Gemma-2-9B
The 9-billion-parameter variant in Google's Gemma 2 model series, trained with a substantially over-trained strategy, achieving a token-to-parameter ratio of approximately 889:1.
Timeline (last 90 days)
On Gemma-2-9B and Qwen3-14B, the researchers successfully switched the facts read by the model by modifying early internal activation vectors while keeping the written instructions unchanged.
Compared to the average of the two graphs, the conditioned response graph reduces mean squared prediction error by 53.6% on Gemma-2-9B.
The large disparity in error reduction between Gemma-2-9B and Qwen3-14B indicates that different architectures differ significantly in their sensitivity to task context.
The token-to-parameter ratios of Llama-2-7B and Gemma-2-9B are approximately 290 and 889, respectively, far exceeding Chinchilla's 20:1.
All Facts (4)
The token-to-parameter ratios of Llama-2-7B and Gemma-2-9B are approximately 290 and 889, respectively, far exceeding Chinchilla's 20:1.
85%UnverifiedOn Gemma-2-9B and Qwen3-14B, the researchers successfully switched the facts read by the model by modifying early internal activation vectors while keeping the written instructions unchanged.
50%UnverifiedCompared to the average of the two graphs, the conditioned response graph reduces mean squared prediction error by 53.6% on Gemma-2-9B.
50%UnverifiedThe large disparity in error reduction between Gemma-2-9B and Qwen3-14B indicates that different architectures differ significantly in their sensitivity to task context.
50%