---
license: apache-2.0
base_model:
- Qwen/Qwen3-VL-235B-A22B-Instruct
---


# Model Overview

- **Model Architecture:** Qwen3-VL-235B-A22B-Instruct
- **Supported Hardware Microarchitecture:** AMD MI350/MI355
- **ROCm**: 7.0
- **Operating System(s):** Linux
- **Inference Engine:** [vLLM](https://docs.vllm.ai/en/latest/)
- **Model Optimizer:** [AMD-Quark](https://quark.docs.amd.com/latest/index.html)
  - **Weight quantization:** Perchannel, FP8E4M3, Static
  - **Activation quantization:** Pertoken, FP8E4M3, Dynamic
- **Calibration Dataset:** [Pile](https://huggingface.co/datasets/mit-han-lab/pile-val-backup)

This model was built with Qwen3-VL-235B-A22B-Instruct model by applying [AMD-Quark](https://quark.docs.amd.com/latest/index.html) for ptpc quantization.

# Model Quantization

The model was quantized from [Qwen/Qwen3-VL-235B-A22B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Instruct) using [AMD-Quark](https://quark.docs.amd.com/latest/index.html). The weights are quantized to FP8 and activations are quantized to FP8. 

**Quantization scripts:**
```
# pip install amd-quark

from transformers import AutoTokenizer, AutoModelForCausalLM
from quark.torch import ModelQuantizer, export_safetensors
from quark.torch.quantization import FP8E4M3PerChannelSpec
from quark.torch.quantization.config.config import Config, QuantizationConfig

ckpt_path = "Qwen/Qwen3-VL-235B-A22B-Instruct"
exclude_layers = ["lm_head","*mlp.gate", "*.visual.*"]
output_dir = ckpt_path.rstrip("/").split("/")[-1] + "-ptpc"

# Load the original floating-point model
model = AutoModelForCausalLM.from_pretrained(ckpt_path, device_map="auto", torch_dtype="auto", trust_remote_code=True)
model.eval()
tokenizer = AutoTokenizer.from_pretrained(ckpt_path)

# Set the quantization configuration
FP8_PER_CHANNEL_SPEC = FP8E4M3PerChannelSpec(is_dynamic=False, ch_axis=0).to_quantization_spec()
FP8_PER_TOKEN_DYNAMIC_SPEC = FP8E4M3PerChannelSpec(is_dynamic=True, ch_axis=1).to_quantization_spec()
W_FP8_PER_CHANNEL_STATIC_A_FP8_PER_TOKEN_DYNAMIC_CONFIG = QuantizationConfig(input_tensors=FP8_PER_TOKEN_DYNAMIC_SPEC, weight=FP8_PER_CHANNEL_SPEC)
quant_config = Config(global_quant_config=W_FP8_PER_CHANNEL_STATIC_A_FP8_PER_TOKEN_DYNAMIC_CONFIG, exclude=exclude_layers)

# Apply quantization
quantizer = ModelQuantizer(quant_config)
model = quantizer.quantize_model(model)

# Export quantized model
model = quantizer.freeze(model)
export_safetensors(model, output_dir)
tokenizer.save_pretrained(output_dir)
```

### Accuracy

<table>
  <tr>
   <td><strong>Benchmark</strong>
   </td>
   <td><strong>Qwen3-VL-235B-A22B-Instruct </strong>
   </td>
   <td><strong>Qwen3-VL-235B-A22B-Instruct-ptpc(this model)</strong>
   </td>
  </tr>
  <tr>
   <td>GSM8K
   </td>
   <td>0.93
   </td>
   <td>0.925
   </td>
  </tr>
</table>


### Reproduction

Docker: rocm/vllm-private:rocm7.1_ubuntu22.04_vllm0.11.2_ptpc_fp8

The result of GSM8K was obtained using [vLLM](https://docs.vllm.ai/en/latest/).

vllm version: main(0b2549)

aiter version: 0.13.20191203

### GSM8K

```
lm_eval --model vllm \
    --model_args pretrained=/model_path/Qwen/Qwen3-VL-235B-A22B-Instruct-ptpc,add_bos_token=true,tensor_parallel_size=4 \
    --tasks gsm8k \
    --num_fewshot 5 \
    --batch_size auto \
    --limit 200
```

# Deployment
### Use with vLLM

This model can be deployed efficiently using the [vLLM](https://docs.vllm.ai/en/latest/) backend.

## Evaluation
The evaluation results and reproduction script are being prepared.


# License
Modifications Copyright(c) 2025 Advanced Micro Devices, Inc. All rights reserved.