Feature #558
Fires on “ an” (not labeled yet: read the examples)
- Fires on
- 1.30% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 94%
- story 5%
- python 0%
- chat 0%
- worksheet 0%
Makes the model say
. like such liked Thank a us<|assistant_end|>
Makes it not say
then wayfirstiescase there tailDon
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.1, write, execute, lsp<|system_end|><|system_start|>You are an AI agent running in
-
agent 3.07search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 3.07search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 3.07search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 3.07search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 3.07search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 3.07search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 2.99emory, github_search<|system_end|><|system_start|>You are an AI agent running in
-
agent 2.99search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 2.99search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 2.99search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 2.99search, write, execute<|system_end|><|system_start|>You are an AI agent running in