AI Research Atlas

InternLM2 (7B, 20B)

Shanghai AI Laboratory (InternLM) · 17 January 2024

Second-generation open InternLM at 7B and 20B with 200K context via 32K training and a new conditional online RLHF method (COOL RLHF).

Tech report (arXiv 2403.17297, 2024-03-26) details data preparation, 4K to 32K context training, strong Needle-in-a-Haystack at 200K, and COOL RLHF to handle conflicting preferences and reward hacking.

Date
Wednesday, 17 January 2024
Lab
Shanghai AI Laboratory (InternLM)
Kind
open-weights
Access
open weights (restricted license)

Figures

MeasureValueMeasured by
Needle-in-a-Haystack context200K tokenscompany

Release 2024-01-17 per the InternLM README news; InternLM-20B came 2023-09-20; InternLM2.5 on 2024-07-03.

Sources

  1. github.com/InternLM/InternLM
  2. arxiv.org/abs/2403.17297

This record was checked against its sources on 6 October 2026. How we check