Modelpublic

nm-testing/Qwen3-30B-A3B-FP8-block

sourceHugging Faceapache-2.0updated 11mo agoView on Hugging Face

0likes1kdownloads

Model Card

Qwen3-30B-A3B-FP8-block

Model Overview

—Model Architecture: Qwen3MoeForCausalLM
—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/Qwen3-30B-A3B.

Model Optimizations

This model was obtained by quantizing the weights and activations of Qwen/Qwen3-30B-A3B 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/Qwen3-30B-A3B-FP8-block --tensor_parallel_size 4

1.Send requests to the server:

python

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/Qwen3-30B-A3B-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>

python

from transformers import AutoProcessor, Qwen3MoeForCausalLM

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

MODEL_ID = "Qwen/Qwen3-30B-A3B"

# 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/Qwen3-30B-A3B-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/Qwen3-30B-A3B-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/Qwen3-30B-A3B-FP8-block" \
                    --dataset "humaneval" \
                    --backend vllm \
                    --tp 4 \
                    --greedy

  evalplus.evaluate --model "nm-testing/Qwen3-30B-A3B-FP8-block" \
                  --dataset "mbpp" \
                  --backend vllm \
                  --tp 4 \
                  --greedy

</details>

Accuracy

<table> <thead> <tr> <th>Category</th> <th>Metric</th> <th>Qwen/Qwen3-30B-A3B</th> <th>nm-testing/Qwen3-30B-A3B-FP8-block</th> <th>Recovery (%)</th> </tr> </thead> <tbody>  <tr> <td rowspan="7"><b>OpenLLM V1</b></td> <td>ARC-Challenge (Acc-Norm, 25-shot)</td> <td>69.88</td> <td>69.71</td> <td>99.76</td> </tr> <tr> <td>GSM8K (Strict-Match, 5-shot)</td> <td>89.16</td> <td>88.63</td> <td>99.40</td> </tr> <tr> <td>HellaSwag (Acc-Norm, 10-shot)</td> <td>77.61</td> <td>77.44</td> <td>99.78</td> </tr> <tr> <td>MMLU (Acc, 5-shot)</td> <td>79.53</td> <td>79.39</td> <td>99.82</td> </tr> <tr> <td>TruthfulQA (MC2, 0-shot)</td> <td>53.25</td> <td>53.17</td> <td>99.84</td> </tr> <tr> <td>Winogrande (Acc, 5-shot)</td> <td>73.09</td> <td>72.77</td> <td>99.57</td> </tr> <tr> <td><b>Average Score</b></td> <td><b>73.75</b></td> <td><b>73.52</b></td> <td><b>99.69</b></td> </tr>  <tr> <td rowspan="7"><b>OpenLLM V2</b></td> <td>IFEval (Inst Level Strict Acc, 0-shot)</td> <td>47.96</td> <td>48.56</td> <td>101.25</td> </tr> <tr> <td>BBH (Acc-Norm, 3-shot)</td> <td>32.01</td> <td>31.71</td> <td>99.08</td> </tr> <tr> <td>Math-Hard (Exact-Match, 4-shot)</td> <td>19.34</td> <td>17.22</td> <td>89.06</td> </tr> <tr> <td>GPQA (Acc-Norm, 0-shot)</td> <td>24.33</td> <td>26.09</td> <td>107.24</td> </tr> <tr> <td>MUSR (Acc-Norm, 0-shot)</td> <td>39.15</td> <td>39.42</td> <td>100.68</td> </tr> <tr> <td>MMLU-Pro (Acc, 5-shot)</td> <td>23.60</td> <td>21.94</td> <td>92.96</td> </tr> <tr> <td><b>Average Score</b></td> <td><b>31.06</b></td> <td><b>30.82</b></td> <td><b>99.23</b></td> </tr> <td rowspan="4" ><strong>Coding</strong> </td> <td>HumanEval pass@1 </td> <td>93.30 </td> <td>93.90 </td> <td>100.64 </td> </tr> <tr> <td>HumanEval+ pass@1 </td> <td>87.80 </td> <td>88.40 </td> <td>100.68 </td> </tr> <tr> <td>MBPP pass@1 </td> <td>86.00 </td> <td>85.20 </td> <td>99.06 </td> </tr> <tr> <td>MBPP+ pass@1 </td> <td>73.00 </td> <td>73.30 </td> <td>100.41 </td> </tr> </tbody> </table>