Qwen3-32B-FP8-block

Qwen3-32B-FP8-block

Model Overview

• Model Architecture: Qwen3ForCausalLM
  • Input: Text
  • Output: Text
• Model Optimizations:
  • Weight quantization: FP8
  • Activation quantization: FP8
• Release Date:
• Version: 1.0
• Model Developers:: Red Hat

Quantized version of Qwen/Qwen3-32B.

Model Optimizations

This model was obtained by quantizing the weights and activations of Qwen/Qwen3-32B to FP8 data type. This optimization reduces the number of bits per parameter from 16 to 8, reducing the disk size and GPU memory requirements by approximately 50%. Only the weights and activations of the linear operators within transformers blocks of the language model are quantized.

Deployment

Use with vLLM

1. Initialize vLLM server:

vllm serve nm-testing/Qwen3-32B-FP8-block --tensor_parallel_size 4

1. Send requests to the server:

from openai import OpenAI

# Modify OpenAI's API key and API base to use vLLM's API server.
openai_api_key = "EMPTY"
openai_api_base = "http://<your-server-host>:8000/v1"

client = OpenAI(
    api_key=openai_api_key,
    base_url=openai_api_base,
)

model = "nm-testing/Qwen3-32B-FP8-block"

messages = [
    {"role": "user", "content": "Explain quantum mechanics clearly and concisely."},
]


outputs = client.chat.completions.create(
    model=model,
    messages=messages,
)

generated_text = outputs.choices[0].message.content
print(generated_text)

Creation

This model was quantized using the llm-compressor library as shown below.

<details> <summary>Creation details</summary>

from transformers import AutoProcessor, Qwen3ForCausalLM

from llmcompressor import oneshot
from llmcompressor.modeling import replace_modules_for_calibration
from llmcompressor.modifiers.quantization import QuantizationModifier

MODEL_ID = ""nm-testing/Qwen3-32B-FP8-block""

# Load model.
model = Qwen3ForCausalLM.from_pretrained(MODEL_ID, dtype="auto")
processor = AutoProcessor.from_pretrained(MODEL_ID)
model = replace_modules_for_calibration(model)

# Configure the quantization algorithm and scheme.
# In this case, we:
#   * quantize the weights to fp8 with per-block quantization
#   * quantize the activations to fp8 with dynamic token activations
recipe = QuantizationModifier(
    targets="Linear",
    scheme="FP8_BLOCK",
    ignore=["lm_head"],
)

# Apply quantization.
oneshot(model=model, recipe=recipe)

# Save to disk in compressed-tensors format.
SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-FP8-block"
model.save_pretrained(SAVE_DIR)
processor.save_pretrained(SAVE_DIR)

</details>

Evaluation

The model was evaluated on the OpenLLM leaderboard task, using lm-evaluation-harness. vLLM was used for all evaluations.

<details> <summary>Evaluation details</summary>

Openllm V1

lm_eval \
  --model vllm \
  --model_args pretrained="nm-testing/Qwen3-32B-FP8-block",dtype=auto,add_bos_token=True,max_model_len=16384,tensor_parallel_size=4,gpu_memory_utilization=0.9,enable_chunked_prefill=True,trust_remote_code=True \
  --tasks openllm \
  --write_out \
  --batch_size auto \
  --show_config

Openllm V2

lm_eval \
  --model vllm \
  --model_args pretrained="nm-testing/Qwen3-32B-FP8-block",dtype=auto,add_bos_token=False,max_model_len=16384,tensor_parallel_size=4,gpu_memory_utilization=0.7,disable_log_stats=True,enable_chunked_prefill=True,trust_remote_code=True \
  --tasks leaderboard \
  --apply_chat_template \
  --fewshot_as_multiturn \
  --write_out \
  --batch_size auto \
  --show_config

Coding Benchmarks

evalplus.evaluate --model "nm-testing/Qwen3-32B-FP8-block" \
                  --dataset "humaneval" \
                  --backend vllm \
                  --tp 4 \
                  --greedy
evalplus.evaluate --model "nm-testing/Qwen3-32B-FP8-block" \
                --dataset "mbpp" \
                --backend vllm \
                --tp 4 \
                --greedy

</details>

Accuracy

<table> <thead> <tr> <th>Category</th> <th>Metric</th> <th>Qwen/Qwen3-32B</th> <th>nm-testing/Qwen3-32B-FP8-block</th> <th>Recovery (%)</th> </tr> </thead> <tbody> <!-- OpenLLM Leaderboard V1 --> <tr> <td rowspan="7"><b>OpenLLM V1</b></td> <td>ARC-Challenge (Acc-Norm, 25-shot)</td> <td>72.95</td> <td>72.78</td> <td>99.77</td> </tr> <tr> <td>GSM8K (Strict-Match, 5-shot)</td> <td>74.15</td> <td>74.53</td> <td>100.51</td> </tr> <tr> <td>HellaSwag (Acc-Norm, 10-shot)</td> <td>84.03</td> <td>83.86</td> <td>99.80</td> </tr> <tr> <td>MMLU (Acc, 5-shot)</td> <td>81.99</td> <td>81.97</td> <td>99.97</td> </tr> <tr> <td>TruthfulQA (MC2, 0-shot)</td> <td>59.18</td> <td>58.72</td> <td>99.22</td> </tr> <tr> <td>Winogrande (Acc, 5-shot)</td> <td>76.01</td> <td>75.22</td> <td>98.96</td> </tr> <tr> <td><b>Average Score</b></td> <td><b>74.72</b></td> <td><b>74.51</b></td> <td><b>99.72</b></td> </tr> <!-- OpenLLM Leaderboard V2 --> <tr> <td rowspan="7"><b>OpenLLM V2</b></td> <td>IFEval (Inst Level Strict Acc, 0-shot)</td> <td>49.04</td> <td>49.28</td> <td>100.49</td> </tr> <tr> <td>BBH (Acc-Norm, 3-shot)</td> <td>35.27</td> <td>33.94</td> <td>96.21</td> </tr> <tr> <td>Math-Hard (Exact-Match, 4-shot)</td> <td>19.94</td> <td>17.52</td> <td>87.88</td> </tr> <tr> <td>GPQA (Acc-Norm, 0-shot)</td> <td>26.01</td> <td>24.41</td> <td>93.87</td> </tr> <tr> <td>MUSR (Acc-Norm, 0-shot)</td> <td>40.34</td> <td>40.21</td> <td>99.67</td> </tr> <tr> <td>MMLU-Pro (Acc, 5-shot)</td> <td>12.38</td> <td>12.35</td> <td>99.73</td> </tr> <tr> <td><b>Average Score</b></td> <td><b>30.50</b></td> <td><b>29.62</b></td> <td><b>97.11</b></td> </tr> <td rowspan="4" ><strong>Coding</strong> </td> <td>HumanEval pass@1 </td> <td>90.20 </td> <td>90.20 </td> <td>100.00 </td> </tr> <tr> <td>HumanEval+ pass@1 </td> <td>84.80 </td> <td>84.10 </td> <td>98.35 </td> </tr> <tr> <td>MBPP pass@1 </td> <td>86.50 </td> <td>86.20 </td> <td>99.65 </td> </tr> <tr> <td>MBPP+ pass@1 </td> <td>73.00 </td> <td>71.40 </td> <td>97.80 </td> </tr> </tbody> </table>