List your company

ResearchFormerly WizWand

Qwen

Benchmarks

TaskDatasetSOTA resultTrendResults

Text Watermarking

Qwen3-8B 100 clients

3.5Perplexity (PPL)

5

Language Modeling

Qwen 2.5 3B

0.506HV

5

Language Modeling

Qwen2.5-1.5B

47.9HV

5

AI-generated text detection

Qwen3-235B-A22B-Thinking-2507 generated text business medical legal (test)

99BERT Score

5

LLM Training Performance

Qwen models on RTX 3090 (24GB VRAM) 2.5 (train)

35.09TFLOPS

5

LLM Training Performance

Qwen models on A6000 (48GB VRAM) 2.5 (train)

56.82TFLOPS

5

Training and Verification System Efficiency

Qwen2.5-14B (2048 samples)

114.1Training Time (s)

5

Training Memory Usage Profiling

Qwen3-32B 16×H100s

38.94Memory Footprint (Seq 128K)

5

Multi-path speculative decoding

Qwen held-out (test)

1.31Throughput Ratio Improvement

5

Adversarial Attack

Qwen VL 235B 3

0.65ASR

5

Adversarial Attack

Qwen-3-VL-30B

55ASR

5

Qwen2.5-7B 64K context (train)

1,560.38Throughput (tokens/sec)

5

Jailbreak Attack

Qwen-2

90.83ASR

5

Sparse Autoencoding

Qwen2.5-3B Layer 24 activations (held-out)

0.591Relative Error

4

Sparse Autoencoder Evaluation

Qwen3-8B layer 20

0.989KL Score

4

LLM Training Step Performance

Qwen3-14B S=2048 (base variant)

266.1Latency (ms)

4

Qwen3-32B

1.33Pruning Time (hours)

4

LLM Jailbreaking

Qwen 14B Instruct 2.5

70Success Rate First (SRF)

4

VLM Inference Efficiency

Qwen2.5-VL-7B

643.68Prefill Latency (ms)

4

Hybrid-Dimension Reconfiguration

Qwen3-235B-A22B

4.36Reconfiguration Time (s)

4

Attention Latency

Qwen3-30B-A3B-Instruct-2507 256K context window

1.64Speedup Factor

4

Attention Latency

Qwen3-30B-A3B-Instruct-2507 128K context window

1.56Speedup Factor

4

Attention Latency

Qwen3-30B-A3B-Instruct-2507 64K context window

1.07Speedup Factor

4

Attention Latency

Qwen3-30B-A3B-Instruct-2507 32K context window

1Speedup Factor

4

Object Placement

Qwen (unseen)

78.53Object Count (CNT)

4