5.05 GB

Ctrl+K

1 contributor

Read up to 2048 state tokens (was 384): the adapters/pointer hold past their training length; on a long-context check (decision-bench quick-v1 in ~1.5k tokens of filler) the end of the text goes from chance to near the start-of-text accuracy

f556a69 verified 15 days ago

.gitattributes

1.52 kB
initial commit 29 days ago
README.md

4.39 kB
v2: trained on human-written typed decisions; Score routed to the vector reading 16 days ago
chat_template.jinja

9.06 kB
minicpm5-2b-decision: bf16 weights for transformers, pointer head and decision.json 29 days ago
config.json

749 Bytes
v2: trained on human-written typed decisions; Score routed to the vector reading 16 days ago
decision.json

1.27 kB
Read up to 2048 state tokens (was 384): the adapters/pointer hold past their training length; on a long-context check (decision-bench quick-v1 in ~1.5k tokens of filler) the end of the text goes from chance to near the start-of-text accuracy 15 days ago
generation_config.json

214 Bytes
v2: trained on human-written typed decisions; Score routed to the vector reading 16 days ago
model.safetensors

5.03 GB
xet

v2: trained on human-written typed decisions; Score routed to the vector reading 16 days ago
pointer_head.npz

4.2 MB
xet

v2: trained on human-written typed decisions; Score routed to the vector reading 16 days ago
tokenizer.json

9.89 MB
minicpm5-2b-decision: bf16 weights for transformers, pointer head and decision.json 29 days ago
tokenizer_config.json

436 Bytes
minicpm5-2b-decision: bf16 weights for transformers, pointer head and decision.json 29 days ago