XuanYuan 2 Models by Du Xiaoman Data Intelligence
Last refreshed 2026-05-19. Next refresh: weekly.
Details
About
XuanYuan2-70B is the second-generation 70B series in Du Xiaoman's XuanYuan line of finance-focused large language models. It builds on the first-generation XuanYuan-70B base, a Chinese-enhanced model built on Llama 2 70B, with more continued pretraining on high-quality data using a bucketed dynamic pretraining method, followed by instruction tuning and reinforcement learning from human feedback. Du Xiaoman says this improves general ability, safety, and financial skills over the first generation, with support for 16K context. The series, open-sourced in March 2024, has a base model, a chat model, and 8-bit and 4-bit quantized chat versions.
Decision facts
- Best fit
- General model comparison
- Capability starting point
- XuanYuan2 70B with 16K context
- Lowest tracked input
- Not tracked
- Closest related family
- XuanYuan
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
Use when the workload needs 16K context and 70B parameters.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| XuanYuan2 70B | Use when the workload needs 16K context and 70B parameters. | 2024-03 | 16K context70B parameters | Current |
Release Timeline
1 release groupSpecifications(1 models)
| Model | Released | Context | Parameters |
|---|---|---|---|
| XuanYuan2 70B | 2024-03 | 16K | 70B |
