Neura Market
Qwen1.5-0.5B|1.8B|4B|7B|14B|32B|72B|110B|MoE-A2.7B logo

Qwen1.5-0.5B|1.8B|4B|7B|14B|32B|72B|110B|MoE-A2.7B

Free

Next-generation open-source LLM series with enhanced multilingual and alignment capabilities

FreeFree tier
Inputs: textOutputs: text
Type
Open Source
Company
Qwen Team
Visit Website GitHub (38 stars)

About Qwen1.5-0.5B|1.8B|4B|7B|14B|32B|72B|110B|MoE-A2.7B

Qwen1.5 is the next iteration of the Qwen series of open-source large language models, released by the Qwen Team in February 2024. It offers base and chat models across eight sizes: 0.5B, 1.8B, 4B, 7B, 14B, 32B, 72B, and 110B, along with a Mixture-of-Experts (MoE) variant. All models support a context length of up to 32,768 tokens. The release brings improved alignment with human preferences and enhanced multilingual capabilities. Quantized versions (Int4, Int8 GPTQ, AWQ, GGUF) are available. Qwen1.5 integrates with Hugging Face Transformers (transformers>=4.37.0 without trust_remote_code) and is supported by deployment frameworks like vLLM and SGLang, quantization tools (AutoAWQ, AutoGPTQ), finetuning frameworks (Axolotl, LLaMA-Factory), and local inference with llama.cpp. It is accessible on platforms such as Ollama, LMStudio, and via API services on DashScope and Together AI. The models demonstrate strong performance on benchmarks including MMLU, C-Eval, GSM8K, MATH, HumanEval, MBPP, BBH, and CMMLU, with Qwen1.5-72B outperforming Llama2-70B across all benchmarks.

Key Features

Eight model sizes: 0.5B, 1.8B, 4B, 7B, 14B, 32B, 72B, 110B plus MoE variant
Context length of up to 32,768 tokens
Improved alignment with human preferences and multilingual support
Quantized models: Int4, Int8 GPTQ, AWQ, GGUF
Integration with Hugging Face Transformers without trust_remote_code
Compatible with vLLM, SGLang, AutoAWQ, AutoGPTQ, Axolotl, LLaMA-Factory, llama.cpp
Available on Ollama, LMStudio, and API services (DashScope, Together AI)
Strong benchmark performance surpassing Llama2-70B on multiple metrics

Pros & Cons

Pros
  • Open-source with multiple model sizes for different hardware constraints
  • Excellent performance relative to model size, especially the 72B variant
  • Broad framework compatibility simplifies deployment and fine-tuning
  • Supports very long context windows (32K tokens)
  • Quantized versions available for efficient inference
Cons
  • Model scalability and deployment cost for larger sizes may limit access
  • Quality of base models only slightly improved; focus on chat models
  • MoE model is not detailed in performance benchmarks on this page
  • Limited to text-based tasks; no multimodal capabilities in this release

Best For

Language understanding and reasoningCoding and code generationMathematical problem solvingMultilingual tasksRetrieval-augmented generation (RAG)Agent and tool useFine-tuning for custom applicationsLocal LLM inference with quantization

FAQ

What model sizes are available in Qwen1.5?
Qwen1.5 offers base and chat models of 0.5B, 1.8B, 4B, 7B, 14B, 32B, 72B, and 110B parameters, plus a Mixture-of-Experts model.
What is the context length of Qwen1.5 models?
All Qwen1.5 models support a context length of up to 32,768 tokens.
Which frameworks and platforms support Qwen1.5?
Qwen1.5 integrates with Hugging Face Transformers (>=4.37.0), vLLM, SGLang, AutoAWQ, AutoGPTQ, Axolotl, LLaMA-Factory, llama.cpp, Ollama, LMStudio, and API services via DashScope and Together AI.
Are quantized versions of Qwen1.5 available?
Yes, quantized models including Int4 and Int8 GPTQ, AWQ, and GGUF are provided.