emulated --data prompt_tokens=<prompt_tokens>,generated_tokens=<generated_tokens>,images=<num_images>,width=<image_width>,height=<image_height> --max seconds 120 --backend aiohttp_server
```

</details>

### Single-stream performance (measured with vLLM version 0.7.2)

<table border="1" class="dataframe">
  <thead>
    <tr>
      <th></th>
      <th></th>
      <th></th>
      <th></th>
      <th style="text-align: center;" colspan="2" >Document Visual Question Answering<br>1680W x 2240H<br>64/128</th>
      <th style="text-align: center;" colspan="2" >Visual Reasoning <br>640W x 480H<br>128/128</th>
      <th style="text-align: center;" colspan="2" >Image Captioning<br>480W x 360H<br>0/128</th>
    </tr>
    <tr>
      <th>Hardware</th>
      <th>Number of GPUs</th>
      <th>Model</th>
      <th>Average Cost Reduction</th>
      <th>Latency (s)</th>
      <th>Queries Per Dollar</th>
      <th>Latency (s)</th>
      <th>Queries Per Dollar</th>
      <th>Latency (s)</th>
      <th>Queries Per Dollar</th>
    </tr>
  </thead>
  <tbody style="text-align: center">
    <tr>
      <th rowspan="3" valign="top">A100</th>
      <td>4</td>
      <td>neuralmagic/Pixtral-Large-Instruct-2411-hf</td>
      <td></td>
      <td>7.5</td>
      <td>67</td>
      <td>6.5</td>
      <td>77</td>
      <td>6.4</td>
      <td>79</td>
    </tr>
    <tr>
      <td>2</td>
      <td>neuralmagic/Pixtral-Large-Instruct-2411-hf-quantized.w8a8</td>
      <td>1.86</td>
      <td>8.1</td>
      <td>124</td>
      <td>7.1</td>
      <td>142</td>
      <td>6.8</td>
      <td>148</td>
    </tr>
    <tr>
      <td>2</td>
      <td>neuralmagic/Pixtral-Large-Instruct-2411-hf-quantized.w4a16</td>
      <td>2.52</td>
      <td>6.9</td>
      <td>147</td>
      <td>5.1</td>
      <td>199</td>
      <td>4.5</td>
      <td>221</td>
    </tr>
    <tr>
      <th rowspan="3" valign="top">H100</th>
      <td>4</td>
      <td>neuralmagic/Pixtral-Large-Instruct-2411-hf</td>
      <td></td>
      <td>4.4</td>
      <td>67</td>
      <td>3.9</td>
      <td>74</td>
      <td>3.7</td>
      <td>79</td>
    </tr>
    <tr>
      <td>2</td>
      <td>neuralmagic/Pixtral-Large-Instruct-2411-hf-FP8-Dynamic</td>
      <td>1.82</td>
      <td>4.7</td>
      <td>120</td>
      <td>4.1</td>
      <td>137</td>
      <td>3.9</td>
      <td>145</td>
    </tr>
    <tr>
      <td>2</td>
      <td>neuralmagic/Pixtral-Large-Instruct-2411-hf-quantized.w4a16</td>
      <td>1.87</td>
      <td>4.7</td>
      <td>120</td>
      <td>3.9</td>
      <td>144</td>
      <td>3.8</td>
      <td>149</td>
    </tr>
  </tbody>
</table>
 
**Use case profiles: Image Size (WxH) / prompt tokens / generation tokens

**QPD: Queries per dollar, based on on-demand cost at [Lambda Labs](https://lambdalabs.com/service/gpu-cloud) (observed on 2/18/2025).

### Multi-stream asynchronous performance (measured with vLLM version 0.7.2)

<table border="1" class="dataframe">
  <thead>
    <tr>
      <th></th>
      <th></th>
      <th></th>
      <th style="text-align: center;" colspan="2" >Document Visual Question Answering<br>1680W x 2240H<br>64/128</th>
      <th style="text-align: center;" colspan="2" >Visual Reasoning <br>640W x 480H<br>128/128</th>
      <th style="text-align: center;" colspan="2" >Image Captioning<br>480W x 360H<br>0/128</th>
    </tr>
    <tr>
      <th>Hardware</th>
      <th>Model</th>
      <th>Average Cost Reduction</th>
      <th>Maximum throughput (QPS)</th>
      <th>Queries Per Dollar</th>
      <th>Maximum throughput (QPS)</th>
      <th>Queries Per Dollar</th>
      <th>Maximum throughput (QPS)</th>
      <th>Queries Per Dollar</th>
    </tr>
  </thead>
  <tbody style="text-align: center">
   <tr>
      <th rowspan="3" valign="top">A100x4</th>
      <td>neuralmagic/Pixtral-Large-Instruct-2411-hf</td>
      <td></td>
      <td>0.4</td>
      <td>222</td>
      <td>0.7</td>
      <td>341</td>
      <td>0.8</td>
      <td>399</td>
    </tr>
    <tr>
      <td>neuralmagic/Pixtral-Large-Instruct-2411-hf-quantized.w8a8</td>
      <td>1.70</td>
      <td>0.8</td>
      <td>383</td>
      <td>1.1</td>
      <td>571</td>
      <td>1.3</td>
      <td>674</td>
    </tr>
    <tr>
      <td>neuralmagic/Pixtral-Large-Instruct-2411-hf-quantized.w4a16</td>
      <td>1.48</td>
      <td>0.5</td>
      <td>276</td>
      <td>1.0</td>
      <td>505</td>
      <td>1.4</td>
      <td>680</td>
    </tr>
    <tr>
      <<th rowspan="3" valign="top">H100x4</th>
      <td>neuralmagic/Pixtral-Large-Instruct-2411-hf</td>
      <td></td>
      <td>1.0</td>
      <td>284</td>
      <td>1.6</td>
      <td>465</td>
      <td>1.8</td>
      <td>511</td>
    </tr>
    <tr>
      <td>neuralmagic/Pixtral-Large-Instruct-2411-hf-FP8-Dynamic</td>
      <td>1.61</td>
      <td>1.7</td>
      <td>467</td>
      <td>2.6</td>
      <td>726</td>
      <td>3.2</td>
      <td>908</td>
    </tr>
    <tr>
      <td>neuralmagic/Pixtral-Large-Instruct-2411-hf-quantized.w4a16</td>
      <td>1.33</td>
      <td>1.4</td>
      <td>393</td>
      <td>2.2</td>
      <td>726</td>
      <td>2.7</td>
      <td>764</td>
    </tr>
  </tbody>
</table>

**Use case profiles: Image Size (WxH) / prompt tokens / generation tokens

**QPS: Queries per second.

**QPD: Queries per dollar, based on on-demand cost at [Lambda Labs](https://lambdalabs.com/service/gpu-cloud) (observed on 2/18/2025).