|
|
9dbfac9dfa
|
PART A: verify kv_norm_w loaded correctly
|
2026-06-03 07:03:39 +00:00 |
|
|
|
a682c6adf4
|
PART A: add raw compressor output diagnostic
|
2026-06-03 06:56:56 +00:00 |
|
|
|
f2c1b3afd5
|
PART A: fix KV diagnostics — compute q_a before indexer, add Q_heads magnitude check
|
2026-06-03 06:33:51 +00:00 |
|
|
|
86e59c16c5
|
PART A: add KV gather diagnostics at blowup layer
|
2026-06-03 06:25:35 +00:00 |
|
|
|
262f844e2e
|
PART A: add detailed blowup diagnostics — capture mHC intermediate values when |X| > 1e6
|
2026-06-03 06:10:33 +00:00 |
|
|
|
6459fbca9a
|
fix: import forward_attention
|
2026-06-03 05:41:33 +00:00 |
|
|
|
91dfac34d8
|
PART A: simplified to production-only diagnostics — track per-layer |X| during prefill and decode, detect blowup early
|
2026-06-03 05:33:22 +00:00 |
|
|
|
d99503732d
|
fix: add BF16 gate weight fallback for dense routers (missing from test)
|
2026-06-03 05:22:47 +00:00 |
|
|
|
801bfc9a83
|
add router mode debug print
|
2026-06-03 05:15:52 +00:00 |
|
|
|
b385ecc05e
|
PART A: decode diagnostics test — production vs reference per-layer X comparison at decode step
|
2026-06-03 05:06:40 +00:00 |
|
