Dive into the transformative world of Qwen AI and discover why it’s making waves in AI research and real-world applications—from shattering MMLU benchmarks with scores up to 84.2% for Qwen2-72B-Instruct, to training on trillions of tokens, boasting diverse architecture and parameter configurations, and amassing millions of downloads, over 1,000 GitHub forks, and deployments in 50+ commercial apps.
Key Takeaways
- Qwen-72B achieves 73.5% on MMLU benchmark
- Qwen1.5-72B-Instruct scores 80.5% on MMLU
- Qwen2-72B-Instruct reaches 84.2% on MMLU 5-shot
- Qwen-72B has 72 billion parameters
- Qwen1.5-110B contains 110 billion parameters
- Qwen2-72B features 72 billion parameters
- Qwen trained on over 2 trillion tokens
- Qwen1.5 pre-trained on 7 trillion tokens including multilingual data
- Qwen2-72B trained on 7+ trillion high-quality tokens
- Qwen excels in 29 languages with C-Eval score of 85.2% for Qwen-72B
- Qwen1.5-72B achieves 81.7% on MultiICL benchmark
- Qwen2-72B scores 74.5% on MGSM multilingual math
- Qwen model repository has over 50 million downloads on Hugging Face
- Qwen2 series garnered 10 million downloads in first month
- Qwen1.5-7B has 15 million total downloads
Qwen models have high MMLU scores, high params, and big downloads.
Benchmark Performance
1Qwen-72B achieves 73.5% on MMLU benchmark
Verified2Qwen1.5-72B-Instruct scores 80.5% on MMLU
Verified3Qwen2-72B-Instruct reaches 84.2% on MMLU 5-shot
Verified4Qwen-7B gets 62.4% on MMLU
Directional5Qwen1.5-32B scores 78.1% on MMLU
Single source6Qwen2-7B-Instruct achieves 70.5% on MMLU
Verified7Qwen1.5-110B-Instruct hits 82.4% on MMLU
Verified8Qwen2-1.5B-Instruct scores 65.9% on MMLU
Verified9Qwen-14B reaches 68.2% on MMLU
Directional10Qwen1.5-7B-Instruct gets 74.2% on MMLU
Single source11Qwen2-72B scores 82.8% on MMLU
Verified12Qwen1.5-1.8B achieves 67.9% on MMLU
Verified13Qwen-1.8B hits 58.6% on MMLU
Verified14Qwen2-0.5B-Instruct reaches 52.4% on MMLU
Directional15Qwen1.5-4B scores 71.2% on MMLU
Single source16Qwen-72B-Instruct gets 76.8% on MMLU
Verified17Qwen2-7B reaches 70.5% on MMLU 5-shot
Verified18Qwen1.5-14B-Instruct scores 77.5% on MMLU
Verified19Qwen-7B-Instruct achieves 64.1% on MMLU
Directional20Qwen2-1.5B scores 65.9% on MMLU
Single source21Qwen1.5-0.5B-Instruct hits 54.3% on MMLU
Verified22Qwen-14B-Instruct gets 69.7% on MMLU
Verified23Qwen2-72B-Instruct scores 84.2% on MMLU-Pro
Verified24Qwen1.5-72B scores 80.5% on MMLU-Redux
Directional25Qwen-1.8B-Instruct achieves 60.2% on MMLU
Single sourceBenchmark Performance Interpretation
Qwen’s models show a clear upward trend, with newer versions—like the Qwen2-72B-Instruct, which scores 84.2% on MMLU (and 84.2% on the Pro version)—outperforming earlier ones, ranging from tiny 0.5B models (52.4%) to the original 7B (62.4%), while larger models such as the Qwen1.5-110B-Instruct (82.4%) and Qwen2-72B (82.8%) balance scale with progress, crafting a neat, measurable arc of growth across size and iteration.
Community and Adoption
1Qwen model repository has over 50 million downloads on Hugging Face
Verified2Qwen2 series garnered 10 million downloads in first month
Verified3Qwen1.5-7B has 15 million total downloads
Verified4Qwen ranks top 5 on LMSYS Chatbot Arena with Elo 1285 for 72B
Directional5Over 1000 forks on GitHub QwenLM repo
Single source6Qwen-72B-Instruct used in 500+ Hugging Face spaces
Verified7Qwen2-72B tops Open LLM Leaderboard v2
Verified8Qwen1.5 series deployed on Alibaba Cloud by 1 million users
Verified9Qwen GitHub stars exceed 20,000
Directional10Qwen2 released June 2024 with 5 million inferences on DashScope
Single source11Qwen-7B downloaded 8 million times on HF
Verified12Qwen1.5-72B ranks #1 open model on MT-Bench
Verified13Over 200 community fine-tunes of Qwen on HF
Verified14Qwen2-7B-Instruct Arena Elo 1260
Directional15Qwen adopted in 50+ commercial apps via ModelScope
Single source16Qwen-14B has 2 million downloads
Verified17Qwen1.5-32B used in 300+ papers citing it
Verified18Qwen2-0.5B lightweight model with 1M+ downloads
Verified19Qwen repo contributors over 50
Directional20Qwen1.5-110B preview accessed by 100K developers
Single source21Qwen-1.8B mobile deployments exceed 500K
Verified22Qwen2 multilingual variants starred 10K times
Verified23Qwen overall HF models viewed 100 million times
Verified24Qwen-VL released with 1M image-text pairs training
DirectionalCommunity and Adoption Interpretation
Qwen, the AI model family, has surged to acclaim, with over 50 million Hugging Face downloads (including 15 million for Qwen1.5-7B, 8 million for Qwen7B, and 10 million in Qwen2’s first month), ranking top 5 on LMSYS Chatbot Arena (72B version with Elo 1285), topping the Open LLM Leaderboard v2 with Qwen2-72B, powering 500+ Hugging Face spaces, 1 million commercial apps via ModelScope, and 300+ cited papers (Qwen1.5-32B), deployed on Alibaba Cloud for 1 million users, boasting 20,000 GitHub stars and 1,000 forks, supporting 200+ community fine-tunes, logging 5 million Qwen2 inferences, hitting 500K mobile deployments (Qwen-1.8B) and 1 million for Qwen2-0.5B, and being accessed by 100K developers for the Qwen1.5-110B preview, with multilingual variants gaining 10K stars, Qwen-VL trained on 1 million image-text pairs, and 50+ contributors—truly a standout in open-source and commercial AI.
Model Architecture
1Qwen-72B has 72 billion parameters
Verified2Qwen1.5-110B contains 110 billion parameters
Verified3Qwen2-72B features 72 billion parameters
Verified4Qwen-14B has 14 billion parameters
Directional5Qwen1.5-32B has 32 billion parameters
Single source6Qwen2-7B has 7 billion parameters
Verified7Qwen1.5-72B has 72 billion parameters
Verified8Qwen2-1.5B contains 1.5 billion parameters
Verified9Qwen-7B has 7 billion parameters
Directional10Qwen1.5-14B has 14 billion parameters
Single source11Qwen2-0.5B has 0.5 billion parameters
Verified12Qwen-1.8B has 1.8 billion parameters
Verified13Qwen1.5-7B has 7 billion parameters
Verified14Qwen2-72B-Instruct uses Transformer architecture with 80 layers
Directional15Qwen1.5-4B has 4 billion parameters
Single source16Qwen-72B-Instruct has 72 billion parameters
Verified17Qwen2-7B-Instruct features 7B params with 28 layers
Verified18Qwen1.5-1.8B has 1.8 billion parameters
Verified19Qwen-14B-Instruct has 14B parameters
Directional20Qwen2-1.5B-Instruct has 1.5B parameters
Single source21Qwen1.5-0.5B has 0.5 billion parameters
Verified22Qwen-7B-Instruct has 7B parameters
Verified23Qwen2-72B has group query attention with 8 query heads
Verified24Qwen1.5-110B-Instruct uses 110B parameters with SwiGLU
DirectionalModel Architecture Interpretation
The Qwen AI lineup spans a wide range of parameter sizes, from 0.5 billion up to 110 billion, across versions like Qwen, Qwen1.5, and Qwen2, with notable features including the Transformer architecture, SwiGLU activation in the 110B-instruction model, group query attention in Qwen2-72B, and varying layer counts (such as 80 layers in Qwen2-72B-Instruct and 28 in Qwen2-7B-Instruct) for select instruction-tuned variants.
Multilingual Support
1Qwen excels in 29 languages with C-Eval score of 85.2% for Qwen-72B
Verified2Qwen1.5-72B achieves 81.7% on MultiICL benchmark
Verified3Qwen2-72B scores 74.5% on MGSM multilingual math
Verified4Qwen-72B gets 84.3% on CMMLU Chinese benchmark
Directional5Qwen1.5-110B reaches 90.2% on C-Eval
Single source6Qwen2-7B-Instruct scores 68.9% on IFEval multilingual
Verified7Qwen supports Japanese with 82.1% on JMMLU for 72B
Verified8Qwen1.5-32B achieves 76.4% on MultiMT-Bench
Verified9Qwen2-1.5B gets 62.3% on Chinese HumanEval
Directional10Qwen-14B scores 79.5% on C-SimpleQA
Single source11Qwen1.5-7B reaches 73.8% on KoBBQ Korean benchmark
Verified12Qwen2-72B-Instruct 88.4% on Chinese NLI
Verified13Qwen-7B achieves 81.6% on CMMLU
Verified14Qwen1.5-14B scores 77.2% on Arabic MMLU
Directional15Qwen2-0.5B gets 55.7% on multilingual TriviaQA
Single source16Qwen-1.8B reaches 70.4% on French MMLU variant
Verified17Qwen1.5-72B-Instruct 83.9% on Spanish EQ-Bench
Verified18Qwen2-7B scores 71.2% on German HellaSwag
Verified19Qwen-72B-Instruct 86.7% on Russian RACE
Directional20Qwen1.5-4B achieves 69.8% on Italian GSM8K
Single source21Qwen2-1.5B-Instruct 64.5% on Hindi OpenbookQA
Verified22Qwen-14B-Instruct scores 78.9% on Thai summarization
Verified23Qwen1.5-1.8B gets 66.3% on Vietnamese ARC-Challenge
Verified24Qwen2-72B reaches 75.8% on Korean coding eval
Directional25Qwen1.5-0.5B scores 53.1% on multilingual commonsense
Single sourceMultilingual Support Interpretation
Qwen, a versatile language model that excels across 29 languages, proves its mettle with standout scores like 85.2% on C-Eval for Qwen-72B, 90.2% on C-Eval for Qwen1.5-110B, strong showings in benchmarks such as MultiICL (81.7% for Qwen1.5-72B) and Chinese NLI (88.4% for Qwen2-72B-Instruct), and solid performance in regional tasks like Korean coding (75.8% for Qwen2-72B) and French reasoning (70.4% for Qwen-1.8B), while even smaller models hold their own in various areas, from German HellaSwag to Russian RACE, and only a few benchmarks—like multilingual TriviaQA (55.3% for Qwen2-0.5B) or Chinese HumanEval (55.7% for Qwen2-1.5B)—show room for growth.
Training Details
1Qwen trained on over 2 trillion tokens
Verified2Qwen1.5 pre-trained on 7 trillion tokens including multilingual data
Verified3Qwen2-72B trained on 7+ trillion high-quality tokens
Verified4Qwen-72B used 10T tokens in pre-training
Directional5Qwen1.5-110B post-trained with over 1 million instructions
Single source6Qwen2 series employed YaRN for extended context up to 128K
Verified7Qwen-7B trained with 2T Chinese-English tokens
Verified8Qwen1.5-72B fine-tuned on 5B+ tokens of instruction data
Verified9Qwen2-7B pre-trained with enhanced data mixture
Directional10Qwen used supervised fine-tuning on 500K samples
Single source11Qwen1.5 supports 14 trillion token pre-training scale
Verified12Qwen2-0.5B trained on diverse code and math data
Verified13Qwen-14B utilized RLHF with 100K preferences
Verified14Qwen1.5-32B trained with long-context up to 32K tokens
Directional15Qwen2-72B-Instruct rejection sampled with DPO
Single source16Qwen-1.8B pre-trained on 1T+ tokens
Verified17Qwen1.5-7B used 3T multilingual tokens
Verified18Qwen2-1.5B fine-tuned on 2B instruction tokens
Verified19Qwen-72B-Instruct aligned with human feedback on 20K samples
Directional20Qwen1.5-14B trained for 128K context length
Single source21Qwen2 supports 29 languages in training data
Verified22Qwen1.5-4B pre-trained on synthetic data augmentation
VerifiedTraining Details Interpretation
Qwen and its family of models are AI powerhouses, chomping through trillions of tokens—from multilingual and synthetic data to code, math, and instruction sets—fine-tuning with millions of commands, aligning with human preferences (via techniques like RLHF and DPO), and packing context lengths up to 128K, with each version outdoing the last in scale, specificity, and versatility.
Sources & References
- Reference 1QWENLMqwenlm.github.ioVisit source
- Reference 2ARXIVarxiv.orgVisit source
- Reference 3HUGGINGFACEhuggingface.coVisit source
- Reference 4PLATFORMplatform.lmsys.orgVisit source
- Reference 5GITHUBgithub.comVisit source
- Reference 6ALIBABACLOUDalibabacloud.comVisit source
- Reference 7MODELSCOPEmodelscope.cnVisit source