Yi-1.5-9B-Chat is a conversational large language model developed by 01.AI as part of the Yi-1.5 model family. It is designed to deliver strong instruction-following, reasoning, and dialogue performance while maintaining efficient deployment requirements.
The model builds on the Yi-1.5 pretrained foundation and is further optimized for chat-style interaction. It supports structured responses, multi-turn conversation, and analytical tasks such as coding, mathematical reasoning, and knowledge-based explanation.
Yi-1.5 models are trained using high-quality large-scale datasets and refined through instruction tuning to improve response usefulness and conversational reliability.
Yi-1.5-9B-Chat is built to provide high-quality conversational performance with strong reasoning ability while remaining practical for deployment.
Key design priorities include:
Reliable instruction-following behavior
Strong performance in reasoning, math, and coding tasks
Stable multi-turn dialogue handling
Flexible long-context processing
Efficient inference compared to larger models
Quantization Details
Q4_K_M
Approx. ~71% size reduction (4.96 GB)
High compression for reduced memory usage
Suitable for CPU inference and limited VRAM systems
Faster generation for local deployments
Slight reduction in reasoning precision for complex tasks
Q5_K_M
Approx. ~67% size reduction (5.83 GB)
Higher numerical precision and response stability
Improved logical consistency and coherence
Better performance on reasoning-heavy prompts
Recommended when additional memory is available
Training Overview
Pretraining Foundation
Yi-1.5 models are continuously pretrained from the original Yi models using large-scale high-quality text corpora. Training data includes hundreds of billions of tokens designed to improve language understanding, reasoning, and knowledge representation.
Instruction Alignment
The chat variant is further refined using millions of diverse instruction and conversation examples to enhance:
Conversational clarity
Prompt understanding
Structured response generation
Task-oriented interaction
This process improves performance in coding, mathematics, reasoning, and instruction-following tasks compared to earlier Yi models.
Core Capabilities
Instruction-following
Executes complex prompts and structured tasks reliably.
