Mistral-Small

Mistral Small 22B Instruct - Details

Last update on 2025-05-18

Mistral Small 22B Instruct is a large language model developed by Mistral Ai, a company specializing in advanced AI research. With 22b parameters, it is designed for high-performance tasks, offering a large context window and rapid response times. The model operates under the Mistral AI Research License (Not Found), emphasizing its open-source nature and focus on practical, instruction-based applications.

Description of Mistral Small 22B Instruct

Mistral Small 22B Instruct is an instruct fine-tuned large language model with 22B parameters designed for efficient and scalable performance. It supports a 32k sequence length, enabling handling of extensive input contexts. The model includes function calling capabilities, making it suitable for interactive and dynamic applications. For production-ready inference pipelines, it is recommended to use the vLLM library to optimize speed and resource utilization.

Parameters & Context Length of Mistral Small 22B Instruct

22b 32k

Mistral Small 22B Instruct features 22b parameters, placing it in the large model category, which balances complexity and resource demands for tasks requiring depth and accuracy. Its 32k context length enables handling extended inputs, ideal for long-form content or detailed interactions, though it requires significant computational resources. This combination makes the model suitable for advanced applications where both scale and context are critical.
- Parameter Size: 22b
- Context Length: 32k

Possible Intended Uses of Mistral Small 22B Instruct

code generation inference pipelines model evaluation distributed computing question answering

Mistral Small 22B Instruct is a versatile large language model with 22b parameters and a 32k context length, making it suitable for a range of possible applications. Its design allows for possible use in generating coherent and contextually relevant text, which could support tasks like drafting documents or creating content. The model’s ability to handle extended contexts also opens possible opportunities for code generation, enabling developers to explore automated coding solutions. Additionally, its structured training may allow for possible effectiveness in question answering, where it could provide detailed responses to complex queries. However, these possible uses require thorough investigation to ensure alignment with specific requirements and constraints.
- text generation
- code generation
- question answering

Possible Applications of Mistral Small 22B Instruct

code assistent text generation technical documentation processing chatbot system content summarization tool

Mistral Small 22B Instruct is a large-scale language model with 22b parameters and a 32k context length, offering possible applications in areas like text generation, where it could assist with creative or technical writing tasks. Its possible ability to generate code might support developers in exploring automated solutions for software tasks. The model’s possible strength in question answering could enable it to provide detailed responses to complex queries, though this would require careful calibration. These possible uses highlight its flexibility but also underscore the need for thorough evaluation to ensure alignment with specific needs. Each application must be thoroughly evaluated and tested before use.
- text generation
- code generation
- question answering

Quantized Versions & Hardware Requirements of Mistral Small 22B Instruct

32 ram 24 vram 20 vram 36 vram

Mistral Small 22B Instruct in its medium q4 version balances precision and performance, requiring a GPU with at least 24GB VRAM for efficient deployment, along with 32GB+ system memory and adequate cooling. This configuration ensures compatibility with mid-to-high-end GPUs like the RTX 3090 Ti or A100, though specific needs may vary based on workload. Possible applications of this quantized version depend on hardware availability, and users should verify their setup aligns with these requirements.
- fp16, q2, q3, q4, q5, q6, q8

Conclusion

Mistral Small 22B Instruct is a large language model with 22b parameters and a 32k context length, designed for high-performance tasks. It is open-source, optimized for efficient inference with the vLLM library, and balances scalability with resource requirements.

References

Huggingface Model Page
Ollama Model Page

Benchmarks

Benchmark Name Score
Instruction Following Evaluation (IFEval) 66.70
Big Bench Hard (BBH) 30.79
Mathematical Reasoning Test (MATH Lvl 5) 14.35
General Purpose Question Answering (GPQA) 9.84
Multimodal Understanding and Reasoning (MUSR) 3.00
Massive Multitask Language Understanding (MMLU-PRO) 32.89
Instruction Following Evaluation (IFEval) 62.83
Big Bench Hard (BBH) 40.56
Mathematical Reasoning Test (MATH Lvl 5) 20.39
General Purpose Question Answering (GPQA) 11.07
Multimodal Understanding and Reasoning (MUSR) 10.22
Massive Multitask Language Understanding (MMLU-PRO) 34.43
Link: Huggingface - Open LLM Leaderboard
Benchmark Graph
Maintainer
Parameters & Context Length
  • Parameters: 22b
  • Context Length: 32K
Statistics
  • Huggingface Likes: 385
  • Huggingface Downloads: 9K
Intended Uses
  • Text Generation
  • Code Generation
  • Question Answering
Languages
  • English