Repository navigation

[CLOSE — superseded by KakeyaLattice] PR-R1c (research, archived): cap-train cross-attn toy + diagnostic mode + vast H200 evidence - #65

Closed

FluffyAIcode wants to merge 3 commits into

AgentMemory/v04-pr-r1b-cross-attn-toy-fix-8e7ffrom

AgentMemory/v04-pr-r1c-cross-attn-toy-cap-train-8e7f

Closed

[CLOSE — superseded by KakeyaLattice] PR-R1c (research, archived): cap-train cross-attn toy + diagnostic mode + vast H200 evidence#65
FluffyAIcode wants to merge 3 commits into
AgentMemory/v04-pr-r1b-cross-attn-toy-fix-8e7ffrom
AgentMemory/v04-pr-r1c-cross-attn-toy-cap-train-8e7f

Commits on Jun 7, 2026
- PR-R1c: cap-train cross-attn toy (2000 steps, capacity bump, o_proj init, needle debug modes)
  
  cursoragent
  and
  FluffyAIcode
  committed
  0e093ee
- vast H200 evidence for PR-R1c (ADR 0011 G-X1 attempt #3)
  
  cursoragent
  and
  FluffyAIcode
  committed
  75ba2ae
- vast H200 evidence: needle-small 6000-step run (3x budget) for PR-R1c
  
  cursoragent
  and
  FluffyAIcode
  committed
  22846e9