26 days ago
-
Publish default Linear vLLM multimodal API and verified 86-label image runtime
-
Add opt-in shared-material reranking with measured multi-question cache reuse
-
Clarify intended use for complex and high-difficulty decisions
-
Reuse backend HTTP connections in verified BF16 linear runtime
-
Remove 180B benchmark conclusions; retain BF16 linear runtime
-
Publish original BF16 linear head runtime and reproducible public evaluation (part 2)
-
Publish original BF16 linear head runtime and reproducible public evaluation
-
initial commit
