InternLM2 Chat 7B
Shanghai AI Laboratory · 2024-01-17 · 7.7B parameters
InternLM2 Chat 7B is Shanghai AI Laboratory's RLHF-aligned 7B chat checkpoint in the InternLM2 family, released after supervised fine-tuning and three rounds of conditional online RLHF. It is a text model for instruction following, conversation, and tool calling; its config specifies 32,768 positions with dynamic RoPE, while official materials separately report extrapolated needle-in-a-haystack performance at 200K contexts.
Benchmark scores
| Benchmark | Score |
|---|---|
| EQ-Bench v2 | 62.6 |
| EQ-Bench v2 + MAGI-Hard Combined | 50.5 |
| MAGI-Hard | 38.4 |