DeepSeek-V2-Chat-0628
DeepSeek · 2024-06-28 · 235.7B parameters
DeepSeek-V2-Chat-0628 is DeepSeek's June 28, 2024 update of DeepSeek-V2-Chat; its downloadable weights contain exactly 235,741,434,880 parameters, while DeepSeek reports only rounded figures of 236B total and 21B activated per token. DeepSeek reports improvements on HumanEval, MATH, BBH, IFEval, Arena-Hard, JSON output, system-message instruction following, and role-playing; the V2 family is documented for 128K context support, while this checkpoint's config declares 163,840 maximum positions.
Benchmark scores
| Benchmark | Score |
|---|---|
| Capability | 118.9 |
| Creative Writing v2 | 69.8 |
| EQ-Bench v2 | 83.2 |
| EQ-Bench v2 + MAGI-Hard Combined | 71.9 |
| MAGI-Hard | 60.6 |
| ZeroEval CRUX | 70.5 |
| ZeroEval GSM8K | 93.9 |
| ZeroEval MMLU-Redux | 80.8 |
| ZeroEval ZebraLogic | 22.7 |