diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/model.safetensors b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..783be8cf2f5e4aca8914a8d89c71fb09d5bd43f0
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48bc8e39e698957d5693df7a8c0a0d78966aff360b6b75640fcbce4db9686272
+size 75550008
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d7fd6b146bfcc0475a1e823b99db5f91f165313e
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e28c53f71f74a065ae57077af5269a989e04645cb45943a6e879161f05eb9ab3
+size 99648
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7898557ea074909faacc3ab65a705dd95e8318d2
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b086675ef55132ea703af5da096f611866d08e09985f3c0855c4f4e9da74329
+size 50471
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2c7690446fbacacc61887f808416111d743a607a
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf470d52f723cce055e16f4e1c4ac9150397d9c3f6fe7977f20574366d040dd6
+size 50382
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e45207054d68b26f83e67167405f3c4d92c2102b
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0b02b61aa710e83e4b8bf520f5df8903f432a4abfe8a6dd6e2ef798d7f4a5c1
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..00b8c29456f450fbd74d1adc2b926920f1274ff4
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d6723f6f7a778dd765ecaa5c4400ae5cf441a33f1914acec85de28766fa112e
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/model.safetensors b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..febda8e1d5b448b27cf9a014bfd569eba996eeaf
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e128024f02b378c4bee46e3ab87104c50c680df37b29621fea1810e93e8acd19
+size 75550008
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f5d43118796345f3148c25db6f93f0cb80a996e7
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:71ee525618dbf3a183276c6a993ec2990ec7f87e8584e53e4636b26c30ff3ae6
+size 99648
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f4d47069da40909026f8f9ca9af3bfd8a536017d
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f4345519baa6bc9a3ec348de48f27fde108271da0de6d559562dceb14664ae2
+size 50471
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..173f92e90f75a3db82bcd4f8663da2d3bdf808f6
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a21f3c77068b7c447109ff3dfa531c942f16c55206e94567740f820982cb754
+size 50382
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..82b3413d5064b0697f4a4b703392c47103701c48
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f6db6f917eeb178365d8d83220afa3757c22b339c58b408300e5145e6514264
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f61ea73632782482c9d444cb9b34dcf90645d289
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ace47ac0862278dc5f1c940fa3a194e19b4dafc4581fb48f645eebfd5ba2550
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/model.safetensors b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1500ad75eaf40aec39175c9dfcb247e64aa8e9a4
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d7c0642336fbb76335ab83b121e03fcce5312cccf1d5c662ddaf603764b469c
+size 75550008
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8b5b1b524c8c431856f2a6bf41f012b8f244ad33
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efaafe8f4853ef7d3cfd866c34578f146a42acbe68e621759c9cf9f843b190af
+size 99648
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..6509e281a639134541d1e5f9c95d09309982937b
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b56cb6d77bc7cb4396df623688de13c1e6aa48b91af03db277c84d9c9b3c03d1
+size 50471
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a993f0809f332c7fbe1b7561f246b2f53e8b6839
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4de9ffbb2c67e7938a4e3393687d79e602735677ef1671afa461bd43110676f6
+size 50382
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d7b1b5171f807babf0c4191c94c924a5450f65c4
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98c0fa12bb683500ebbe5eaa0416c488041e90f640b88fed7a6ff5d7fc3cb131
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..998799f92c1b6ffbb9ed2b9a0464bcf7e523fb62
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e24363b3ed6cd1ab8a6ceb8a11785b14c8c8a1bc79495c157cf09b77c666d18d
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/model.safetensors b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a60cec461773ff95509c12aac26550931c7999d5
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7cb962e4e685b5529cd2126e05310b93be500337c522abb979ea3b9e7788be26
+size 75550008
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4638ee8df6df90b8740eb3eb8ec4b25b36bebde1
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36a9f019c96260e9ce2251674689c674435c4479b75b4861f832a95aa0ae9075
+size 99648
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f2e3b61d1b03d7dc6cb966898023046344848aa0
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:773a7adf718655cdeec995850a46b467606b4d420ab839f3a7bf8b06ca9c07ba
+size 50471
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..430d9e6cb6164a06bd9ac8c3ed2a66ebaed718c1
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bee0eef9f116344b82c6951f2c1cb377e80a9f21de5c63d5c9be31c13cdaf1b0
+size 50382
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..eda06e87041b25332c64ac2b16a9686649a39df1
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25822cdc54edb775e745429312823823d72ac8fede3c1db3faff247964672aad
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..23dc0db64439814554233fb345f9ab4bed8f885a
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c9e046e5c22e4a8a945b0da026ec58a1aec6f19cae8ecc2bdf4a45cc22d03a48
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/model.safetensors b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..09d654e5510607e083813b645828b4aefb945746
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:356b0edcb0420ab3e3b286bda60da924d92b540ce7765eb2d1a5790e3c8ee791
+size 75550008
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f96829acf44f552929b10caeb0c5bd6a2e4c036f
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff25de822d7d830913d57ffe3bd2ac6c2726b50a8fa174943c7b1e008787b3f5
+size 99648
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d072b3683a738ff7b71c4af930fb5f2ca1f990d1
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2bcd191e85a80b9e835ef507efbebc1ded8f7d95311c99bca5e88f2b1e322eb
+size 50471
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8cd8ea7b7f84948e99216ead80770a731bad0e64
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66c5f64d78082aeb37adddc9fb4b6f2bae69afe533439f364ca7bdb54a89b167
+size 50382
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..6009011eeb986a20ee4116664bb800b33d82837d
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee080bfe9e41319eee4b23d0f2c12893e14292fcfc9424ed089604397a530f95
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..86e39512d47ca31052a45035b089e1af0e81a104
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b47b7c779643b65334dbcb7c7f6c2c2032a2e5ee428498211d5e11ed4602f725
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/model.safetensors b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c600835205520b5062864c940039729e7743232a
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54d62e5777de311220a39c1c5f04cd2f11b7b6703ba84e81222db22b315c4fe2
+size 75550008
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9e1bad2745a58da041cfc628159248d9fd6da136
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ef943d548e806248d1ca5557a5c8e1dec8cb84bb6c7457dae149e6b2d025e05
+size 99648
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..40989e3228041d7a110a4bd5afe159e00442fe60
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:04aa93f42d1bf760d8840b2ea13826d096fd4e60a9c475b6dc92ccedc7a9f9d1
+size 50471
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a9363862b6c54bf282ba437628cc655fa504f51e
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ffc68d2202391ec536bd0af863b08d4c06649419e4944b7cfb5db13c34cbc00e
+size 50382
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9c5148990fd3052ab2c4d48cc72088cceb07fe84
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a07245beb75a6fae85ca7747ea9cbc2f5a3136eaeef6581324aec1b1e53bc830
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9af45b8ec1657a6f32fcb9782dfbfce56192ffbf
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:613afaaafd0cf83818dc25281420b7f70740bfdf6c5b80f4a0a784df8cde29a1
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/model.safetensors b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9bcac67c985dce9c30018692d9906003d445ae36
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3527206ac9b7c45f279cdc0114b376c659e2dd8c33a83fc73d12a1796fe1d0e
+size 75550008
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b2ed60549745ff9fd5101440c7cdd36df32db84e
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba46c110f7a9a77aaa536f6e50f5015334be0b73c11bfd95ea04740ccabe2c39
+size 99648
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..29370274f15d6321ab5d79d21783d30f611d6566
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cbfe2631a1e2112ee078c30004dbfebc27e412446ed987d981c4dbff5db1eb24
+size 50471
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..10116f831d3961fd22b5d0089772ee6df8419a89
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bbc39436f26c7d075d146a74f37bb55194d2bb7be11107c83cd8ce6898287c4b
+size 50382
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..741b4e3f0a16060f2301b7b3361f2a0eaed976c6
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3fc2c84e3a6ca2667bd4164745ad8836acebc98e110b2df977036e563c7c0968
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d3080e928ca43780a3992bc70c80b311bac8a514
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94cbc051a223b82de6a47dfc15aa265a081fcfb7f136f904410a0ba301d77ba2
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/model.safetensors b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..72db6994a489d0271a659c8c0430c5ea3fbb4abe
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e4d7b4f33bae19268e49b7902f9cd93fc2b4025c38de1216af4d0932f1bc5bf0
+size 75550008
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d65e1bff6e6d0bb2001943fedb4d9faed0a97369
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e710f0cd047e3fc35b7077506f6357b7187392eae7536d012a88f4a6ddd1fa2c
+size 99648
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9c9c95bdbe28a020629a8faf2a897ba388767343
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e68c38573f7c7591be45483bd4278c2e1e274704e9287d57606fd5ef26c7027
+size 50471
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c1c39d2390ce7b990b561193431092859a510db9
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:801f4cea6d6ab0a339b091d69431e2832c0cc9053c1ca17270b87d1a1f7dddfc
+size 50382
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..35073f9302d73eb8486b6969faadcf9fb8858887
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7e917b3e065f656e2235311d055f018db76572e6a3909a7fbe3526225be524c2
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..ef1daf3bb028a737e6a97eb31789dd4ed60a4f49
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0d63eb0c5fe53cf7c7f307657ca9d98e45db4f5c9fe031b6d102509a20b81c8
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/model.safetensors b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8df758f244785ab6aa7add66acedc15e17f77cae
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e581ba55c5c1d2a7af7a1026809e3129d41256c76bf11cdf512c4b5cab1c3b6
+size 75550008
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8571432da5adc0ad390c9f45a3eb30a51fd18a1e
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6668a00b51d993e1f29ffe64c47504efe9fdd035e6076c3296c60ad259d646b
+size 99648
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e1f6005b11edabcec040717ec5b27edb1bfd9460
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3fb745f08615ed20fb74230d13c6fd33283e413382bc0cc1110553c5681b7403
+size 50471
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..81f234633b020e2bbb7968c2e76ed4da77ae5312
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e061a2349a0edac16c655f24e26d93e06f34ce4540d2aace8e5c41d15a79601
+size 50382
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..6aa1529ed57937d63b404bb606fed6d046ebf3ec
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b07e783bfa13416f3e33506d91d15e6150f952984f051208b47089165cc5573a
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..cd3c462a92ca7d4cd54f230ad7a3c5f0b905aacb
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e43fa79f727788d4c72f1a152d343d8533991090f4e8113bde56ab943e1484a7
+size 50387
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/model.safetensors b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f62c0eb96a6761fc0199f2400b9ae92fc4d95000
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9821e020bdca4bcd94bda2e3bb37443d315d4dcc4456e4ef0c71a7ab1549cee8
+size 18890032
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d13be9b9107f8b38614b862c6771b01e14d4b7e1
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:204ebc940bdca83c3425988910f39aad4008c9defc6f35bf09114290b67a9f33
+size 25920
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c1ca27c5cf4c1b9cb13af4e88ad8da996f877070
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ee6e5a5625dae60119c84849127a989a2c10123a6ee17bdd637663be1303064
+size 13607
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..ee223fd48cee7775ac0e686b427d471ffea967a4
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f5db9fa97d739532c2eac1b1b719e8cb0b38ce1662ce4ac5d65b326e2a23c93
+size 13518
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..44be7bd0337fd4f22b8990d595f68c59c3eab5a6
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1207b99701a4d9e3d279287502dbd992ec115342c8531d7d25cc55bf1d91e5d1
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..602d8ac815fe58cc056f295e7a5e9580c9cd120a
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efff07d48707b4d5dd08a6e68fed42a3faf37b04a3af4fc70d4a8075b8517e2a
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/model.safetensors b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..67dd51ea09c297371f0dabb86c40b4d8fb2032da
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e43fdf0fc847f0a4665acf76ec6660e0abd9172200542da74f82b2abfd3dd07
+size 18890032
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..cdb127ecc949897bbe4b3be41ad886d1dc8ccee6
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:60762d3879e72adea3a5cc65cae559b90861d0cd573bc54bb0c6f8cc92b84c97
+size 25920
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b1d0bd81fa283fc398df79cdb6691c9c85da1eea
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e059760597c93f9b7ae78b3827a82a4ee3236269e7da116e1a204759bbb8976d
+size 13607
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0778d117ade306ec281252fb837fcd7f2a85c795
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:65f477416579521b1edcdac7e1cce71a2607d2a490a08ba99e9060e4273da825
+size 13518
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f9a30c6f2aa31fc3c9b3e72e287568a53502a781
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fcfc8adb0bf9b4ce62714a41f4ed4fbac4d091db616230a068bfc1195f8ed293
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..917188205a59c18782d4b424da1ef7fca6707577
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94e793bbda70a256e807020f1ae1b74bd9f38bbde4996207eef5899854a576e4
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/model.safetensors b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..13779acd7d4ee93e17bb7e75e1dd33a397f2699d
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:90b93560a66f5ebb817c15bca1f7795b7b76d0c3dfa63719ff7e005f5bb697ff
+size 18890032
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..254cfa6987ab515708e9d2bc256af5c31a22a110
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b3dd499bc89d77ca98060efe52e9ba5a932929e61920d7147ffbc961ef7e477
+size 25920
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4f8a8ef14e31e2fb1e5d0cceed0280d85813243c
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5627662b0c55c38c5483257c22fedad0039b83e3a3ccf888af2ba4c33aaef07
+size 13607
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..035f3a8a1d6ee9c6f99460fdfa60c8fd088c5092
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eccf0d8ec89a5a46e3b3c4eb5a905404ecc77258a134d97a5e57002adca413ad
+size 13518
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a5684cb258468351d573991c11bfb752afa3c16b
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:44689046c906160dabca0cbc4620f333418d9241269c5fdffbdee49a02a87e38
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..bf5bf623116064c7da639d9c901cb6ceb7fe2469
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ae589de95ae5b5b21b0527b0f7279f79f4659f0a6e9d9b8b086bc60dab71716
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/model.safetensors b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e7960b36f6015530cd0ef7bbd8409a9d84f123c5
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:febdfb461d2d0b27ba309c2c34529708cb10cbecc71f06a816924e2c4cb55650
+size 18890032
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..6f4b2b101987cdc9ada9e966e265f874a34856ba
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f720640708626699bcba34b370c9bc7bcc98b2ba858a350b362a4a3091cce830
+size 25920
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3cbcfc8e6473398e22b694ab8151ddac33650e72
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aee04dd8a3e1bc50898c8a36a99a0c32f4ad5824c850fb439a2ec0b5bd62a74c
+size 13607
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..007d1ad755402abc2b8015f18f5af973c0b73dc2
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:542edd874c3cad544c09e4bd57686b7a91c5c26cd971e602da12f01599cbd7be
+size 13518
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1067388c1bedac17b61b82b4afa51a56dcc3aaff
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72c559892a6bfad6a77a7393e52320d126ceac3b44d91ead8d398b8e800d102d
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fc362309776926c6531010d37cd67a4fab3cefee
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1868615ee8a21c1dffea8bb78c68d127e5389a942d7596f3cfa12d2c59e3a594
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/model.safetensors b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..faaa87b5986c5c9d69f8dd93026e86ef0bda31d3
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03d7901ecaea9f4916046d2064e0775299c6bdc4f0705874c67ab12637d2e521
+size 18890032
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8bb35e07d78a35c165fa8c29d7b4c144fd279918
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1debf85c7c0131fbe6aeaaadb236b07a468ba2da0eb339a74196799019fab308
+size 25920
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..26c08fa74c3cf4d30179fe1a13d492562cc14e8a
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3104af49585120c7a5975a220e558e139cbcd9d084105a1727663d9a6a5ff5a8
+size 13607
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..086441de2b33c88eb53f793f54fd27e36dcf59c1
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4fb9cb882a80b099a5fee5a5c630ab8f6d402653b89a9c4f241e329504af7138
+size 13518
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f954e4a28918a9fe7200931aa787985162aa6989
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ecc5f2899c2e794583ecf5bdf3ae4638d44b5625c1ef0395bd494b94bdc085b
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..697f732fa412d5b17e099e3ca21b67285988a186
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:205139b6cd944047682e6a7d9248362b778340e82f30ee5b8106cecefe0ade42
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/model.safetensors b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..184b311e8e49022ca4909c17cad007f18f960565
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d589418112c9cac7f4eda26aea182a5c98a269e8747005a74755d40a6d12e543
+size 18890032
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a14627410b156e4ec747038ba88e8772404a774f
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48a1ef9d0bd7dddf3ec342e5402dee4abb7c79911a2476f149a094df75195046
+size 25920
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..97814668d63351ddcd6ae4e4d33373baeebe13ad
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c148d321449b80bbdfc191c4eadc6a43422d6bc0bbfc1b4525765f0bb05c9de
+size 13607
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c7cc93545fb76cf60f17216564b267cbf991af28
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e011be71b081ed90a4b83e44245f19baa663488fe9a435804a91654ae88bd84f
+size 13518
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e524765efebc4d97d0c20b96226d6946a2c680d1
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:219e351d956d622d52fdc211d2bb4ce5c5570245bae0dd3d073819c46a07ff7a
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5dd12f448c898b7a644c5ef3aa2b6cc49c494618
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df99b9e49561d4bf1aaf474b82bd6b514c00b73fb66acfa1eb4890583a8e86a7
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/model.safetensors b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a31cf8fbb9cf319531c2bc2c0d4a041d33ab900e
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f25a93aebad75848e24c17ec8839054484d934d9e0fe31004608f2167fbdaf62
+size 18890032
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..455536e11008816adcad0148dba524c6b83c33ad
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bad086876419c6df25565e4269b3d3369425cff43411556f91ee629747d7eddb
+size 25920
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..53ee986ed2bc53f63e296639068e8884540f87c6
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d1a21cbf1f3ba1d284d345d29ecde64b4387b50dd01aec0c9b8023b829e53f9
+size 13607
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e3f1a1166a5f83c6f2d5c91f5fb6b094d2ecd7d3
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8fa8371bbb202d84646f792d0eb04c85dc7e556778bd52ab054d9e009ea0f33c
+size 13518
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..475424b48084980f7ea20a2a9ca153ec0b5f0fd2
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12cdecbf2587b1ecb4607ae3d5c768ca25baf2f34d46dd14def5feac93d6d291
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e8daff11ca2f8b74422b910e116af2d28d7cd0d0
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d756b242bb8272ea6862ab8525f077a8952955ead6b2500c4870e38eedd37a90
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/model.safetensors b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8fb49cbf162938ec4c2fc2c086913e5cc0839c9e
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf3ac757159e4128d74dc8efb632ae2118bdb3cbb34e7d7051d6f99da4541d8b
+size 18890032
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5fa93523e3aa809f6918f54e465fcb26ace50079
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe1f328ecf55d2e3f206610cbcb7df9498f28926eb176683beffa6451f0815f5
+size 25920
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5d456a87151eca9497439957fb3e0c12d5a709e6
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:59eb572b5f60b88f3155b442c819b1790012164c899d508c8cfb990ea5bc4d7c
+size 13607
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c01d5e4ab570740f7ba6e9fcc62dc87826fd6fa4
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3839e1727cb08cadd48b4e05531b041c9ec444424f78689bd53fccdcd81dd4d9
+size 13518
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f8086cc0e7050a572c83be83bb5b8e2712471f6b
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4211067cf5de473eb0571a4d8d8e043442bf5cac4f38e56e773172cb00dd39f
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f8ea44f7f708abef9671c145753cab5b9ad03692
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08837d18cab609dcac2a531f7da2cf48e0d6cd7ff4b98684e572c57853f5af80
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/model.safetensors b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5f5571140f0278c3917b33c366a0ac48273d0501
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83bca20231bace7fd25f3dcda0549919e3465a37746e1b2dd4582f5819178faa
+size 18890032
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fc87545b05e90e73c603637e42759e19c13304ee
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e570fd147e94f142bc2f0d5a1ae4b60d6ff05b1c711236208e3a95b4b738a9ce
+size 25920
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f60bd585ff74c93e866bf41700f1505497e4afb6
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:655e7e967e48ec4b1ac76054938111b34bd65b236c4902cb9e56ce96d6006922
+size 13607
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..360387c977fc42e801603503b89b906169b5e8f1
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db4a211ff586bfac97db4e7df2e6630a4e6b00a02280fc04846c8bac80549cc5
+size 13518
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e2df439d7a7f7d52b09a954b77ec7ed9c7764dfc
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3b1a7ba3647cfccfc3e4da9a285deef641ea358a54a6343b8bbbd628b45f15f
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a427eab3489c1f72c3aaea82d1c4d7e03ba026a9
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efbbd04ec521afe6e4a27be5371003550ba0c0981c26990fcb3d3d954720eafd
+size 13523
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/model.safetensors b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7c4c4167717cba4527a42b789313f265c42cc844
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f30b86331736e5f6e20e0d2e986bc65de8f34d57b4e812b1c3955d20fd2d21d4
+size 37776696
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8fbf075b77f62e19dc4a20ea62e2e62ac9b759f6
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f1ff71072738a91cdbf7f7c4675d03db658d499fb32289435df2908c247f45a
+size 50496
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..70b64fe46cfc2ab9d8f756397ab01d88bc9a1219
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:04ae5489a74f3399a222f1408f0b1d5e6244b238ea851014692d62840fbba9df
+size 25895
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1d8336b489c2d12c5441464e18f2d08ac55bdfef
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f54a64f77ae70b9552d6ab02dad2304454871d4f38b4b452a9749382e87c163
+size 25806
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8d4ec9055572a1eedf29ed9ac2ce2bce7fb9ed9c
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:947fd06cb4970b4043757a2affc7abf511ee6abe18a5913c4a7dfb3f06c82f89
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b0e0efe85c83c04f838dab1dc5c549492a3cca88
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df50dd4834b70c9bee39095bd38d6172dcda11487a5f018e9222c7dabef1f867
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/model.safetensors b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..859a2cf31117b0ec5dcdbd348d465a6b7d8f741c
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fceff22007aefbe62c3ee3c480c9858e905534c8a576e4e9b9072993c3b74077
+size 37776696
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..406d1ac19b201beccbfd6f2e906f6f52efc00f05
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd1170f5ff16caea2557c65b472e19468bb0109a175df40d13a16b85438e306d
+size 50496
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f8901025a07dece37025e7c44c7e141107daec89
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf15a84cadd7ad1808306174696d88d41d2ad9b05b1518c1edcface70a843fab
+size 25895
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..89c2d8377d8a20c6f8d1f2f01b8c505334ae2759
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b259fc22424d2f0e66f5f047775cafa957ce3e9ce8f9af26f05061dd1f61a1ac
+size 25806
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fb578dce08f5e1d99b9ba7d7c1a15aede2b6d358
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35574c715d0aadf55b6c5be4301a8b0592c33b3b5b9068456f0332da72105c8e
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..173c0a0045a2b85d0301d4dd29fcb2dba90d10b9
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5bfbd8b64af8ec047652f8b66a78ca4d10524593144459cf709115d909f57ec
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/model.safetensors b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b4e7934637119fcb0ca245920ef0d5413a4cd7f0
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c26ab37b90215080fdc78880df76c042d5e8e82eea476bfabd86c232f84d67a
+size 37776696
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0f9228ee2791afc10f9c5254bce57f011b7cc1f1
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1134ca826030402d3288612e0c9fdc911aa551d89a0d2f84de10be3f8c78d489
+size 50496
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..cdeaacc571c0ccc7d11002a2cacbad58439d6597
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4764b379e5067994b7d4003de691eee2b24107056befa7b17ff0b330d91cdcad
+size 25895
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..69f95548284d30442cb85e77af04877bec58e910
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:15afdc97ed828bd85d77cf887686493ecb8185e24cb3ee29886a2aacdf9d940f
+size 25806
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fe4ba493ff58f81b350a193f6309ebfe68cc22ed
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aecb341f9d4f9c6cebafb67aaf5195db0744b06626ad197ae38153e916d9e79e
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..80dbfb0ffe7a00612eb124e3bbf32e342cc5a4b1
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08b1263e6cb3583249f87e5ec8a2593e979ff2c5c86c1cd251e4988d05ca0afc
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/model.safetensors b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f89aa63baeb3814231c0ba34c615aeffd695d4ee
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f8080e8b2b90f96cb4f7ffe9748c3695e2943237fc2476f088d5888dc6e2b151
+size 37776696
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..642d7c7b659637315225d87e9115ba1addf9e907
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c7548b6af7bda0a2b8617aab837cf87d0ded17b7797218bbb9b2ed2f285667b
+size 50496
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9f236769130de11af64d283db19a916c46c38521
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91d0de2155971894ad027b62cceb76e50e8a74d082392ebcc1a203369de36c7c
+size 25895
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..92790e3e0e1970c793832d87621cffd8908ed8ad
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7fadd41a35628d8eab9c77241fa6c98f427bc5e72b8f198714fb3ec21c0c5ee
+size 25806
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..dc3f30081ab037daf79ad87a3bfe3fbb804ceff6
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd0e28b5c0a3b38bfbf6daf863421b884300cf61d0dbd3d5e8d86c4c1e842033
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5dd3ca0323586258b34ea49f69ffaca9ba2145ea
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:388df1cee46f5a3e72e87866b93b259745ed5656a7a38b951486d2fb9c7e0c2f
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/model.safetensors b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..afd86fece552fe98286f4932222354f92a6d4b33
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20acad14003f1bd04281a12cfc41b5cf2bf14000ae8f020ce67a03dd96e99f08
+size 37776696
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e44e65cc3e190ac466cc965f079a69ca79eceb0c
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2b980f556e99376f54c7bf8237da69c5a61bdd29b9bcaa90b57719de581b535
+size 50496
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..83eb46ef26d545de1fc0461ce43c3b99289e6c54
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4933fcc7f7bd72eb6321fe4b381c9152d0b7c513afe2b85d8a2b296785d9ad5c
+size 25895
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..bc4027493e94ed6cf82ec100db490ebc5e6a8147
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f705c0d6251b8449a080917bf9a2982291969f73d294f8d812d2c453f355d0b
+size 25806
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b6da77ae3fc3080dfae0ae95e3dbefdfbf263bbb
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f6187fccca24a944fb30147e2848d3454266ef683ea05333ddfd5fbc652ddef
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9d0c05d121bab1bac4d7e35b2eb54827cbc43603
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:754fdf62701c8d0fb7e0881401034671d3d7410c6b98e3925d3c07b9d9bd0610
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/model.safetensors b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..65cf240145bd83d7b1eeb9c90c3567f5bad92f5f
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3cf78fef0bf243833936d2d8fc8110a4770ee5cd7eb27b1e781442b0370df8f
+size 37776696
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5cc4d4d3d2cdbbdeef51f44748b0d396a6a9d4f5
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:119c955ba35b32fb76b656b76d87af16d2dc70d9d8b1b6686d962f5aa6d50133
+size 50496
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c8b50051686a88637190b3df554a47b3f6c56fb3
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:504d98ebea35c088bd31426e6b6bb75491dc278b02d6c818bb15bbbd3b978d9b
+size 25895
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..88713291f0f73d820a49ad61e2760e09d03d05cb
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e076e0353d47616a40dc2d317f0f1938a0fe0cbccb45cd8323d0cc1587d5da9d
+size 25806
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1e22596c668e234a9df0c4efc0b45a98c135a049
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:056e98cd0128ea363e923c8eda6afeb01276191441e2143e5bc0148041f07903
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c627f4cc0b0af5c091f39d25f2b48b7595d5f9a5
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1656f593f8918ab149f353a083bd3492083a9f76a4a5b72b5fec26c68866d61a
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/model.safetensors b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..afbb97df3d0cb190086d66ff1e74511158eb2373
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93671928dbc40b9debdf5f258e090f430981f0132443cbc1aa0c6b4cf52d87eb
+size 37776696
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..000f8152cb368df17bb81f51acc8ff94f4e70b7c
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:212080ce37e88576033d6aa8ede6e35d200d2094ad3eb2ac222013e6900ff9fd
+size 50496
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..033e0e460133f83e51d93c57a6fbb814445e2771
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:173152a9dfe5827d558a35b370c94b4fe85df0e6f0ceb2656a9287780c52959f
+size 25895
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..ddeb6f72b1424ef8003369c7800a4709ec22a0e0
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96d4242faf3badeb441287c25c765b60c0811ce7c1194383c6211b506657e50f
+size 25806
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..eb6b0364d2801802e1a43fae88a365f167b55e3e
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d42de6ac5db9f718d67df9452ca4115b288f1a6c26497a70e768991148793576
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1d0c0ef39f738c7d4ac78baf0f44435732aa0b3c
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d6e5f9293d37b913ff86ff2d48794178296eb78a2d06f81ef4eeb64495501b57
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/model.safetensors b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3398a4a486a2cad85413b7e22069e56b0abb9b67
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:62f1c65ee4d190c5ae7a9de41e334d462744981b7dff34871ab06317ec4ef1b0
+size 37776696
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..17fe2f7d9726a915e4cc408133680fb91809db91
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09299448d922df0a657d24d6eafc74a5d7e7fe69e637011382272d4b5abb476a
+size 50496
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..92e074abe5071dbf3d7f1bf0f09570a25090f237
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b4d53c9efb2078f06a3fbf14a4f1ae34de0abbdfb256675d510a2f9d149f70f
+size 25895
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..77dd1d7c29c166cbc53071ecb595a2605c0d7f65
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6b23378ecd744d664a68848fd7d589d75ff16dc8fff1550a9ba994c70987f7b
+size 25806
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0f68f92f6c2051bc7a92cdad34107769d4e9dfae
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b92bb310168a230f92a81c3137d6c91eb162ce0a6079b4f5100b9d917b1a6b18
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b23791cd26e40c621dd64b223aa1f959b01c58c2
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2d4dcf387324b5d0c9484968343ec27371bdc635082ca225c501cebdefb9cb3
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/model.safetensors b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..12a22f807f9b9c7e3a75ca0701dc7cf89503cec8
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c46155375c1c39fb7e2d3a3033c98f9b06d9b2ee40213a51b70797afd4d35e04
+size 37776696
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..bb184f4402ac23714fc5e384364b010372177139
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91bdac0315b99afa20afadf602551543e1882d7ea6003060c4c3bcd72ec32dbf
+size 50496
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fb00c6d9fd79ecb34da785ef9fbc0f71296c2721
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a89e7559a9a15f9fb38af1ad77fc19edc87dcacb0394948dcf3e180d89610758
+size 25895
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0c839aa70b5e0e757a80662d04440a047cf0d60f
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b2fdf31f3efdc179f1e5aaba83cbb61c32bd5c56f8c608059d5726b2206b60d
+size 25806
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7f5a31407c23bb11ba2bb891bf66346aa9a3f8af
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b64e6969b1a96eee7947caa804617938fe0be00090d1025128de161922c4b012
+size 25811
diff --git a/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3d303299266a69c7e1b4b7cb01796acd925e339f
--- /dev/null
+++ b/lm/gpt2-small_layer8/consdl/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f74643037d4b4e24599ef6ecf1d6ad785ee9cd598e643e94b3dac7dae5ac8050
+size 25811
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_128/code_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x16/l0_128/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..adad61cca7034ff34703a8e03860a4c4b941ae00
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_128/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 128.0,
+   "recon_r2": 0.945589449529994,
+   "l1_per_token": 5.18737556473359
+  },
+  "seed1": {
+   "l0": 128.0,
+   "recon_r2": 0.9455424848237902,
+   "l1_per_token": 5.199535936852198
+  },
+  "seed2": {
+   "l0": 128.0,
+   "recon_r2": 0.945648819029503,
+   "l1_per_token": 5.1838029704085375
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.44661482050032525,
+   "mcc_abs": 0.45489543985943,
+   "act_iou": 0.220281468509374,
+   "act_iou_n_pairs": 12288,
+   "input_r2": 0.9668218684965307,
+   "matched_mcc_signed": {
+    "0.6": 0.7865074049468922,
+    "0.7": 0.8179921652213032,
+    "0.8": 0.8531424749796515,
+    "0.9": 0.9022205161618765
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.7867254883624136,
+    "0.7": 0.8155344001649791,
+    "0.8": 0.8516281284036048,
+    "0.9": 0.9027916299190725
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.4515007741023263,
+   "mcc_abs": 0.4579362586753671,
+   "act_iou": 0.22091080282348405,
+   "act_iou_n_pairs": 12288,
+   "input_r2": 0.9670525579214755,
+   "matched_mcc_signed": {
+    "0.6": 0.7798976757975423,
+    "0.7": 0.8129910683696655,
+    "0.8": 0.8477365819257775,
+    "0.9": 0.9018931582635713
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.7841120325956743,
+    "0.7": 0.8134188101684917,
+    "0.8": 0.8489396538110224,
+    "0.9": 0.9032446830234664
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.4474619596107365,
+   "mcc_abs": 0.45473851992018094,
+   "act_iou": 0.21982558542596534,
+   "act_iou_n_pairs": 12288,
+   "input_r2": 0.9672638579455872,
+   "matched_mcc_signed": {
+    "0.6": 0.7825279135520201,
+    "0.7": 0.8143180289812273,
+    "0.8": 0.8516078891529016,
+    "0.9": 0.904005432434756
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.7846441745097401,
+    "0.7": 0.8149010222840262,
+    "0.8": 0.8521412827268509,
+    "0.9": 0.9041290439457793
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 128.0,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9455935844610958,
+  "recon_r2_std": 5.3287560157660226e-05,
+  "l1_per_token_mean": 5.190238157331442,
+  "l1_per_token_std": 0.00824787157788542,
+  "input_r2_mean": 0.9670460947878645,
+  "input_r2_std": 0.0002210655951100718,
+  "mcc_signed_mean": 0.4485258514044627,
+  "mcc_signed_std": 0.0026109452238212965,
+  "mcc_abs_mean": 0.45585673948499267,
+  "mcc_abs_std": 0.0018026247568989698,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.7829776647654848,
+   "0.7": 0.8151004208573988,
+   "0.8": 0.8508289820194435,
+   "0.9": 0.9027063689534013
+  },
+  "matched_mcc_signed_band": 0.8379033591489321,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.7851605651559427,
+   "0.7": 0.8146180775391656,
+   "0.8": 0.8509030216471594,
+   "0.9": 0.903388452296106
+  },
+  "matched_mcc_abs_band": 0.8385175291595934,
+  "act_iou_mean": 0.22033928558627447,
+  "act_iou_std": 0.0005449140397854574,
+  "act_iou_n_pairs": 12288.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_128/dict_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x16/l0_128/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..6987f9410260f3cf0577f18610f4d556772d2849
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_128/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 128.0,
+   "recon_r2": 0.9455109683486793,
+   "l1_per_token": 5.18835042448215,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 196,
+   "dense_rate": 0.015950520833333332,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 2,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 1,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 2,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.15897931158542633,
+   "n_dup_abs": {
+    "0.5": 2,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 1,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 2,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.16850043833255768
+  },
+  "seed1": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 128.0,
+   "recon_r2": 0.9454681899064855,
+   "l1_per_token": 5.200436899440258,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 198,
+   "dense_rate": 0.01611328125,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.15786662697792053,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.16716869175434113
+  },
+  "seed2": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 128.0,
+   "recon_r2": 0.9455750816523579,
+   "l1_per_token": 5.184698194675075,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 189,
+   "dense_rate": 0.015380859375,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.1597777158021927,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.16926686465740204
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.46659669280052185,
+   "dcs_abs": 0.7028051018714905,
+   "frac_sign_flipped": 0.5033365885416666,
+   "alive_source": "stats",
+   "n_alive_a": 12288,
+   "n_alive_b": 12288,
+   "matched_cos_signed": {
+    "0.5": 0.466621994972229,
+    "0.6": 0.46659669280052185,
+    "0.7": 0.46659669280052185,
+    "0.8": 0.46659669280052185,
+    "0.9": 0.46659669280052185
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9999593098958334,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.46660300807464716,
+    "0.6": 0.46659669280052185,
+    "0.7": 0.46659669280052185,
+    "0.8": 0.46659669280052185,
+    "0.9": 0.46659669280052185
+   },
+   "matched_frac_signed": {
+    "0.5": 0.43359375,
+    "0.6": 0.3683268229166667,
+    "0.7": 0.2920735677083333,
+    "0.8": 0.20499674479166666,
+    "0.9": 0.09342447916666667
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7028377652168274,
+    "0.6": 0.70280522108078,
+    "0.7": 0.70280522108078,
+    "0.8": 0.70280522108078,
+    "0.9": 0.70280522108078
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9999593098958334,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7028091666749485,
+    "0.6": 0.70280522108078,
+    "0.7": 0.70280522108078,
+    "0.8": 0.70280522108078,
+    "0.9": 0.70280522108078
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8250325520833334,
+    "0.6": 0.731201171875,
+    "0.7": 0.5926106770833334,
+    "0.8": 0.41162109375,
+    "0.9": 0.179443359375
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.47105374932289124,
+   "dcs_abs": 0.7047373652458191,
+   "frac_sign_flipped": 0.496826171875,
+   "alive_source": "stats",
+   "n_alive_a": 12288,
+   "n_alive_b": 12288,
+   "matched_cos_signed": {
+    "0.5": 0.47107118368148804,
+    "0.6": 0.47105374932289124,
+    "0.7": 0.47105374932289124,
+    "0.8": 0.47105374932289124,
+    "0.9": 0.47105374932289124
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9999593098958334,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.4710520157459541,
+    "0.6": 0.47105374932289124,
+    "0.7": 0.47105374932289124,
+    "0.8": 0.47105374932289124,
+    "0.9": 0.47105374932289124
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4457194010416667,
+    "0.6": 0.3793131510416667,
+    "0.7": 0.2977701822916667,
+    "0.8": 0.20841471354166666,
+    "0.9": 0.09098307291666667
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7047717571258545,
+    "0.6": 0.7047373652458191,
+    "0.7": 0.7047373652458191,
+    "0.8": 0.7047373652458191,
+    "0.9": 0.7047373652458191
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9999593098958334,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7047430798896434,
+    "0.6": 0.7047373652458191,
+    "0.7": 0.7047373652458191,
+    "0.8": 0.7047373652458191,
+    "0.9": 0.7047373652458191
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8318684895833334,
+    "0.6": 0.7347005208333334,
+    "0.7": 0.595703125,
+    "0.8": 0.4143880208333333,
+    "0.9": 0.17879231770833334
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.4698200523853302,
+   "dcs_abs": 0.7023658156394958,
+   "frac_sign_flipped": 0.5045572916666666,
+   "alive_source": "stats",
+   "n_alive_a": 12288,
+   "n_alive_b": 12288,
+   "matched_cos_signed": {
+    "0.5": 0.4698200523853302,
+    "0.6": 0.4698200523853302,
+    "0.7": 0.4698200523853302,
+    "0.8": 0.4698200523853302,
+    "0.9": 0.4698200523853302
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.4698200523853302,
+    "0.6": 0.4698200523853302,
+    "0.7": 0.4698200523853302,
+    "0.8": 0.4698200523853302,
+    "0.9": 0.4698200523853302
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4418131510416667,
+    "0.6": 0.3749186197916667,
+    "0.7": 0.2976888020833333,
+    "0.8": 0.20621744791666666,
+    "0.9": 0.09016927083333333
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7023658752441406,
+    "0.6": 0.7023658752441406,
+    "0.7": 0.7023658752441406,
+    "0.8": 0.7023658752441406,
+    "0.9": 0.7023658752441406
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7023658752441406,
+    "0.6": 0.7023658752441406,
+    "0.7": 0.7023658752441406,
+    "0.8": 0.7023658752441406,
+    "0.9": 0.7023658752441406
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8255208333333334,
+    "0.6": 0.7317708333333334,
+    "0.7": 0.59033203125,
+    "0.8": 0.4092610677083333,
+    "0.9": 0.18115234375
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 194,
+  "n_dense_std": 4.725815626252609,
+  "l0_mean": 128,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9455180799691743,
+  "recon_r2_std": null,
+  "l1_per_token_mean": 5.191161839532494,
+  "l1_per_token_std": 0.008237399338406275,
+  "babel_mu1_signed_mean": 0.15887455145517984,
+  "babel_mu1_signed_std": 0.0009598417250940007,
+  "n_dup_signed_per_threshold": {
+   "0.5": 0.6666666666666666,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_signed_band": 0.0,
+  "n_dup_signed_band_std": 0.0,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 0.3333333333333333,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_signed_band": 0.0,
+  "n_dup_clusters_signed_band_std": 0.0,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 1.3333333333333333,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_signed_band": 1.0,
+  "largest_dup_signed_band_std": 0.0,
+  "dcs_signed_mean": 0.46915683150291443,
+  "dcs_signed_std": 0.0023013549628540866,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.46915835873531053,
+   "0.6": 0.46915683150291443,
+   "0.7": 0.46915683150291443,
+   "0.8": 0.46915683150291443,
+   "0.9": 0.46915683150291443
+  },
+  "dedup_dcs_signed_band": 0.46915683150291443,
+  "dedup_dcs_signed_band_std": 0.0023013549628540866,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.46917107701301575,
+   "0.6": 0.46915683150291443,
+   "0.7": 0.46915683150291443,
+   "0.8": 0.46915683150291443,
+   "0.9": 0.46915683150291443
+  },
+  "matched_cos_signed_band": 0.46915683150291443,
+  "matched_cos_signed_band_std": 0.0023013549628540866,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.999972873263889,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_signed_band": 1.0,
+  "unique_frac_signed_band_std": 0.0,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.44037543402777785,
+   "0.6": 0.3741861979166667,
+   "0.7": 0.29584418402777773,
+   "0.8": 0.20654296875,
+   "0.9": 0.0915256076388889
+  },
+  "matched_frac_signed_band": 0.24202473958333331,
+  "matched_frac_signed_band_std": 0.002215930173655546,
+  "babel_mu1_abs_mean": 0.16831199824810028,
+  "babel_mu1_abs_std": 0.0010617036459660003,
+  "n_dup_abs_per_threshold": {
+   "0.5": 0.6666666666666666,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_abs_band": 0.0,
+  "n_dup_abs_band_std": 0.0,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 0.3333333333333333,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_abs_band": 0.0,
+  "n_dup_clusters_abs_band_std": 0.0,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 1.3333333333333333,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_abs_band": 1.0,
+  "largest_dup_abs_band_std": 0.0,
+  "dcs_abs_mean": 0.7033027609189352,
+  "dcs_abs_std": 0.0012616696395810702,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.7033060406029108,
+   "0.6": 0.70330282052358,
+   "0.7": 0.70330282052358,
+   "0.8": 0.70330282052358,
+   "0.9": 0.70330282052358
+  },
+  "dedup_dcs_abs_band": 0.70330282052358,
+  "dedup_dcs_abs_band_std": 0.0012616239975211717,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.7033251325289408,
+   "0.6": 0.70330282052358,
+   "0.7": 0.70330282052358,
+   "0.8": 0.70330282052358,
+   "0.9": 0.70330282052358
+  },
+  "matched_cos_abs_band": 0.70330282052358,
+  "matched_cos_abs_band_std": 0.0012616239975211717,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.999972873263889,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_abs_band": 1.0,
+  "unique_frac_abs_band_std": 0.0,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.8274739583333334,
+   "0.6": 0.7325575086805557,
+   "0.7": 0.5928819444444445,
+   "0.8": 0.4117567274305555,
+   "0.9": 0.17979600694444445
+  },
+  "matched_frac_abs_band": 0.47924804687500006,
+  "matched_frac_abs_band_std": 0.001457336402917159,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_128/seed0/config.json b/lm/gpt2-small_layer8/isae/x16/l0_128/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..d1a4836058c29cfdbfdadd94e9d4c640e079a7be
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_128/seed0/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..190fc03309a037e877e0e97e7c51eca305ee0467
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 128.0,
+  "l0_per_token_0.001": 128.0,
+  "l0_per_token_0.01": 127.80913598632813,
+  "l0_per_token_0.05": 24.870790087890626,
+  "l0_per_token_0.1": 7.483444873046875,
+  "l0_per_token_0.5": 0.001005615234375,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 5.18835042448215,
+  "r2_score": 0.9455109683486793,
+  "l2_loss": 1.277900409557822
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_128/seed1/config.json b/lm/gpt2-small_layer8/isae/x16/l0_128/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..28bd0a9a09fabf165083e2ee78027182c7ba5be3
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_128/seed1/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..15fc72308a47ed71de3917671ae6d86e0ecfa0d5
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1d0e143d9ffbaf873485e1df60ad7dbfd0d8c3ca3214bbd4fe8924032938705
+size 99648
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..6426b9a9d8c84bce2fb71b56e91b6363b96390d5
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 128.0,
+  "l0_per_token_0.001": 128.0,
+  "l0_per_token_0.01": 127.82004497070312,
+  "l0_per_token_0.05": 24.91362802734375,
+  "l0_per_token_0.1": 7.550155029296875,
+  "l0_per_token_0.5": 0.001125927734375,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 5.200436899440258,
+  "r2_score": 0.9454681899064855,
+  "l2_loss": 1.278903668142952
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_128/seed2/config.json b/lm/gpt2-small_layer8/isae/x16/l0_128/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..dc9d411495e2547d51bacace5eb57c467f1a668e
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_128/seed2/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..75801de33b1245dbb9ebf23fd12d333298e47038
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 128.0,
+  "l0_per_token_0.001": 128.0,
+  "l0_per_token_0.01": 127.81857329101562,
+  "l0_per_token_0.05": 24.772639013671874,
+  "l0_per_token_0.1": 7.4826326171875,
+  "l0_per_token_0.5": 0.00121044921875,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 5.184698194675075,
+  "r2_score": 0.9455750816523579,
+  "l2_loss": 1.27639679654533
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/code_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x16/l0_32/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..63e8377bb308ff1bba82767bad959c91d2e3a1a3
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 32.0,
+   "recon_r2": 0.9076311373317318,
+   "l1_per_token": 2.7880744834361444
+  },
+  "seed1": {
+   "l0": 32.0,
+   "recon_r2": 0.9032327229880623,
+   "l1_per_token": 2.741785103854914
+  },
+  "seed2": {
+   "l0": 32.0,
+   "recon_r2": 0.9077510756923542,
+   "l1_per_token": 2.7297401373159818
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.3282044006177079,
+   "mcc_abs": 0.3311824572280302,
+   "act_iou": 0.1815798897317736,
+   "act_iou_n_pairs": 12286,
+   "input_r2": 0.9590658116682543,
+   "matched_mcc_signed": {
+    "0.6": 0.8013857648228031,
+    "0.7": 0.8287989271582785,
+    "0.8": 0.8611609945154639,
+    "0.9": 0.8956226119891885
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8138433875632928,
+    "0.7": 0.8331688623205072,
+    "0.8": 0.8631321917022747,
+    "0.9": 0.8985541946800916
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.41680489894685513,
+   "mcc_abs": 0.42017986338030083,
+   "act_iou": 0.24097852763549765,
+   "act_iou_n_pairs": 12288,
+   "input_r2": 0.9636486827654452,
+   "matched_mcc_signed": {
+    "0.6": 0.8136763959756719,
+    "0.7": 0.8385592996844159,
+    "0.8": 0.8653836059153709,
+    "0.9": 0.8985684262374758
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8214555397384108,
+    "0.7": 0.8411278135728227,
+    "0.8": 0.8668740952643538,
+    "0.9": 0.8997132983905047
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.332541877641126,
+   "mcc_abs": 0.3353912739302589,
+   "act_iou": 0.18464022933944774,
+   "act_iou_n_pairs": 12286,
+   "input_r2": 0.9595575648021941,
+   "matched_mcc_signed": {
+    "0.6": 0.8005186496630173,
+    "0.7": 0.8287522793386951,
+    "0.8": 0.8583413204414153,
+    "0.9": 0.9005710607762802
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8134331002936288,
+    "0.7": 0.8343242351464992,
+    "0.8": 0.8609170943492641,
+    "0.9": 0.8990340336034716
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 32.0,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9062049786707161,
+  "recon_r2_std": 0.00257474740215192,
+  "l1_per_token_mean": 2.753199908202347,
+  "l1_per_token_std": 0.030796871643632224,
+  "input_r2_mean": 0.9607573530786313,
+  "input_r2_std": 0.002516007910276497,
+  "mcc_signed_mean": 0.35918372573522966,
+  "mcc_signed_std": 0.049948504766862405,
+  "mcc_abs_mean": 0.36225119817953,
+  "mcc_abs_std": 0.050211813584708234,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8051936034871642,
+   "0.7": 0.8320368353937965,
+   "0.8": 0.8616286402907501,
+   "0.9": 0.8982540330009815
+  },
+  "matched_mcc_signed_band": 0.8492782780431731,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.8162440091984441,
+   "0.7": 0.8362069703466096,
+   "0.8": 0.8636411271052976,
+   "0.9": 0.899100508891356
+  },
+  "matched_mcc_abs_band": 0.8537981538854268,
+  "act_iou_mean": 0.20239954890223966,
+  "act_iou_std": 0.03344539758628164,
+  "act_iou_n_pairs": 12286.666666666666,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/dict_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x16/l0_32/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..00839d0b424642781fb2986565daf5795e165636
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 32.0,
+   "recon_r2": 0.9075041504389101,
+   "l1_per_token": 2.788173881703426,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 10,
+   "dense_rate": 0.0008138020833333334,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.21489746868610382,
+   "n_dup_abs": {
+    "0.5": 2,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 1,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 2,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.22976021468639374
+  },
+  "seed1": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 32.0,
+   "recon_r2": 0.903132408550283,
+   "l1_per_token": 2.741906528374584,
+   "stats_source": "stats",
+   "n_dead": 3530,
+   "dead_rate": 0.2872721354166667,
+   "alive_source": "stats",
+   "n_dense": 10,
+   "dense_rate": 0.0008138020833333334,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 1865,
+    "0.6": 1102,
+    "0.7": 569,
+    "0.8": 190,
+    "0.9": 10
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 841,
+    "0.6": 519,
+    "0.7": 275,
+    "0.8": 92,
+    "0.9": 5
+   },
+   "largest_dup_signed": {
+    "0.5": 5,
+    "0.6": 5,
+    "0.7": 5,
+    "0.8": 3,
+    "0.9": 2
+   },
+   "babel_mu1_signed": 0.32149019837379456,
+   "n_dup_abs": {
+    "0.5": 2465,
+    "0.6": 1538,
+    "0.7": 828,
+    "0.8": 275,
+    "0.9": 18
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 1024,
+    "0.6": 684,
+    "0.7": 388,
+    "0.8": 132,
+    "0.9": 8
+   },
+   "largest_dup_abs": {
+    "0.5": 6,
+    "0.6": 6,
+    "0.7": 6,
+    "0.8": 5,
+    "0.9": 4
+   },
+   "babel_mu1_abs": 0.3567313253879547
+  },
+  "seed2": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 32.0,
+   "recon_r2": 0.9076146852669282,
+   "l1_per_token": 2.729719622368297,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 10,
+   "dense_rate": 0.0008138020833333334,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.21519283950328827,
+   "n_dup_abs": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 2,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 2,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.23018653690814972
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.45811787247657776,
+   "dcs_abs": 0.6325723528862,
+   "frac_sign_flipped": 0.4933268229166667,
+   "alive_source": "stats",
+   "n_alive_a": 12288,
+   "n_alive_b": 8758,
+   "matched_cos_signed": {
+    "0.5": 0.5591102242469788,
+    "0.6": 0.5581484436988831,
+    "0.7": 0.556789755821228,
+    "0.8": 0.5558221340179443,
+    "0.9": 0.5557036995887756
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9959612277867528,
+    "0.6": 0.9971015869999049,
+    "0.7": 0.9987170958852039,
+    "0.8": 0.9998099401311413,
+    "0.9": 0.9999524850327853
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5568521054091476,
+    "0.6": 0.5565306989936833,
+    "0.7": 0.5560754479524086,
+    "0.8": 0.5557164945360441,
+    "0.9": 0.5556772953457086
+   },
+   "matched_frac_signed": {
+    "0.5": 0.43105578257151,
+    "0.6": 0.3903829706357503,
+    "0.7": 0.3434381830276537,
+    "0.8": 0.27986315689442176,
+    "0.9": 0.17742088757958757
+   },
+   "matched_cos_abs": {
+    "0.5": 0.824739396572113,
+    "0.6": 0.8216115832328796,
+    "0.7": 0.8180738091468811,
+    "0.8": 0.8155664801597595,
+    "0.9": 0.8149334788322449
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9932053596883018,
+    "0.6": 0.9953435332129621,
+    "0.7": 0.9977667965409104,
+    "0.8": 0.9994773353606387,
+    "0.9": 0.9999049700655707
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8191355890215185,
+    "0.6": 0.8177857761837101,
+    "0.7": 0.8162468838865037,
+    "0.8": 0.8151402123995316,
+    "0.9": 0.8148560357571871
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7729734866482942,
+    "0.6": 0.7420887579587571,
+    "0.7": 0.6810795400551174,
+    "0.8": 0.5564952960182458,
+    "0.9": 0.35484177515917514
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.5144454836845398,
+   "dcs_abs": 0.7818900942802429,
+   "frac_sign_flipped": 0.5045572916666666,
+   "alive_source": "stats",
+   "n_alive_a": 12288,
+   "n_alive_b": 12288,
+   "matched_cos_signed": {
+    "0.5": 0.5144454836845398,
+    "0.6": 0.5144454836845398,
+    "0.7": 0.5144454836845398,
+    "0.8": 0.5144454836845398,
+    "0.9": 0.5144454836845398
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5144454836845398,
+    "0.6": 0.5144454836845398,
+    "0.7": 0.5144454836845398,
+    "0.8": 0.5144454836845398,
+    "0.9": 0.5144454836845398
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4568684895833333,
+    "0.6": 0.4185384114583333,
+    "0.7": 0.3766276041666667,
+    "0.8": 0.3180338541666667,
+    "0.9": 0.21818033854166666
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7818711400032043,
+    "0.6": 0.7819421291351318,
+    "0.7": 0.7819421291351318,
+    "0.8": 0.7818900942802429,
+    "0.9": 0.7818900942802429
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9999186197916666,
+    "0.6": 0.9999593098958334,
+    "0.7": 0.9999593098958334,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.781807511166941,
+    "0.6": 0.7819103118284451,
+    "0.7": 0.7819103118284451,
+    "0.8": 0.7818900942802429,
+    "0.9": 0.7818900942802429
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8575846354166666,
+    "0.6": 0.817138671875,
+    "0.7": 0.75537109375,
+    "0.8": 0.6463216145833334,
+    "0.9": 0.451171875
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.45954182744026184,
+   "dcs_abs": 0.6315269470214844,
+   "frac_sign_flipped": 0.49267578125,
+   "alive_source": "stats",
+   "n_alive_a": 8758,
+   "n_alive_b": 12288,
+   "matched_cos_signed": {
+    "0.5": 0.5638695359230042,
+    "0.6": 0.562885046005249,
+    "0.7": 0.5614797472953796,
+    "0.8": 0.5604748725891113,
+    "0.9": 0.5603472590446472
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9959612277867528,
+    "0.6": 0.9971015869999049,
+    "0.7": 0.9987170958852039,
+    "0.8": 0.9998099401311413,
+    "0.9": 0.9999524850327853
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5615921953094217,
+    "0.6": 0.5612535726703483,
+    "0.7": 0.5607594226171997,
+    "0.8": 0.5603683488083284,
+    "0.9": 0.5603206341630049
+   },
+   "matched_frac_signed": {
+    "0.5": 0.43637745889955337,
+    "0.6": 0.3944692578162121,
+    "0.7": 0.3482847096835503,
+    "0.8": 0.2839494440748836,
+    "0.9": 0.18008172574360923
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8230814933776855,
+    "0.6": 0.8199750781059265,
+    "0.7": 0.8164289593696594,
+    "0.8": 0.814014732837677,
+    "0.9": 0.8133817911148071
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9932053596883018,
+    "0.6": 0.9952960182457474,
+    "0.7": 0.9977192815736957,
+    "0.8": 0.9994773353606387,
+    "0.9": 0.9999049700655707
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8174889506829688,
+    "0.6": 0.8161179302995744,
+    "0.7": 0.8145669147982566,
+    "0.8": 0.8135892761209036,
+    "0.9": 0.8133044954965315
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7724033070417181,
+    "0.6": 0.7386676803193006,
+    "0.7": 0.6740473249073459,
+    "0.8": 0.5543096075263708,
+    "0.9": 0.3530362064050176
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 1177,
+  "n_dead_std": 2038.0464502393788,
+  "n_dense_mean": 10,
+  "n_dense_std": 0.0,
+  "l0_mean": 32,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9060837480853738,
+  "recon_r2_std": null,
+  "l1_per_token_mean": 2.7532666774821024,
+  "l1_per_token_std": 0.0308385286518895,
+  "babel_mu1_signed_mean": 0.2505268355210622,
+  "babel_mu1_signed_std": 0.061456252419963955,
+  "n_dup_signed_per_threshold": {
+   "0.5": 621.6666666666666,
+   "0.6": 367.3333333333333,
+   "0.7": 189.66666666666666,
+   "0.8": 63.333333333333336,
+   "0.9": 3.3333333333333335
+  },
+  "n_dup_signed_band": 155.91666666666669,
+  "n_dup_signed_band_std": 270.0555884134475,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 280.3333333333333,
+   "0.6": 173.0,
+   "0.7": 91.66666666666667,
+   "0.8": 30.666666666666668,
+   "0.9": 1.6666666666666667
+  },
+  "n_dup_clusters_signed_band": 74.25000000000001,
+  "n_dup_clusters_signed_band_std": 128.60477246198914,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 2.3333333333333335,
+   "0.6": 2.3333333333333335,
+   "0.7": 2.3333333333333335,
+   "0.8": 1.6666666666666667,
+   "0.9": 1.3333333333333333
+  },
+  "largest_dup_signed_band": 1.9166666666666667,
+  "largest_dup_signed_band_std": 1.5877132402714706,
+  "dcs_signed_mean": 0.47736839453379315,
+  "dcs_signed_std": 0.03211759357196554,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.5442965948010364,
+   "0.6": 0.5440765851161905,
+   "0.7": 0.543760118084716,
+   "0.8": 0.5435101090096374,
+   "0.9": 0.5434811377310844
+  },
+  "dedup_dcs_signed_band": 0.543706987485407,
+  "dedup_dcs_signed_band_std": 0.025448807492542885,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.5458084146181742,
+   "0.6": 0.545159657796224,
+   "0.7": 0.5442383289337158,
+   "0.8": 0.5435808300971985,
+   "0.9": 0.5434988141059875
+  },
+  "matched_cos_signed_band": 0.5441194077332815,
+  "matched_cos_signed_band_std": 0.02580472085057212,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.9973074851911686,
+   "0.6": 0.9980677246666033,
+   "0.7": 0.9991447305901359,
+   "0.8": 0.9998732934207609,
+   "0.9": 0.9999683233551903
+  },
+  "unique_frac_signed_band": 0.9992635180081726,
+  "unique_frac_signed_band_std": 0.0006378121143522995,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.44143391035146556,
+   "0.6": 0.40113021330343185,
+   "0.7": 0.35611683229262353,
+   "0.8": 0.29394881837865733,
+   "0.9": 0.19189431728828785
+  },
+  "matched_frac_signed_band": 0.31077254531575016,
+  "matched_frac_signed_band_std": 0.019215572758895984,
+  "babel_mu1_abs_mean": 0.2722260256608327,
+  "babel_mu1_abs_std": 0.07318404675294496,
+  "n_dup_abs_per_threshold": {
+   "0.5": 823.0,
+   "0.6": 513.3333333333334,
+   "0.7": 276.6666666666667,
+   "0.8": 91.66666666666667,
+   "0.9": 6.0
+  },
+  "n_dup_abs_band": 221.91666666666666,
+  "n_dup_abs_band_std": 383.50524225013316,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 342.0,
+   "0.6": 228.33333333333334,
+   "0.7": 129.66666666666666,
+   "0.8": 44.0,
+   "0.9": 2.6666666666666665
+  },
+  "n_dup_clusters_abs_band": 101.16666666666667,
+  "n_dup_clusters_abs_band_std": 174.79297278018169,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 3.3333333333333335,
+   "0.6": 3.0,
+   "0.7": 3.0,
+   "0.8": 2.3333333333333335,
+   "0.9": 2.0
+  },
+  "largest_dup_abs_band": 2.5833333333333335,
+  "largest_dup_abs_band_std": 2.322893310794392,
+  "dcs_abs_mean": 0.6819964647293091,
+  "dcs_abs_std": 0.08651199995954259,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.8061440169571427,
+   "0.6": 0.8052713394372432,
+   "0.7": 0.8042413701710686,
+   "0.8": 0.8035398609335594,
+   "0.9": 0.8033502085113206
+  },
+  "dedup_dcs_abs_band": 0.8041006947632979,
+  "dedup_dcs_abs_band_std": 0.01924308893891894,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.8098973433176676,
+   "0.6": 0.8078429301579794,
+   "0.7": 0.8054816325505575,
+   "0.8": 0.8038237690925598,
+   "0.9": 0.803401788075765
+  },
+  "matched_cos_abs_band": 0.8051375299692155,
+  "matched_cos_abs_band_std": 0.020126168566350064,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.9954431130560901,
+   "0.6": 0.9968662871181809,
+   "0.7": 0.9984817960034799,
+   "0.8": 0.9996515569070924,
+   "0.9": 0.9999366467103804
+  },
+  "unique_frac_abs_band": 0.9987340716847833,
+  "unique_frac_abs_band_std": 0.001078772150899022,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.8009871430355596,
+   "0.6": 0.7659650367176859,
+   "0.7": 0.7034993195708211,
+   "0.8": 0.5857088393759833,
+   "0.9": 0.3863499521880642
+  },
+  "matched_frac_abs_band": 0.6103807869631386,
+  "matched_frac_abs_band_std": 0.04950033513106061,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/config.json b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..453661d3a2732a1758a572f2b2d5b55d9d69be6e
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..afbab17d2a7c87272f0ca9958f131e0b739dc5fb
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35aececace33b219d2bd406cb84ddcd74a2bd62111eeca19e185f0abe4b21f70
+size 99648
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9926bb72cc09acaf53727f0180edd8cadb76545e
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc8c2c78bb88967fe77055378a5b165b236695e4f1e4fb09f1f2aedef270d6eb
+size 50471
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..1da0d34b7f4c907f95a9f971a19b3c968363583b
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 32.0,
+  "l0_per_token_0.001": 32.0,
+  "l0_per_token_0.01": 31.980476171875,
+  "l0_per_token_0.05": 24.695182568359375,
+  "l0_per_token_0.1": 7.659456884765625,
+  "l0_per_token_0.5": 0.008879248046875,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 2.788173881703426,
+  "r2_score": 0.9075041504389101,
+  "l2_loss": 2.169252791880922
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..70aae5135b511850564337e737dfb92adcfab377
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b037e6c7eee9ccfb5706a05a3489dd2cc2025b25a006258edbb343d53d8175ed
+size 50382
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9160c63d17c13fdf0138120cd77fd41b5e8b4e7f
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92103cca5f49e111c7a15d761cf11c15fed480b6bd48929d76193d2fea1f6aaa
+size 50387
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9b3c3f2202646c2c3863e588fdbbae8b787fc839
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c21aafb6ae64519618843ddbb38d62e47ba0f9974464b0f6a89328909ca45c1
+size 50387
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/config.json b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..ac5d516faa65f1748a0ccabb4d1d287f5a2b30c8
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/model.safetensors b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..54da2f6b7852d1af99fdf2c23b76a43e324f9bb3
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5716777f37953890cf4384122c5b1bc155e1d4fead4feb1f335ae1ab967f135b
+size 75550008
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..afbab17d2a7c87272f0ca9958f131e0b739dc5fb
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35aececace33b219d2bd406cb84ddcd74a2bd62111eeca19e185f0abe4b21f70
+size 99648
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..89dd4dd6b541333e769d2de4c837f753e8258b2a
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b8046070faa6237c845b2175b92ef73fdbd6c2b7c422c469015fbeff432cea8
+size 50471
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..27a3292da3259fc385a9bf3ea6bc20a7d18d1cbb
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 32.0,
+  "l0_per_token_0.001": 32.0,
+  "l0_per_token_0.01": 31.9814556640625,
+  "l0_per_token_0.05": 24.825621044921874,
+  "l0_per_token_0.1": 7.532770849609375,
+  "l0_per_token_0.5": 0.00570791015625,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 2.741906528374584,
+  "r2_score": 0.903132408550283,
+  "l2_loss": 2.271780779269413
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4f0c800c52fc252e8e9b14f8bbc878f7a7e08f89
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:95b12370c24dbaf893aa43da19ced221b3d0448b68ee43b2aac2107d8b113ee9
+size 50382
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f43e025f129ab17a5eba27cf29960104c2bccd83
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1260abb70fc7fcb3230e0c1d86bf6cd2febc378d034450cf5de28b0024d00702
+size 50387
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..20e33f65f74393c048747565ccc82b660d5508f7
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6824c828d1d75d2168e1a58ac91ac553af0442ecaee95dbfd44c3087e209af47
+size 50387
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/config.json b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..06369c818260ca5170c0dfe7473da75248c01f80
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/model.safetensors b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4a5856a14a46a1e5f52aa7e88985c07d5fc08326
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7778cfa86aa969bab86b5d913fc605ec69dedbf307b7283f0859258b05423ca1
+size 75550008
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..afbab17d2a7c87272f0ca9958f131e0b739dc5fb
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35aececace33b219d2bd406cb84ddcd74a2bd62111eeca19e185f0abe4b21f70
+size 99648
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5ea45779542a1e99c41e01e61622a6865d95b74e
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e46680c85e8f370323d6f8a792e242ef60cb55f46cb4db37f2ff4ed887b817f
+size 50471
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..293ff5399b3bf8275c13ca42850c7b92666141c5
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 32.0,
+  "l0_per_token_0.001": 32.0,
+  "l0_per_token_0.01": 31.979508984375,
+  "l0_per_token_0.05": 24.596323974609376,
+  "l0_per_token_0.1": 7.5045275390625,
+  "l0_per_token_0.5": 0.00736611328125,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 2.729719622368297,
+  "r2_score": 0.9076146852669282,
+  "l2_loss": 2.1666604811403207
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f4f6ff8e2282b98144b98e065e38282880babf52
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:686247134a82398f35f4f52de5257ed6792b1a2810ddb5c0d06845660d30c1bb
+size 50382
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..80337f35553e03ae20729d6dfe480d60142bd964
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a63333143e8e3149a3032c6cc4d0bd32c25f05c4a6169a6508e9ea8deac87efb
+size 50387
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..34bf40f9c9f79c8ba00aee37ed494697abaa29bb
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c087793d7af08bf87b54b1256181870e4b120eee3e8d64a2385182a1efac1d10
+size 50387
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/code_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x16/l0_64/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..eb0871d51b417fca064aa13957d8cac6788ff71d
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 64.0,
+   "recon_r2": 0.9269195970961246,
+   "l1_per_token": 3.722019245364733
+  },
+  "seed1": {
+   "l0": 64.0,
+   "recon_r2": 0.9269815367136025,
+   "l1_per_token": 3.7283920260221195
+  },
+  "seed2": {
+   "l0": 64.0,
+   "recon_r2": 0.9271251690954544,
+   "l1_per_token": 3.729575516361588
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.44074188066666425,
+   "mcc_abs": 0.44714264636398554,
+   "act_iou": 0.2383903007836438,
+   "act_iou_n_pairs": 12288,
+   "input_r2": 0.9659676711196822,
+   "matched_mcc_signed": {
+    "0.6": 0.8053341496681989,
+    "0.7": 0.8300272093133901,
+    "0.8": 0.8596593692594189,
+    "0.9": 0.8978988307027671
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8092910614287908,
+    "0.7": 0.8304760870305998,
+    "0.8": 0.8595327928252863,
+    "0.9": 0.8974035114038839
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.4465127915972937,
+   "mcc_abs": 0.45332933667583264,
+   "act_iou": 0.24328324473363186,
+   "act_iou_n_pairs": 12288,
+   "input_r2": 0.9665342539110171,
+   "matched_mcc_signed": {
+    "0.6": 0.8053749973124064,
+    "0.7": 0.8322994247177166,
+    "0.8": 0.8610475457004325,
+    "0.9": 0.8976801317410574
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8103171475047142,
+    "0.7": 0.8322902419128676,
+    "0.8": 0.8597311340758257,
+    "0.9": 0.8974318991191593
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.44722796223383066,
+   "mcc_abs": 0.45413844161763994,
+   "act_iou": 0.24305100186981118,
+   "act_iou_n_pairs": 12288,
+   "input_r2": 0.9664387858331096,
+   "matched_mcc_signed": {
+    "0.6": 0.8035003840595296,
+    "0.7": 0.830781293422454,
+    "0.8": 0.8589589152279573,
+    "0.9": 0.8993687470440612
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8082935192554529,
+    "0.7": 0.8309570772703431,
+    "0.8": 0.8591261237514658,
+    "0.9": 0.8981550821989127
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 64.0,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9270087676350606,
+  "recon_r2_std": 0.00010545664530839875,
+  "l1_per_token_mean": 3.7266622625828134,
+  "l1_per_token_std": 0.004064279642377905,
+  "input_r2_mean": 0.9663135702879364,
+  "input_r2_std": 0.0003033368003127958,
+  "mcc_signed_mean": 0.4448275448325962,
+  "mcc_signed_std": 0.0035563121372323946,
+  "mcc_abs_mean": 0.4515368082191527,
+  "mcc_abs_std": 0.00382689907219525,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8047365103467117,
+   "0.7": 0.8310359758178536,
+   "0.8": 0.8598886100626029,
+   "0.9": 0.8983159031626285
+  },
+  "matched_mcc_signed_band": 0.8484942498474491,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.809300576062986,
+   "0.7": 0.8312411354046035,
+   "0.8": 0.859463350217526,
+   "0.9": 0.8976634975739852
+  },
+  "matched_mcc_abs_band": 0.8494171398147752,
+  "act_iou_mean": 0.24157484912902893,
+  "act_iou_std": 0.0027603433319100734,
+  "act_iou_n_pairs": 12288.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/dict_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x16/l0_64/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..fcab95fe88f1bb720289a408e8ffa91c7f9fb06a
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 64.0,
+   "recon_r2": 0.9268360661429926,
+   "l1_per_token": 3.7229563668976313,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 44,
+   "dense_rate": 0.0035807291666666665,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 4,
+    "0.6": 2,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 2,
+    "0.6": 1,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.17768387496471405,
+   "n_dup_abs": {
+    "0.5": 4,
+    "0.6": 2,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 2,
+    "0.6": 1,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.18925358355045319
+  },
+  "seed1": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 64.0,
+   "recon_r2": 0.9268818873063308,
+   "l1_per_token": 3.7292008090152566,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 50,
+   "dense_rate": 0.004069010416666667,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 4,
+    "0.6": 2,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 2,
+    "0.6": 1,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.1765168160200119,
+   "n_dup_abs": {
+    "0.5": 6,
+    "0.6": 2,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 3,
+    "0.6": 1,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.1880100965499878
+  },
+  "seed2": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 64.0,
+   "recon_r2": 0.9270372596384872,
+   "l1_per_token": 3.7307122829036277,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 45,
+   "dense_rate": 0.003662109375,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 2,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 1,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 2,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.17857377231121063,
+   "n_dup_abs": {
+    "0.5": 2,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 1,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 2,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.19025833904743195
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.49421754479408264,
+   "dcs_abs": 0.763469934463501,
+   "frac_sign_flipped": 0.5159505208333334,
+   "alive_source": "stats",
+   "n_alive_a": 12288,
+   "n_alive_b": 12288,
+   "matched_cos_signed": {
+    "0.5": 0.4941669702529907,
+    "0.6": 0.49418017268180847,
+    "0.7": 0.49421754479408264,
+    "0.8": 0.49421754479408264,
+    "0.9": 0.49421754479408264
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9998372395833334,
+    "0.6": 0.9999186197916666,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.49408653943100944,
+    "0.6": 0.4941399561964014,
+    "0.7": 0.49421754479408264,
+    "0.8": 0.49421754479408264,
+    "0.9": 0.49421754479408264
+   },
+   "matched_frac_signed": {
+    "0.5": 0.457763671875,
+    "0.6": 0.406494140625,
+    "0.7": 0.3531901041666667,
+    "0.8": 0.2803548177083333,
+    "0.9": 0.16357421875
+   },
+   "matched_cos_abs": {
+    "0.5": 0.763466477394104,
+    "0.6": 0.7634544372558594,
+    "0.7": 0.763469934463501,
+    "0.8": 0.763469934463501,
+    "0.9": 0.763469934463501
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9997965494791666,
+    "0.6": 0.9999186197916666,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7633111497416394,
+    "0.6": 0.7633923071747024,
+    "0.7": 0.763469934463501,
+    "0.8": 0.763469934463501,
+    "0.9": 0.763469934463501
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8695475260416666,
+    "0.6": 0.8147786458333334,
+    "0.7": 0.7250162760416666,
+    "0.8": 0.5780436197916666,
+    "0.9": 0.3373209635416667
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.5009267330169678,
+   "dcs_abs": 0.7650578618049622,
+   "frac_sign_flipped": 0.50244140625,
+   "alive_source": "stats",
+   "n_alive_a": 12288,
+   "n_alive_b": 12288,
+   "matched_cos_signed": {
+    "0.5": 0.5009841322898865,
+    "0.6": 0.5009605884552002,
+    "0.7": 0.5009267330169678,
+    "0.8": 0.5009267330169678,
+    "0.9": 0.5009267330169678
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9998779296875,
+    "0.6": 0.9999593098958334,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5009229770003003,
+    "0.6": 0.5009402043166726,
+    "0.7": 0.5009267330169678,
+    "0.8": 0.5009267330169678,
+    "0.9": 0.5009267330169678
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4654947916666667,
+    "0.6": 0.417724609375,
+    "0.7": 0.3626302083333333,
+    "0.8": 0.2888997395833333,
+    "0.9": 0.17447916666666666
+   },
+   "matched_cos_abs": {
+    "0.5": 0.765082597732544,
+    "0.6": 0.7650459408760071,
+    "0.7": 0.7650578618049622,
+    "0.8": 0.7650578618049622,
+    "0.9": 0.7650578618049622
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9998779296875,
+    "0.6": 0.9999593098958334,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7649892038607504,
+    "0.6": 0.7650148110769805,
+    "0.7": 0.7650578618049622,
+    "0.8": 0.7650578618049622,
+    "0.9": 0.7650578618049622
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8717447916666666,
+    "0.6": 0.8148600260416666,
+    "0.7": 0.7255859375,
+    "0.8": 0.5829264322916666,
+    "0.9": 0.3451334635416667
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.5017046332359314,
+   "dcs_abs": 0.764442503452301,
+   "frac_sign_flipped": 0.50048828125,
+   "alive_source": "stats",
+   "n_alive_a": 12288,
+   "n_alive_b": 12288,
+   "matched_cos_signed": {
+    "0.5": 0.5017666220664978,
+    "0.6": 0.5017353296279907,
+    "0.7": 0.5017046332359314,
+    "0.8": 0.5017046332359314,
+    "0.9": 0.5017046332359314
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9998779296875,
+    "0.6": 0.9999593098958334,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5017053712581401,
+    "0.6": 0.5017149139651641,
+    "0.7": 0.5017046332359314,
+    "0.8": 0.5017046332359314,
+    "0.9": 0.5017046332359314
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4703776041666667,
+    "0.6": 0.4202473958333333,
+    "0.7": 0.361572265625,
+    "0.8": 0.2880859375,
+    "0.9": 0.166259765625
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7645003199577332,
+    "0.6": 0.7644320130348206,
+    "0.7": 0.7644426226615906,
+    "0.8": 0.7644426226615906,
+    "0.9": 0.7644426226615906
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9998372395833334,
+    "0.6": 0.9999593098958334,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7643758895671151,
+    "0.6": 0.7644009082165819,
+    "0.7": 0.7644426226615906,
+    "0.8": 0.7644426226615906,
+    "0.9": 0.7644426226615906
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8722330729166666,
+    "0.6": 0.8177897135416666,
+    "0.7": 0.7257486979166666,
+    "0.8": 0.5787760416666666,
+    "0.9": 0.336181640625
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 46,
+  "n_dense_std": 3.2145502536643185,
+  "l0_mean": 64,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9269184043626035,
+  "recon_r2_std": null,
+  "l1_per_token_mean": 3.7276231529388384,
+  "l1_per_token_std": 0.004111606412901383,
+  "babel_mu1_signed_mean": 0.1775914877653122,
+  "babel_mu1_signed_std": 0.0010315855960294801,
+  "n_dup_signed_per_threshold": {
+   "0.5": 3.3333333333333335,
+   "0.6": 1.3333333333333333,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_signed_band": 0.3333333333333333,
+  "n_dup_signed_band_std": 0.2886751345948129,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 1.6666666666666667,
+   "0.6": 0.6666666666666666,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_signed_band": 0.16666666666666666,
+  "n_dup_clusters_signed_band_std": 0.14433756729740646,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 2.0,
+   "0.6": 1.6666666666666667,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_signed_band": 1.1666666666666667,
+  "largest_dup_signed_band_std": 0.14433756729740646,
+  "dcs_signed_mean": 0.4989496370156606,
+  "dcs_signed_std": 0.004116528243897387,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.4989049625631499,
+   "0.6": 0.498931691492746,
+   "0.7": 0.4989496370156606,
+   "0.8": 0.4989496370156606,
+   "0.9": 0.4989496370156606
+  },
+  "dedup_dcs_signed_band": 0.49894515063493194,
+  "dedup_dcs_signed_band_std": 0.0041293462128722865,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.4989725748697917,
+   "0.6": 0.49895869692166644,
+   "0.7": 0.4989496370156606,
+   "0.8": 0.4989496370156606,
+   "0.9": 0.4989496370156606
+  },
+  "matched_cos_signed_band": 0.49895190199216205,
+  "matched_cos_signed_band_std": 0.004126499005503968,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.9998643663194445,
+   "0.6": 0.9999457465277778,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_signed_band": 0.9999864366319444,
+  "unique_frac_signed_band_std": 5.8731106485374604e-06,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.46454535590277785,
+   "0.6": 0.4148220486111111,
+   "0.7": 0.359130859375,
+   "0.8": 0.2857801649305555,
+   "0.9": 0.16810438368055555
+  },
+  "matched_frac_signed_band": 0.3069593641493056,
+  "matched_frac_signed_band_std": 0.005329329339531221,
+  "babel_mu1_abs_mean": 0.1891740063826243,
+  "babel_mu1_abs_std": 0.001126231759478585,
+  "n_dup_abs_per_threshold": {
+   "0.5": 4.0,
+   "0.6": 1.3333333333333333,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_abs_band": 0.3333333333333333,
+  "n_dup_abs_band_std": 0.2886751345948129,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 2.0,
+   "0.6": 0.6666666666666666,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_abs_band": 0.16666666666666666,
+  "n_dup_clusters_abs_band_std": 0.14433756729740646,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 2.0,
+   "0.6": 1.6666666666666667,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_abs_band": 1.1666666666666667,
+  "largest_dup_abs_band_std": 0.14433756729740646,
+  "dcs_abs_mean": 0.7643234332402548,
+  "dcs_abs_std": 0.0008006319984784258,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.764225414389835,
+   "0.6": 0.7642693421560883,
+   "0.7": 0.7643234729766846,
+   "0.8": 0.7643234729766846,
+   "0.9": 0.7643234729766846
+  },
+  "dedup_dcs_abs_band": 0.7643099402715355,
+  "dedup_dcs_abs_band_std": 0.0008052761103128397,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.7643497983614603,
+   "0.6": 0.7643107970555624,
+   "0.7": 0.7643234729766846,
+   "0.8": 0.7643234729766846,
+   "0.9": 0.7643234729766846
+  },
+  "matched_cos_abs_band": 0.764320303996404,
+  "matched_cos_abs_band_std": 0.0008011419219994612,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.9998372395833334,
+   "0.6": 0.9999457465277778,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_abs_band": 0.9999864366319444,
+  "unique_frac_abs_band_std": 5.8731106485374604e-06,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.8711751302083334,
+   "0.6": 0.8158094618055555,
+   "0.7": 0.7254503038194443,
+   "0.8": 0.5799153645833334,
+   "0.9": 0.33954535590277785
+  },
+  "matched_frac_abs_band": 0.6151801215277777,
+  "matched_frac_abs_band_std": 0.0017364157303489144,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/config.json b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..fc536eebc53f7224072b82e212ac5b2c9a4043f6
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/model.safetensors b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8acee4f33ef724de3dcda3e6074eb6d0a4665776
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:195690e8f8de6c698531d0296de5530b070f0e140843b89bd90fd6c81030fa7a
+size 75550008
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8de83be649f2f0cc84109b3da72b06a86da913fc
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53a972b5048d9cd93f2f67cbe0716fdc3f9f7eef130ca0bcd46ceccf0b967cde
+size 99648
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9853da34cdb1a97a22ee3c24734fa036f536bc07
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ebe8c00d3a0d27f1c6160ffbdf519ca76ebba7c681675e8a9edb5b2b942d563
+size 50471
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..d0ddbc62482b0b638f937a99ac112879e8b6294e
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.0,
+  "l0_per_token_0.001": 64.0,
+  "l0_per_token_0.01": 63.9051845703125,
+  "l0_per_token_0.05": 22.507840185546875,
+  "l0_per_token_0.1": 7.52195322265625,
+  "l0_per_token_0.5": 0.00433623046875,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 3.7229563668976313,
+  "r2_score": 0.9268360661429926,
+  "l2_loss": 1.7158723179193245
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..6fefb4ec1d3ed230481ab6930984722ae4356996
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9545c6e6f742a9ab0e271e5be6e5d946f295bea1c4334f561f9d1ac1a43d5b16
+size 50382
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0cb196c8f4a061afb17fe5804be8d788eaf8cf3b
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ce1d912b07c393ac018c859ded62c0bd5af2f713d9ee3106ce410fc77728db4
+size 50387
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3d7d5e8d178645664f49a756d9f4b4a9c90d4668
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:497603cea84da5054f3b66ad6ee8ab9f96b61f84b30b8290acb9f7336bd3e0ba
+size 50387
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/config.json b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..ba8c1310f91975954f6b4828f0c9c0dc066d6f63
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/model.safetensors b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..65410bd5bb88f3fc47c8e11cac8ee3a3b74970eb
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc62a492db6c9ee562984b76e4b0432dfeabd0f1fe7d6c5e77cb77c53788b415
+size 75550008
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8de83be649f2f0cc84109b3da72b06a86da913fc
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53a972b5048d9cd93f2f67cbe0716fdc3f9f7eef130ca0bcd46ceccf0b967cde
+size 99648
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..76ac332012aaac52a57b90ab2dbb69bb5ff454d1
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23f74df7d9b8dd300b78ffdd787b0bb9c13346435c4a9727046671386c90d7aa
+size 50471
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..85656c9b2d093f3897e4b4c3022d7f8b361083ab
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.0,
+  "l0_per_token_0.001": 64.0,
+  "l0_per_token_0.01": 63.911101123046876,
+  "l0_per_token_0.05": 22.53399365234375,
+  "l0_per_token_0.1": 7.5544091796875,
+  "l0_per_token_0.5": 0.0039447265625,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 3.7292008090152566,
+  "r2_score": 0.9268818873063308,
+  "l2_loss": 1.714797700117875
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0dd99eab60af45085bea466bc051d3d6a4689c4a
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:30b761635484cbb44530b13e286dec74eb61d2e4ddd95d1d777c1386e1ac032d
+size 50382
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3689689992ea031e63dac5a459782c850d4425d6
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0eed64b1c67dda414c37cee29a7fb22b06347e90f240f35660c63a405ebcc9a5
+size 50387
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..dc12dedef11a5f23734c31b146e862984189b40e
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e57515cb0efdab43e023b638108d86068ee72aee99e55c434f5e56b2c8c66fc7
+size 50387
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/config.json b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..72c099534805adbcab623b82c98d6f0a5bd89862
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/model.safetensors b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..52c33cbb7a5963e6ff7ccf2bc60db04f461e0566
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83924bbead36351425e7da54e7a9339125c04c55c546d1f806272e08a49d1c4d
+size 75550008
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8de83be649f2f0cc84109b3da72b06a86da913fc
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53a972b5048d9cd93f2f67cbe0716fdc3f9f7eef130ca0bcd46ceccf0b967cde
+size 99648
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7168e149c6fa7b8f1748decc8dceeb3ba08fe241
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:494df049dacf2326cd38c756556e6db75fe756aeb48c4d7651c07bb4787c87ba
+size 50471
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..207f53484c95c66fe107196f5a637b40a68eed33
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.0,
+  "l0_per_token_0.001": 64.0,
+  "l0_per_token_0.01": 63.90602529296875,
+  "l0_per_token_0.05": 22.464368115234375,
+  "l0_per_token_0.1": 7.50871357421875,
+  "l0_per_token_0.5": 0.004605029296875,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 3.7307122829036277,
+  "r2_score": 0.9270372596384872,
+  "l2_loss": 1.7111538407781253
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9df1531a0ed7c3f1ba22426f4ff00de35f061d24
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d004cacdedad45dcde5fd8bc209d48d2cfba6abe72e282562eb450ebb1a839c
+size 50382
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9831ceb2c27bef2ddb914b715cf0606a13ddce22
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a13523ef5785d958916e26c4c00215e28cb374028eddbb75325f596b43fde12a
+size 50387
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..222633cff944f4c1a9abf4576a4e1981828996fd
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a32eaf41c4270fe1c66dca8519c2e3f33c2891bd9476b396bd6f09c919ce8d2
+size 50387
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_128/code_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x4/l0_128/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..181bd2174cb3e3ac56bd6283d69ec3e19bf93ce9
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_128/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 128.0,
+   "recon_r2": 0.9355496733024188,
+   "l1_per_token": 5.431357130433026
+  },
+  "seed1": {
+   "l0": 128.0,
+   "recon_r2": 0.9355322484470104,
+   "l1_per_token": 5.444456709538326
+  },
+  "seed2": {
+   "l0": 128.0,
+   "recon_r2": 0.9355475010813245,
+   "l1_per_token": 5.427247523817316
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.4689063167324012,
+   "mcc_abs": 0.47577241424059435,
+   "act_iou": 0.2632600057287496,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.959797770989561,
+   "matched_mcc_signed": {
+    "0.6": 0.8374511706930756,
+    "0.7": 0.8591525144399056,
+    "0.8": 0.8834345343211321,
+    "0.9": 0.9154739859254306
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8408480702952644,
+    "0.7": 0.8599389017333158,
+    "0.8": 0.8823589319583227,
+    "0.9": 0.9152752773718864
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.4753887027631684,
+   "mcc_abs": 0.4815590415758871,
+   "act_iou": 0.2665553072761587,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9600181992221298,
+   "matched_mcc_signed": {
+    "0.6": 0.8366004253831654,
+    "0.7": 0.8578250300098375,
+    "0.8": 0.8792930179190083,
+    "0.9": 0.9139191550063906
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8403661400234352,
+    "0.7": 0.8589248453531247,
+    "0.8": 0.8805149473768648,
+    "0.9": 0.9138207681765547
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.4764583193711205,
+   "mcc_abs": 0.4836027970389276,
+   "act_iou": 0.26596651273608646,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9594343232496322,
+   "matched_mcc_signed": {
+    "0.6": 0.8387584564538226,
+    "0.7": 0.8570460140721418,
+    "0.8": 0.8804548659155181,
+    "0.9": 0.9144575615291486
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8394904036616863,
+    "0.7": 0.8575065701383571,
+    "0.8": 0.8788541381555114,
+    "0.9": 0.9134829280209882
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 128.0,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9355431409435845,
+  "recon_r2_std": 9.495498793299303e-06,
+  "l1_per_token_mean": 5.434353787929556,
+  "l1_per_token_std": 0.0089874348625556,
+  "input_r2_mean": 0.9597500978204411,
+  "input_r2_std": 0.00029484290072083464,
+  "mcc_signed_mean": 0.47358444628889673,
+  "mcc_signed_std": 0.004086525667669223,
+  "mcc_abs_mean": 0.4803114176184697,
+  "mcc_abs_std": 0.004061545007315533,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8376033508433546,
+   "0.7": 0.8580078528406282,
+   "0.8": 0.8810608060518862,
+   "0.9": 0.9146169008203233
+  },
+  "matched_mcc_signed_band": 0.872822227639048,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.8402348713267953,
+   "0.7": 0.8587901057415991,
+   "0.8": 0.8805760058302329,
+   "0.9": 0.9141929911898098
+  },
+  "matched_mcc_abs_band": 0.8734484935221092,
+  "act_iou_mean": 0.26526060858033157,
+  "act_iou_std": 0.00175740677702212,
+  "act_iou_n_pairs": 3072.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_128/dict_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x4/l0_128/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..763945970daa6f904432c205d8d40be4579c31b8
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_128/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 128.0,
+   "recon_r2": 0.9354827466104612,
+   "l1_per_token": 5.432239386713446,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 278,
+   "dense_rate": 0.09049479166666667,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.1379949301481247,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.1493125706911087
+  },
+  "seed1": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 128.0,
+   "recon_r2": 0.9354652294663645,
+   "l1_per_token": 5.445538012804615,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 275,
+   "dense_rate": 0.08951822916666667,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.13749313354492188,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.14824135601520538
+  },
+  "seed2": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 128.0,
+   "recon_r2": 0.9354828049669269,
+   "l1_per_token": 5.428469289761648,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 267,
+   "dense_rate": 0.0869140625,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.13940806686878204,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.15003608167171478
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.48847532272338867,
+   "dcs_abs": 0.7799339890480042,
+   "frac_sign_flipped": 0.4957682291666667,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.48847532272338867,
+    "0.6": 0.48847532272338867,
+    "0.7": 0.48847532272338867,
+    "0.8": 0.48847532272338867,
+    "0.9": 0.48847532272338867
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.48847532272338867,
+    "0.6": 0.48847532272338867,
+    "0.7": 0.48847532272338867,
+    "0.8": 0.48847532272338867,
+    "0.9": 0.48847532272338867
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4593098958333333,
+    "0.6": 0.4212239583333333,
+    "0.7": 0.3798828125,
+    "0.8": 0.3206380208333333,
+    "0.9": 0.20638020833333334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7799339890480042,
+    "0.6": 0.7799339890480042,
+    "0.7": 0.7799339890480042,
+    "0.8": 0.7799339890480042,
+    "0.9": 0.7799339890480042
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7799339890480042,
+    "0.6": 0.7799339890480042,
+    "0.7": 0.7799339890480042,
+    "0.8": 0.7799339890480042,
+    "0.9": 0.7799339890480042
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8753255208333334,
+    "0.6": 0.8284505208333334,
+    "0.7": 0.7578125,
+    "0.8": 0.6432291666666666,
+    "0.9": 0.4052734375
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.4960308074951172,
+   "dcs_abs": 0.7824799418449402,
+   "frac_sign_flipped": 0.4931640625,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.4960308372974396,
+    "0.6": 0.4960308372974396,
+    "0.7": 0.4960308372974396,
+    "0.8": 0.4960308372974396,
+    "0.9": 0.4960308372974396
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.4960308372974396,
+    "0.6": 0.4960308372974396,
+    "0.7": 0.4960308372974396,
+    "0.8": 0.4960308372974396,
+    "0.9": 0.4960308372974396
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4684244791666667,
+    "0.6": 0.4339192708333333,
+    "0.7": 0.3909505208333333,
+    "0.8": 0.3323567708333333,
+    "0.9": 0.2119140625
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7824799418449402,
+    "0.6": 0.7824799418449402,
+    "0.7": 0.7824799418449402,
+    "0.8": 0.7824799418449402,
+    "0.9": 0.7824799418449402
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7824799418449402,
+    "0.6": 0.7824799418449402,
+    "0.7": 0.7824799418449402,
+    "0.8": 0.7824799418449402,
+    "0.9": 0.7824799418449402
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8753255208333334,
+    "0.6": 0.8336588541666666,
+    "0.7": 0.7600911458333334,
+    "0.8": 0.6507161458333334,
+    "0.9": 0.4212239583333333
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.4956958293914795,
+   "dcs_abs": 0.7783279418945312,
+   "frac_sign_flipped": 0.4918619791666667,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.4956958293914795,
+    "0.6": 0.4956958293914795,
+    "0.7": 0.4956958293914795,
+    "0.8": 0.4956958293914795,
+    "0.9": 0.4956958293914795
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.4956958293914795,
+    "0.6": 0.4956958293914795,
+    "0.7": 0.4956958293914795,
+    "0.8": 0.4956958293914795,
+    "0.9": 0.4956958293914795
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4703776041666667,
+    "0.6": 0.4345703125,
+    "0.7": 0.39453125,
+    "0.8": 0.3326822916666667,
+    "0.9": 0.20149739583333334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.778328001499176,
+    "0.6": 0.778328001499176,
+    "0.7": 0.778328001499176,
+    "0.8": 0.778328001499176,
+    "0.9": 0.778328001499176
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.778328001499176,
+    "0.6": 0.778328001499176,
+    "0.7": 0.778328001499176,
+    "0.8": 0.778328001499176,
+    "0.9": 0.778328001499176
+   },
+   "matched_frac_abs": {
+    "0.5": 0.875,
+    "0.6": 0.8304036458333334,
+    "0.7": 0.7607421875,
+    "0.8": 0.6477864583333334,
+    "0.9": 0.3955078125
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 273,
+  "n_dense_std": 5.686240703077327,
+  "l0_mean": 128,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9354769270145842,
+  "recon_r2_std": null,
+  "l1_per_token_mean": 5.435415563093237,
+  "l1_per_token_std": 0.008966682528889923,
+  "babel_mu1_signed_mean": 0.1382987101872762,
+  "babel_mu1_signed_std": 0.0009929522359346937,
+  "n_dup_signed_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_signed_band": 0.0,
+  "n_dup_signed_band_std": 0.0,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_signed_band": 0.0,
+  "n_dup_clusters_signed_band_std": 0.0,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_signed_band": 1.0,
+  "largest_dup_signed_band_std": 0.0,
+  "dcs_signed_mean": 0.49340065320332843,
+  "dcs_signed_std": 0.004268748392087197,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.4934006631374359,
+   "0.6": 0.4934006631374359,
+   "0.7": 0.4934006631374359,
+   "0.8": 0.4934006631374359,
+   "0.9": 0.4934006631374359
+  },
+  "dedup_dcs_signed_band": 0.4934006631374359,
+  "dedup_dcs_signed_band_std": 0.00426875757334019,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.4934006631374359,
+   "0.6": 0.4934006631374359,
+   "0.7": 0.4934006631374359,
+   "0.8": 0.4934006631374359,
+   "0.9": 0.4934006631374359
+  },
+  "matched_cos_signed_band": 0.4934006631374359,
+  "matched_cos_signed_band_std": 0.00426875757334019,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_signed_band": 1.0,
+  "unique_frac_signed_band_std": 0.0,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.4660373263888889,
+   "0.6": 0.42990451388888884,
+   "0.7": 0.3884548611111111,
+   "0.8": 0.32855902777777773,
+   "0.9": 0.20659722222222224
+  },
+  "matched_frac_signed_band": 0.33837890625,
+  "matched_frac_signed_band_std": 0.005545808931445612,
+  "babel_mu1_abs_mean": 0.14919666945934296,
+  "babel_mu1_abs_std": 0.0009029589509921737,
+  "n_dup_abs_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_abs_band": 0.0,
+  "n_dup_abs_band_std": 0.0,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_abs_band": 0.0,
+  "n_dup_clusters_abs_band_std": 0.0,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_abs_band": 1.0,
+  "largest_dup_abs_band_std": 0.0,
+  "dcs_abs_mean": 0.7802472909291586,
+  "dcs_abs_std": 0.0020936557617240005,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.7802473107973734,
+   "0.6": 0.7802473107973734,
+   "0.7": 0.7802473107973734,
+   "0.8": 0.7802473107973734,
+   "0.9": 0.7802473107973734
+  },
+  "dedup_dcs_abs_band": 0.7802473107973734,
+  "dedup_dcs_abs_band_std": 0.002093628440690213,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.7802473107973734,
+   "0.6": 0.7802473107973734,
+   "0.7": 0.7802473107973734,
+   "0.8": 0.7802473107973734,
+   "0.9": 0.7802473107973734
+  },
+  "matched_cos_abs_band": 0.7802473107973734,
+  "matched_cos_abs_band_std": 0.002093628440690213,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_abs_band": 1.0,
+  "unique_frac_abs_band_std": 0.0,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.875217013888889,
+   "0.6": 0.8308376736111112,
+   "0.7": 0.7595486111111112,
+   "0.8": 0.6472439236111112,
+   "0.9": 0.4073350694444444
+  },
+  "matched_frac_abs_band": 0.6612413194444444,
+  "matched_frac_abs_band_std": 0.004487241027279635,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_128/seed0/config.json b/lm/gpt2-small_layer8/isae/x4/l0_128/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..58683470b63be99124282c520391d3e3a47e767a
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_128/seed0/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..9fd22094b3fb1464cb1a396b53637ad48ac8c3bb
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 128.0,
+  "l0_per_token_0.001": 128.0,
+  "l0_per_token_0.01": 127.83978579101563,
+  "l0_per_token_0.05": 26.177788330078126,
+  "l0_per_token_0.1": 8.010144140625,
+  "l0_per_token_0.5": 0.001235009765625,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 5.432239386713446,
+  "r2_score": 0.9354827466104612,
+  "l2_loss": 1.513086616360139
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_128/seed1/config.json b/lm/gpt2-small_layer8/isae/x4/l0_128/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..aa6d57cd26b5f2599a20f38d50b6f722e672d2fa
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_128/seed1/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..d3648c999f313770686145c6f9441d6e150e614f
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 128.0,
+  "l0_per_token_0.001": 128.0,
+  "l0_per_token_0.01": 127.83923134765625,
+  "l0_per_token_0.05": 26.180233349609374,
+  "l0_per_token_0.1": 8.00692744140625,
+  "l0_per_token_0.5": 0.003945703125,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 5.445538012804615,
+  "r2_score": 0.9354652294663645,
+  "l2_loss": 1.513497436023054
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_128/seed2/config.json b/lm/gpt2-small_layer8/isae/x4/l0_128/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..2bd41291b9360c7eb800a0914203b4894f6d83f4
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_128/seed2/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..0e124b09df15f684c0e49607ac71390eb026bfdb
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 128.0,
+  "l0_per_token_0.001": 128.0,
+  "l0_per_token_0.01": 127.8473763671875,
+  "l0_per_token_0.05": 26.12139033203125,
+  "l0_per_token_0.1": 8.014390869140625,
+  "l0_per_token_0.5": 0.0022693359375,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 5.428469289761648,
+  "r2_score": 0.9354828049669269,
+  "l2_loss": 1.5130852477589882
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_32/code_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x4/l0_32/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..7cc8d05e8e39d93d8dc3e65d53913c74a16dbbc0
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_32/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 32.0,
+   "recon_r2": 0.8875370040776929,
+   "l1_per_token": 2.8079154671711506
+  },
+  "seed1": {
+   "l0": 32.0,
+   "recon_r2": 0.8874652144328223,
+   "l1_per_token": 2.7736678656223135
+  },
+  "seed2": {
+   "l0": 32.0,
+   "recon_r2": 0.88756996275891,
+   "l1_per_token": 2.7712378482291515
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.464715559526163,
+   "mcc_abs": 0.46703811904523573,
+   "act_iou": 0.3002170007913603,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9631761748875203,
+   "matched_mcc_signed": {
+    "0.6": 0.8708684022004921,
+    "0.7": 0.8884056935848356,
+    "0.8": 0.9051978325363762,
+    "0.9": 0.9240902381927091
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8749735088278999,
+    "0.7": 0.8873812702973838,
+    "0.8": 0.9044830061319694,
+    "0.9": 0.9243264663627367
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.475302272133674,
+   "mcc_abs": 0.4778130370153442,
+   "act_iou": 0.3074976937938832,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9624183789970823,
+   "matched_mcc_signed": {
+    "0.6": 0.8684291724446661,
+    "0.7": 0.8840612941085491,
+    "0.8": 0.9024237405585073,
+    "0.9": 0.9244572316021649
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8734143004608453,
+    "0.7": 0.8863969693108041,
+    "0.8": 0.902838114004868,
+    "0.9": 0.9241634221482655
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.4633499602183373,
+   "mcc_abs": 0.4665355200867079,
+   "act_iou": 0.29822993683249194,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9625062140480241,
+   "matched_mcc_signed": {
+    "0.6": 0.8713021541070228,
+    "0.7": 0.8842870140054235,
+    "0.8": 0.9017211099874671,
+    "0.9": 0.9236053099355767
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8742253168335271,
+    "0.7": 0.8861280484507187,
+    "0.8": 0.9032780007778443,
+    "0.9": 0.9245713984784981
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 32.0,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.8875240604231417,
+  "recon_r2_std": 5.3560308059190186e-05,
+  "l1_per_token_mean": 2.784273727007539,
+  "l1_per_token_std": 0.020510367003190796,
+  "input_r2_mean": 0.9627002559775422,
+  "input_r2_std": 0.0004144910803618411,
+  "mcc_signed_mean": 0.46778926395939147,
+  "mcc_signed_std": 0.00654218497426593,
+  "mcc_abs_mean": 0.47046222538242927,
+  "mcc_abs_std": 0.006370947745447806,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8701999095840603,
+   "0.7": 0.885584667232936,
+   "0.8": 0.9031142276941169,
+   "0.9": 0.9240509265768169
+  },
+  "matched_mcc_signed_band": 0.8957374327719826,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.8742043753740908,
+   "0.7": 0.8866354293529689,
+   "0.8": 0.9035330403048939,
+   "0.9": 0.9243537623298335
+  },
+  "matched_mcc_abs_band": 0.8971816518404467,
+  "act_iou_mean": 0.30198154380591186,
+  "act_iou_std": 0.00487934819535416,
+  "act_iou_n_pairs": 3072.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_32/dict_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x4/l0_32/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..7c872a734d6a5c776f6524fff9c967b56b14e83f
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_32/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 32.0,
+   "recon_r2": 0.8874386198342771,
+   "l1_per_token": 2.808196093640639,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 12,
+   "dense_rate": 0.00390625,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.1835392713546753,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.19814515113830566
+  },
+  "seed1": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 32.0,
+   "recon_r2": 0.8873449927951378,
+   "l1_per_token": 2.773869812895541,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 12,
+   "dense_rate": 0.00390625,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.18215899169445038,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.19700586795806885
+  },
+  "seed2": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 32.0,
+   "recon_r2": 0.887441285834079,
+   "l1_per_token": 2.7713824994471796,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 11,
+   "dense_rate": 0.0035807291666666665,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.18303143978118896,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.19809861481189728
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.5319955945014954,
+   "dcs_abs": 0.8495085835456848,
+   "frac_sign_flipped": 0.4993489583333333,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.5319956541061401,
+    "0.6": 0.5319956541061401,
+    "0.7": 0.5319956541061401,
+    "0.8": 0.5319956541061401,
+    "0.9": 0.5319956541061401
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5319956541061401,
+    "0.6": 0.5319956541061401,
+    "0.7": 0.5319956541061401,
+    "0.8": 0.5319956541061401,
+    "0.9": 0.5319956541061401
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4781901041666667,
+    "0.6": 0.4518229166666667,
+    "0.7": 0.4244791666666667,
+    "0.8": 0.3883463541666667,
+    "0.9": 0.32421875
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8495085835456848,
+    "0.6": 0.8495085835456848,
+    "0.7": 0.8495085835456848,
+    "0.8": 0.8495085835456848,
+    "0.9": 0.8495085835456848
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8495085835456848,
+    "0.6": 0.8495085835456848,
+    "0.7": 0.8495085835456848,
+    "0.8": 0.8495085835456848,
+    "0.9": 0.8495085835456848
+   },
+   "matched_frac_abs": {
+    "0.5": 0.9078776041666666,
+    "0.6": 0.8850911458333334,
+    "0.7": 0.8512369791666666,
+    "0.8": 0.7799479166666666,
+    "0.9": 0.6471354166666666
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.5431373715400696,
+   "dcs_abs": 0.8474366068840027,
+   "frac_sign_flipped": 0.478515625,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.5431373715400696,
+    "0.6": 0.5431373715400696,
+    "0.7": 0.5431373715400696,
+    "0.8": 0.5431373715400696,
+    "0.9": 0.5431373715400696
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5431373715400696,
+    "0.6": 0.5431373715400696,
+    "0.7": 0.5431373715400696,
+    "0.8": 0.5431373715400696,
+    "0.9": 0.5431373715400696
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4925130208333333,
+    "0.6": 0.470703125,
+    "0.7": 0.4427083333333333,
+    "0.8": 0.4020182291666667,
+    "0.9": 0.3326822916666667
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8474366068840027,
+    "0.6": 0.8474366068840027,
+    "0.7": 0.8474366068840027,
+    "0.8": 0.8474366068840027,
+    "0.9": 0.8474366068840027
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8474366068840027,
+    "0.6": 0.8474366068840027,
+    "0.7": 0.8474366068840027,
+    "0.8": 0.8474366068840027,
+    "0.9": 0.8474366068840027
+   },
+   "matched_frac_abs": {
+    "0.5": 0.9095052083333334,
+    "0.6": 0.8850911458333334,
+    "0.7": 0.8453776041666666,
+    "0.8": 0.77734375,
+    "0.9": 0.6409505208333334
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.5317707657814026,
+   "dcs_abs": 0.8464012145996094,
+   "frac_sign_flipped": 0.4983723958333333,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.5317707657814026,
+    "0.6": 0.5317707657814026,
+    "0.7": 0.5317707657814026,
+    "0.8": 0.5317707657814026,
+    "0.9": 0.5317707657814026
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5317707657814026,
+    "0.6": 0.5317707657814026,
+    "0.7": 0.5317707657814026,
+    "0.8": 0.5317707657814026,
+    "0.9": 0.5317707657814026
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4791666666666667,
+    "0.6": 0.4514973958333333,
+    "0.7": 0.4280598958333333,
+    "0.8": 0.39453125,
+    "0.9": 0.3193359375
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8464012145996094,
+    "0.6": 0.8464012145996094,
+    "0.7": 0.8464012145996094,
+    "0.8": 0.8464012145996094,
+    "0.9": 0.8464012145996094
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8464012145996094,
+    "0.6": 0.8464012145996094,
+    "0.7": 0.8464012145996094,
+    "0.8": 0.8464012145996094,
+    "0.9": 0.8464012145996094
+   },
+   "matched_frac_abs": {
+    "0.5": 0.9085286458333334,
+    "0.6": 0.8818359375,
+    "0.7": 0.8440755208333334,
+    "0.8": 0.7776692708333334,
+    "0.9": 0.6389973958333334
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 12,
+  "n_dense_std": 0.5773502691896257,
+  "l0_mean": 32,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.8874082994878313,
+  "recon_r2_std": null,
+  "l1_per_token_mean": 2.784482801994453,
+  "l1_per_token_std": 0.02057393572600456,
+  "babel_mu1_signed_mean": 0.1829099009434382,
+  "babel_mu1_signed_std": 0.0006981201557891049,
+  "n_dup_signed_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_signed_band": 0.0,
+  "n_dup_signed_band_std": 0.0,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_signed_band": 0.0,
+  "n_dup_clusters_signed_band_std": 0.0,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_signed_band": 1.0,
+  "largest_dup_signed_band_std": 0.0,
+  "dcs_signed_mean": 0.5356345772743225,
+  "dcs_signed_std": 0.006498582794266761,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.5356345971425375,
+   "0.6": 0.5356345971425375,
+   "0.7": 0.5356345971425375,
+   "0.8": 0.5356345971425375,
+   "0.9": 0.5356345971425375
+  },
+  "dedup_dcs_signed_band": 0.5356345971425375,
+  "dedup_dcs_signed_band_std": 0.006498566106061298,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.5356345971425375,
+   "0.6": 0.5356345971425375,
+   "0.7": 0.5356345971425375,
+   "0.8": 0.5356345971425375,
+   "0.9": 0.5356345971425375
+  },
+  "matched_cos_signed_band": 0.5356345971425375,
+  "matched_cos_signed_band_std": 0.006498566106061298,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_signed_band": 1.0,
+  "unique_frac_signed_band_std": 0.0,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.4832899305555556,
+   "0.6": 0.4580078125,
+   "0.7": 0.4317491319444444,
+   "0.8": 0.39496527777777785,
+   "0.9": 0.3254123263888889
+  },
+  "matched_frac_signed_band": 0.4025336371527778,
+  "matched_frac_signed_band_std": 0.008242064935952437,
+  "babel_mu1_abs_mean": 0.19774987796942392,
+  "babel_mu1_abs_std": 0.0006447515646870034,
+  "n_dup_abs_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_abs_band": 0.0,
+  "n_dup_abs_band_std": 0.0,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_abs_band": 0.0,
+  "n_dup_clusters_abs_band_std": 0.0,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_abs_band": 1.0,
+  "largest_dup_abs_band_std": 0.0,
+  "dcs_abs_mean": 0.8477821350097656,
+  "dcs_abs_std": 0.0015822381950984605,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.8477821350097656,
+   "0.6": 0.8477821350097656,
+   "0.7": 0.8477821350097656,
+   "0.8": 0.8477821350097656,
+   "0.9": 0.8477821350097656
+  },
+  "dedup_dcs_abs_band": 0.8477821350097656,
+  "dedup_dcs_abs_band_std": 0.0015822381950984605,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.8477821350097656,
+   "0.6": 0.8477821350097656,
+   "0.7": 0.8477821350097656,
+   "0.8": 0.8477821350097656,
+   "0.9": 0.8477821350097656
+  },
+  "matched_cos_abs_band": 0.8477821350097656,
+  "matched_cos_abs_band_std": 0.0015822381950984605,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_abs_band": 1.0,
+  "unique_frac_abs_band_std": 0.0,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.9086371527777778,
+   "0.6": 0.884006076388889,
+   "0.7": 0.8468967013888888,
+   "0.8": 0.7783203125,
+   "0.9": 0.6423611111111112
+  },
+  "matched_frac_abs_band": 0.7878960503472222,
+  "matched_frac_abs_band_std": 0.0026748392386239787,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_32/seed0/config.json b/lm/gpt2-small_layer8/isae/x4/l0_32/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7698993f019b0b847dd5844e172c86abe89b8c5d
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_32/seed0/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..53e68a41376be7a69195e288bfb5671650e929b8
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 32.0,
+  "l0_per_token_0.001": 32.0,
+  "l0_per_token_0.01": 31.98144501953125,
+  "l0_per_token_0.05": 25.1997927734375,
+  "l0_per_token_0.1": 7.73191875,
+  "l0_per_token_0.5": 0.0067970703125,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 2.808196093640639,
+  "r2_score": 0.8874386198342771,
+  "l2_loss": 2.6398383207583485
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_32/seed1/config.json b/lm/gpt2-small_layer8/isae/x4/l0_32/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..8dd350b779af726863c7ffbafe9691b8d48a8977
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_32/seed1/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..451176aa4c78e5687767623c7f1208dfd2630e62
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 32.0,
+  "l0_per_token_0.001": 32.0,
+  "l0_per_token_0.01": 31.982439990234376,
+  "l0_per_token_0.05": 25.18191796875,
+  "l0_per_token_0.1": 7.677237744140625,
+  "l0_per_token_0.5": 0.0073236328125,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 2.773869812895541,
+  "r2_score": 0.8873449927951378,
+  "l2_loss": 2.642034102698966
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_32/seed2/config.json b/lm/gpt2-small_layer8/isae/x4/l0_32/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..c365ecc852813ac60a03d7d03f4fd43a7ef83f74
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_32/seed2/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..3149e1739ee34680458da5f4a97261abe981be43
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 32.0,
+  "l0_per_token_0.001": 32.0,
+  "l0_per_token_0.01": 31.983414404296877,
+  "l0_per_token_0.05": 25.161342431640627,
+  "l0_per_token_0.1": 7.68944931640625,
+  "l0_per_token_0.5": 0.006118603515625,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 2.7713824994471796,
+  "r2_score": 0.887441285834079,
+  "l2_loss": 2.6397757965743884
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_64/code_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x4/l0_64/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..a632b24ec52b69a8745a7a19f47a64a203ebb523
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_64/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 64.0,
+   "recon_r2": 0.9108616901343901,
+   "l1_per_token": 3.8337769074773695
+  },
+  "seed1": {
+   "l0": 64.0,
+   "recon_r2": 0.9108816176486899,
+   "l1_per_token": 3.8666136246139953
+  },
+  "seed2": {
+   "l0": 64.0,
+   "recon_r2": 0.9109332184901348,
+   "l1_per_token": 3.826730621349334
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.4742599955202273,
+   "mcc_abs": 0.48143606067247524,
+   "act_iou": 0.2922721278744978,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9630603877310121,
+   "matched_mcc_signed": {
+    "0.6": 0.8535681486295492,
+    "0.7": 0.874268387703731,
+    "0.8": 0.8940728737778997,
+    "0.9": 0.9196479017481659
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8601421983808936,
+    "0.7": 0.8762891720431838,
+    "0.8": 0.8945456353239528,
+    "0.9": 0.9201494670315562
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.48291037983405705,
+   "mcc_abs": 0.4875990548901559,
+   "act_iou": 0.29784525239666454,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9634618582792625,
+   "matched_mcc_signed": {
+    "0.6": 0.8581090579594017,
+    "0.7": 0.8780839981089704,
+    "0.8": 0.896521012952878,
+    "0.9": 0.922528363038735
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8613067210969036,
+    "0.7": 0.8782151936706958,
+    "0.8": 0.8953196731294842,
+    "0.9": 0.9218530450083606
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.47866420274004273,
+   "mcc_abs": 0.48237277783618177,
+   "act_iou": 0.2922316279382962,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9640425369389779,
+   "matched_mcc_signed": {
+    "0.6": 0.8595982705361255,
+    "0.7": 0.874970831613146,
+    "0.8": 0.8961525262583623,
+    "0.9": 0.9200397854006559
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8624948090270002,
+    "0.7": 0.8762316314801522,
+    "0.8": 0.8967160513348938,
+    "0.9": 0.9202798429536153
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 64.0,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9108921754244049,
+  "recon_r2_std": 3.691444689981787e-05,
+  "l1_per_token_mean": 3.8423737178135666,
+  "l1_per_token_std": 0.02128596599451495,
+  "input_r2_mean": 0.9635215943164175,
+  "input_r2_std": 0.0004937920233035384,
+  "mcc_signed_mean": 0.47861152603144236,
+  "mcc_signed_std": 0.004325432732219943,
+  "mcc_abs_mean": 0.4838026311329376,
+  "mcc_abs_std": 0.0033209915340864883,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8570918257083587,
+   "0.7": 0.8757744058086159,
+   "0.8": 0.8955821376630467,
+   "0.9": 0.9207386833958523
+  },
+  "matched_mcc_signed_band": 0.8872967631439684,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.8613145761682658,
+   "0.7": 0.8769119990646773,
+   "0.8": 0.8955271199294437,
+   "0.9": 0.9207607849978441
+  },
+  "matched_mcc_abs_band": 0.8886286200400577,
+  "act_iou_mean": 0.2941163360698195,
+  "act_iou_std": 0.0032293997570239912,
+  "act_iou_n_pairs": 3072.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_64/dict_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x4/l0_64/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..7c27e79e36e992037bd796fc388116a2c5405f38
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_64/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 64.0,
+   "recon_r2": 0.9107767216876593,
+   "l1_per_token": 3.83468987757705,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 55,
+   "dense_rate": 0.017903645833333332,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.15144942700862885,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.1619657427072525
+  },
+  "seed1": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 64.0,
+   "recon_r2": 0.9107735178029157,
+   "l1_per_token": 3.8675073870753316,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 52,
+   "dense_rate": 0.016927083333333332,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.14922931790351868,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.16083137691020966
+  },
+  "seed2": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 64.0,
+   "recon_r2": 0.9108304181709048,
+   "l1_per_token": 3.827908278020669,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 52,
+   "dense_rate": 0.016927083333333332,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.15053828060626984,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.16286377608776093
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.5160282254219055,
+   "dcs_abs": 0.829708993434906,
+   "frac_sign_flipped": 0.494140625,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.5160282254219055,
+    "0.6": 0.5160282254219055,
+    "0.7": 0.5160282254219055,
+    "0.8": 0.5160282254219055,
+    "0.9": 0.5160282254219055
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5160282254219055,
+    "0.6": 0.5160282254219055,
+    "0.7": 0.5160282254219055,
+    "0.8": 0.5160282254219055,
+    "0.9": 0.5160282254219055
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4759114583333333,
+    "0.6": 0.4527994791666667,
+    "0.7": 0.4169921875,
+    "0.8": 0.3720703125,
+    "0.9": 0.2887369791666667
+   },
+   "matched_cos_abs": {
+    "0.5": 0.829708993434906,
+    "0.6": 0.829708993434906,
+    "0.7": 0.829708993434906,
+    "0.8": 0.829708993434906,
+    "0.9": 0.829708993434906
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.829708993434906,
+    "0.6": 0.829708993434906,
+    "0.7": 0.829708993434906,
+    "0.8": 0.829708993434906,
+    "0.9": 0.829708993434906
+   },
+   "matched_frac_abs": {
+    "0.5": 0.90625,
+    "0.6": 0.8779296875,
+    "0.7": 0.8245442708333334,
+    "0.8": 0.7434895833333334,
+    "0.9": 0.5758463541666666
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.5227801203727722,
+   "dcs_abs": 0.8288562893867493,
+   "frac_sign_flipped": 0.4905598958333333,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.5227801203727722,
+    "0.6": 0.5227801203727722,
+    "0.7": 0.5227801203727722,
+    "0.8": 0.5227801203727722,
+    "0.9": 0.5227801203727722
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5227801203727722,
+    "0.6": 0.5227801203727722,
+    "0.7": 0.5227801203727722,
+    "0.8": 0.5227801203727722,
+    "0.9": 0.5227801203727722
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4791666666666667,
+    "0.6": 0.4527994791666667,
+    "0.7": 0.4228515625,
+    "0.8": 0.3831380208333333,
+    "0.9": 0.2962239583333333
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8288562893867493,
+    "0.6": 0.8288562893867493,
+    "0.7": 0.8288562893867493,
+    "0.8": 0.8288562893867493,
+    "0.9": 0.8288562893867493
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8288562893867493,
+    "0.6": 0.8288562893867493,
+    "0.7": 0.8288562893867493,
+    "0.8": 0.8288562893867493,
+    "0.9": 0.8288562893867493
+   },
+   "matched_frac_abs": {
+    "0.5": 0.9020182291666666,
+    "0.6": 0.8756510416666666,
+    "0.7": 0.8255208333333334,
+    "0.8": 0.7447916666666666,
+    "0.9": 0.5699869791666666
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.5153018832206726,
+   "dcs_abs": 0.8296567797660828,
+   "frac_sign_flipped": 0.4977213541666667,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.5153018832206726,
+    "0.6": 0.5153018832206726,
+    "0.7": 0.5153018832206726,
+    "0.8": 0.5153018832206726,
+    "0.9": 0.5153018832206726
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5153018832206726,
+    "0.6": 0.5153018832206726,
+    "0.7": 0.5153018832206726,
+    "0.8": 0.5153018832206726,
+    "0.9": 0.5153018832206726
+   },
+   "matched_frac_signed": {
+    "0.5": 0.474609375,
+    "0.6": 0.4462890625,
+    "0.7": 0.416015625,
+    "0.8": 0.3697916666666667,
+    "0.9": 0.2926432291666667
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8296567797660828,
+    "0.6": 0.8296567797660828,
+    "0.7": 0.8296567797660828,
+    "0.8": 0.8296567797660828,
+    "0.9": 0.8296567797660828
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8296567797660828,
+    "0.6": 0.8296567797660828,
+    "0.7": 0.8296567797660828,
+    "0.8": 0.8296567797660828,
+    "0.9": 0.8296567797660828
+   },
+   "matched_frac_abs": {
+    "0.5": 0.9039713541666666,
+    "0.6": 0.875,
+    "0.7": 0.8284505208333334,
+    "0.8": 0.7415364583333334,
+    "0.9": 0.5833333333333334
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 53,
+  "n_dense_std": 1.7320508075688772,
+  "l0_mean": 64,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9107935525538267,
+  "recon_r2_std": null,
+  "l1_per_token_mean": 3.84336851422435,
+  "l1_per_token_std": 0.021178088184923977,
+  "babel_mu1_signed_mean": 0.1504056751728058,
+  "babel_mu1_signed_std": 0.0011159790591164379,
+  "n_dup_signed_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_signed_band": 0.0,
+  "n_dup_signed_band_std": 0.0,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_signed_band": 0.0,
+  "n_dup_clusters_signed_band_std": 0.0,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_signed_band": 1.0,
+  "largest_dup_signed_band_std": 0.0,
+  "dcs_signed_mean": 0.5180367430051168,
+  "dcs_signed_std": 0.004123907720513679,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.5180367430051168,
+   "0.6": 0.5180367430051168,
+   "0.7": 0.5180367430051168,
+   "0.8": 0.5180367430051168,
+   "0.9": 0.5180367430051168
+  },
+  "dedup_dcs_signed_band": 0.5180367430051168,
+  "dedup_dcs_signed_band_std": 0.004123907720513679,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.5180367430051168,
+   "0.6": 0.5180367430051168,
+   "0.7": 0.5180367430051168,
+   "0.8": 0.5180367430051168,
+   "0.9": 0.5180367430051168
+  },
+  "matched_cos_signed_band": 0.5180367430051168,
+  "matched_cos_signed_band_std": 0.004123907720513679,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_signed_band": 1.0,
+  "unique_frac_signed_band_std": 0.0,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.4765625,
+   "0.6": 0.45062934027777785,
+   "0.7": 0.4186197916666667,
+   "0.8": 0.375,
+   "0.9": 0.29253472222222227
+  },
+  "matched_frac_signed_band": 0.38419596354166674,
+  "matched_frac_signed_band_std": 0.00401411538253204,
+  "babel_mu1_abs_mean": 0.16188696523507437,
+  "babel_mu1_abs_std": 0.001018487124032431,
+  "n_dup_abs_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_abs_band": 0.0,
+  "n_dup_abs_band_std": 0.0,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_abs_band": 0.0,
+  "n_dup_clusters_abs_band_std": 0.0,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_abs_band": 1.0,
+  "largest_dup_abs_band_std": 0.0,
+  "dcs_abs_mean": 0.8294073541959127,
+  "dcs_abs_std": 0.00047794966756628895,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.8294073541959127,
+   "0.6": 0.8294073541959127,
+   "0.7": 0.8294073541959127,
+   "0.8": 0.8294073541959127,
+   "0.9": 0.8294073541959127
+  },
+  "dedup_dcs_abs_band": 0.8294073541959127,
+  "dedup_dcs_abs_band_std": 0.00047794966756628895,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.8294073541959127,
+   "0.6": 0.8294073541959127,
+   "0.7": 0.8294073541959127,
+   "0.8": 0.8294073541959127,
+   "0.9": 0.8294073541959127
+  },
+  "matched_cos_abs_band": 0.8294073541959127,
+  "matched_cos_abs_band_std": 0.00047794966756628895,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_abs_band": 1.0,
+  "unique_frac_abs_band_std": 0.0,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.904079861111111,
+   "0.6": 0.8761935763888888,
+   "0.7": 0.826171875,
+   "0.8": 0.7432725694444445,
+   "0.9": 0.5763888888888888
+  },
+  "matched_frac_abs_band": 0.7555067274305556,
+  "matched_frac_abs_band_std": 0.0015469376550981014,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_64/seed0/config.json b/lm/gpt2-small_layer8/isae/x4/l0_64/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..584f6af16d8082cd6a6b59a6ab852e3e73b8bac5
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_64/seed0/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..7bec11d6b2c4c4bc6e0ff386c03dc101264465a1
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.0,
+  "l0_per_token_0.001": 64.0,
+  "l0_per_token_0.01": 63.91413120117188,
+  "l0_per_token_0.05": 24.144826025390625,
+  "l0_per_token_0.1": 7.724731005859375,
+  "l0_per_token_0.5": 0.00370234375,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 3.83468987757705,
+  "r2_score": 0.9107767216876593,
+  "l2_loss": 2.0925030311979875
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_64/seed1/config.json b/lm/gpt2-small_layer8/isae/x4/l0_64/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..aa6ea96215736297f15273926f2155d9e50197ab
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_64/seed1/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..6c95b33eab615a76856f33e167326963d3132481
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.0,
+  "l0_per_token_0.001": 64.0,
+  "l0_per_token_0.01": 63.915991845703125,
+  "l0_per_token_0.05": 24.15197890625,
+  "l0_per_token_0.1": 7.733937939453125,
+  "l0_per_token_0.5": 0.003730322265625,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 3.8675073870753316,
+  "r2_score": 0.9107735178029157,
+  "l2_loss": 2.092578170092953
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_64/seed2/config.json b/lm/gpt2-small_layer8/isae/x4/l0_64/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..d896eb6ecc74aff8c480e9b1efebbc1e969adec3
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_64/seed2/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..44c5d2e57fc9f96d656e5cc9778cae83de005f58
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.0,
+  "l0_per_token_0.001": 64.0,
+  "l0_per_token_0.01": 63.91634897460938,
+  "l0_per_token_0.05": 24.052953759765625,
+  "l0_per_token_0.1": 7.691286865234375,
+  "l0_per_token_0.5": 0.0039111328125,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 3.827908278020669,
+  "r2_score": 0.9108304181709048,
+  "l2_loss": 2.0912437179774788
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_128/code_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x8/l0_128/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..93d507a6a50518f455ba9e2dc9f8d47f9c97d23b
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_128/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 128.0,
+   "recon_r2": 0.9408276570372041,
+   "l1_per_token": 5.290462849130255
+  },
+  "seed1": {
+   "l0": 128.0,
+   "recon_r2": 0.9408370207178236,
+   "l1_per_token": 5.3008274887035665
+  },
+  "seed2": {
+   "l0": 128.0,
+   "recon_r2": 0.9408645230954712,
+   "l1_per_token": 5.273586491955457
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.4590823325598114,
+   "mcc_abs": 0.46545024851458683,
+   "act_iou": 0.24023442971760356,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9647008559768026,
+   "matched_mcc_signed": {
+    "0.6": 0.8080772881798189,
+    "0.7": 0.8336381971097293,
+    "0.8": 0.8641042739605231,
+    "0.9": 0.9093470140606147
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8143425207618272,
+    "0.7": 0.837117047809041,
+    "0.8": 0.8652009897145425,
+    "0.9": 0.9096347693281328
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.46367191134518526,
+   "mcc_abs": 0.4709734976487929,
+   "act_iou": 0.247012408309398,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.965133606229361,
+   "matched_mcc_signed": {
+    "0.6": 0.8144688993251197,
+    "0.7": 0.8366313126248468,
+    "0.8": 0.8648888835438711,
+    "0.9": 0.9081970889074081
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8169154010997357,
+    "0.7": 0.8369340442247206,
+    "0.8": 0.8649955784796279,
+    "0.9": 0.9084479201753581
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.4548615377821463,
+   "mcc_abs": 0.4622666254019812,
+   "act_iou": 0.2424570759634932,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9664870435364208,
+   "matched_mcc_signed": {
+    "0.6": 0.8161578408397389,
+    "0.7": 0.8397025966364745,
+    "0.8": 0.8698957399942938,
+    "0.9": 0.9113361523672596
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8165571689688713,
+    "0.7": 0.8392185385234451,
+    "0.8": 0.8677045424522182,
+    "0.9": 0.910615312868582
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 128.0,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9408430669501663,
+  "recon_r2_std": 1.9162313461460514e-05,
+  "l1_per_token_mean": 5.288292276596426,
+  "l1_per_token_std": 0.013749600532418771,
+  "input_r2_mean": 0.9654405019141947,
+  "input_r2_std": 0.0009318021359499812,
+  "mcc_signed_mean": 0.45920526056238103,
+  "mcc_signed_std": 0.004406472971712851,
+  "mcc_abs_mean": 0.4662301238551203,
+  "mcc_abs_std": 0.004405514753199016,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8129013427815591,
+   "0.7": 0.8366573687903501,
+   "0.8": 0.8662962991662293,
+   "0.9": 0.9096267517784274
+  },
+  "matched_mcc_signed_band": 0.8563704406291415,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.8159383636101447,
+   "0.7": 0.8377565435190689,
+   "0.8": 0.8659670368821296,
+   "0.9": 0.9095660007906909
+  },
+  "matched_mcc_abs_band": 0.8573069862005086,
+  "act_iou_mean": 0.24323463799683157,
+  "act_iou_std": 0.0034552424639195615,
+  "act_iou_n_pairs": 6144.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_128/dict_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x8/l0_128/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..48e7ec174e1e765a1f2988a74ed8b59e65e62867
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_128/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 128.0,
+   "recon_r2": 0.9407592506788783,
+   "l1_per_token": 5.2918284352840885,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 219,
+   "dense_rate": 0.03564453125,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.15194782614707947,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.16206467151641846
+  },
+  "seed1": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 128.0,
+   "recon_r2": 0.940756346849262,
+   "l1_per_token": 5.301968848106155,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 223,
+   "dense_rate": 0.036295572916666664,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.15116338431835175,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.16105018556118011
+  },
+  "seed2": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 128.0,
+   "recon_r2": 0.9407873927865034,
+   "l1_per_token": 5.274746725079388,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 210,
+   "dense_rate": 0.0341796875,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 12,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 6,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 2,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.15357999503612518,
+   "n_dup_abs": {
+    "0.5": 11,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 5,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 3,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.16334520280361176
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.4834640920162201,
+   "dcs_abs": 0.75323885679245,
+   "frac_sign_flipped": 0.4977213541666667,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.4834640920162201,
+    "0.6": 0.4834640920162201,
+    "0.7": 0.4834640920162201,
+    "0.8": 0.4834640920162201,
+    "0.9": 0.4834640920162201
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.4834640920162201,
+    "0.6": 0.4834640920162201,
+    "0.7": 0.4834640920162201,
+    "0.8": 0.4834640920162201,
+    "0.9": 0.4834640920162201
+   },
+   "matched_frac_signed": {
+    "0.5": 0.45947265625,
+    "0.6": 0.4098307291666667,
+    "0.7": 0.3505859375,
+    "0.8": 0.2705078125,
+    "0.9": 0.14095052083333334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.75323885679245,
+    "0.6": 0.75323885679245,
+    "0.7": 0.75323885679245,
+    "0.8": 0.75323885679245,
+    "0.9": 0.75323885679245
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.75323885679245,
+    "0.6": 0.75323885679245,
+    "0.7": 0.75323885679245,
+    "0.8": 0.75323885679245,
+    "0.9": 0.75323885679245
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8644205729166666,
+    "0.6": 0.8050130208333334,
+    "0.7": 0.70703125,
+    "0.8": 0.5556640625,
+    "0.9": 0.2947591145833333
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.4891701936721802,
+   "dcs_abs": 0.7573849558830261,
+   "frac_sign_flipped": 0.4969075520833333,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.48936715722084045,
+    "0.6": 0.4891701936721802,
+    "0.7": 0.4891701936721802,
+    "0.8": 0.4891701936721802,
+    "0.9": 0.4891701936721802
+   },
+   "unique_frac_signed": {
+    "0.5": 0.99951171875,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.4891282084136037,
+    "0.6": 0.4891701936721802,
+    "0.7": 0.4891701936721802,
+    "0.8": 0.4891701936721802,
+    "0.9": 0.4891701936721802
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4611002604166667,
+    "0.6": 0.412109375,
+    "0.7": 0.3614908854166667,
+    "0.8": 0.2887369791666667,
+    "0.9": 0.15966796875
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7577136754989624,
+    "0.6": 0.7573849558830261,
+    "0.7": 0.7573849558830261,
+    "0.8": 0.7573849558830261,
+    "0.9": 0.7573849558830261
+   },
+   "unique_frac_abs": {
+    "0.5": 0.99951171875,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7573436981183477,
+    "0.6": 0.7573849558830261,
+    "0.7": 0.7573849558830261,
+    "0.8": 0.7573849558830261,
+    "0.9": 0.7573849558830261
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8680013020833334,
+    "0.6": 0.8064778645833334,
+    "0.7": 0.7164713541666666,
+    "0.8": 0.5725911458333334,
+    "0.9": 0.3115234375
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.48126277327537537,
+   "dcs_abs": 0.7586879730224609,
+   "frac_sign_flipped": 0.5138346354166666,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.4814332127571106,
+    "0.6": 0.48126277327537537,
+    "0.7": 0.48126277327537537,
+    "0.8": 0.48126277327537537,
+    "0.9": 0.48126277327537537
+   },
+   "unique_frac_signed": {
+    "0.5": 0.99951171875,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.48119813794619404,
+    "0.6": 0.48126277327537537,
+    "0.7": 0.48126277327537537,
+    "0.8": 0.48126277327537537,
+    "0.9": 0.48126277327537537
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4495442708333333,
+    "0.6": 0.4000651041666667,
+    "0.7": 0.3466796875,
+    "0.8": 0.27490234375,
+    "0.9": 0.15397135416666666
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7591173052787781,
+    "0.6": 0.7586879730224609,
+    "0.7": 0.7586879730224609,
+    "0.8": 0.7586879730224609,
+    "0.9": 0.7586879730224609
+   },
+   "unique_frac_abs": {
+    "0.5": 0.99951171875,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7587466425320599,
+    "0.6": 0.7586879730224609,
+    "0.7": 0.7586879730224609,
+    "0.8": 0.7586879730224609,
+    "0.9": 0.7586879730224609
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8663736979166666,
+    "0.6": 0.8087565104166666,
+    "0.7": 0.7112630208333334,
+    "0.8": 0.5685221354166666,
+    "0.9": 0.31640625
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 217,
+  "n_dense_std": 6.6583281184793925,
+  "l0_mean": 128,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9407676634382146,
+  "recon_r2_std": null,
+  "l1_per_token_mean": 5.2895146694898765,
+  "l1_per_token_std": 0.013757766080541866,
+  "babel_mu1_signed_mean": 0.15223040183385214,
+  "babel_mu1_signed_std": 0.0012328376228768509,
+  "n_dup_signed_per_threshold": {
+   "0.5": 4.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_signed_band": 0.0,
+  "n_dup_signed_band_std": 0.0,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 2.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_signed_band": 0.0,
+  "n_dup_clusters_signed_band_std": 0.0,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 1.3333333333333333,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_signed_band": 1.0,
+  "largest_dup_signed_band_std": 0.0,
+  "dcs_signed_mean": 0.48463235298792523,
+  "dcs_signed_std": 0.0040811088697139935,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.48459681279200595,
+   "0.6": 0.48463235298792523,
+   "0.7": 0.48463235298792523,
+   "0.8": 0.48463235298792523,
+   "0.9": 0.48463235298792523
+  },
+  "dedup_dcs_signed_band": 0.48463235298792523,
+  "dedup_dcs_signed_band_std": 0.0040811088697139935,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.4847548206647237,
+   "0.6": 0.48463235298792523,
+   "0.7": 0.48463235298792523,
+   "0.8": 0.48463235298792523,
+   "0.9": 0.48463235298792523
+  },
+  "matched_cos_signed_band": 0.48463235298792523,
+  "matched_cos_signed_band_std": 0.0040811088697139935,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.9996744791666666,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_signed_band": 1.0,
+  "unique_frac_signed_band_std": 0.0,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.4567057291666667,
+   "0.6": 0.4073350694444445,
+   "0.7": 0.3529188368055556,
+   "0.8": 0.2780490451388889,
+   "0.9": 0.15152994791666666
+  },
+  "matched_frac_signed_band": 0.2974582248263889,
+  "matched_frac_signed_band_std": 0.006981209286807236,
+  "babel_mu1_abs_mean": 0.16215335329373678,
+  "babel_mu1_abs_std": 0.0011500758144513277,
+  "n_dup_abs_per_threshold": {
+   "0.5": 3.6666666666666665,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_abs_band": 0.0,
+  "n_dup_abs_band_std": 0.0,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 1.6666666666666667,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_abs_band": 0.0,
+  "n_dup_clusters_abs_band_std": 0.0,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 1.6666666666666667,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_abs_band": 1.0,
+  "largest_dup_abs_band_std": 0.0,
+  "dcs_abs_mean": 0.7564372618993124,
+  "dcs_abs_std": 0.0028454893844675933,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.7564430658142859,
+   "0.6": 0.7564372618993124,
+   "0.7": 0.7564372618993124,
+   "0.8": 0.7564372618993124,
+   "0.9": 0.7564372618993124
+  },
+  "dedup_dcs_abs_band": 0.7564372618993124,
+  "dedup_dcs_abs_band_std": 0.0028454893844675933,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.7566899458567301,
+   "0.6": 0.7564372618993124,
+   "0.7": 0.7564372618993124,
+   "0.8": 0.7564372618993124,
+   "0.9": 0.7564372618993124
+  },
+  "matched_cos_abs_band": 0.7564372618993124,
+  "matched_cos_abs_band_std": 0.0028454893844675933,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.9996744791666666,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_abs_band": 1.0,
+  "unique_frac_abs_band_std": 0.0,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.8662651909722222,
+   "0.6": 0.8067491319444445,
+   "0.7": 0.7115885416666666,
+   "0.8": 0.5655924479166666,
+   "0.9": 0.30756293402777773
+  },
+  "matched_frac_abs_band": 0.5978732638888888,
+  "matched_frac_abs_band_std": 0.0062897916642721124,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_128/seed0/config.json b/lm/gpt2-small_layer8/isae/x8/l0_128/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..40df04c960386578c223597af1001d2e978efd94
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_128/seed0/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..5806f7b632ac4f4c265e66af4f378beaeb18dce9
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 128.0,
+  "l0_per_token_0.001": 128.0,
+  "l0_per_token_0.01": 127.82762529296875,
+  "l0_per_token_0.05": 25.20435791015625,
+  "l0_per_token_0.1": 7.753847412109375,
+  "l0_per_token_0.5": 0.0038908203125,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 5.2918284352840885,
+  "r2_score": 0.9407592506788783,
+  "l2_loss": 1.3893397538133472
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_128/seed1/config.json b/lm/gpt2-small_layer8/isae/x8/l0_128/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..2577533fca09be994e11f088889f54105f929fcf
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_128/seed1/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..b7c0ac5e772803266cd0117d14ca01bb89ef37e8
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 128.0,
+  "l0_per_token_0.001": 128.0,
+  "l0_per_token_0.01": 127.83055766601562,
+  "l0_per_token_0.05": 25.208123828125,
+  "l0_per_token_0.1": 7.652070166015625,
+  "l0_per_token_0.5": 0.0039263671875,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 5.301968848106155,
+  "r2_score": 0.940756346849262,
+  "l2_loss": 1.3894078556852931
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_128/seed2/config.json b/lm/gpt2-small_layer8/isae/x8/l0_128/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..a1db69203b94d5ad580031e9ccd7364464b046c4
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_128/seed2/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..eece43380d34cb5bd8be17d3a60a19af4e0b4703
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 128.0,
+  "l0_per_token_0.001": 128.0,
+  "l0_per_token_0.01": 127.82975166015625,
+  "l0_per_token_0.05": 25.199023681640625,
+  "l0_per_token_0.1": 7.7287826171875,
+  "l0_per_token_0.5": 0.001089208984375,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 5.274746725079388,
+  "r2_score": 0.9407873927865034,
+  "l2_loss": 1.3886797528962815
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_32/code_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x8/l0_32/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..6e5cc2d9d9d8680b481ee945155d7b720762b8c7
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_32/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 32.0,
+   "recon_r2": 0.8987467791167935,
+   "l1_per_token": 2.7483979971123844
+  },
+  "seed1": {
+   "l0": 32.0,
+   "recon_r2": 0.8988496644990407,
+   "l1_per_token": 2.748708808692522
+  },
+  "seed2": {
+   "l0": 32.0,
+   "recon_r2": 0.8989595762250762,
+   "l1_per_token": 2.7434077741462115
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.4437741101530879,
+   "mcc_abs": 0.4476767233526024,
+   "act_iou": 0.2693536878887275,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9633327383904989,
+   "matched_mcc_signed": {
+    "0.6": 0.8406822078874171,
+    "0.7": 0.8633917718332393,
+    "0.8": 0.8834862170023867,
+    "0.9": 0.9104208743282491
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8497203768782522,
+    "0.7": 0.8655182192304486,
+    "0.8": 0.8852571433383318,
+    "0.9": 0.9118939795039405
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.45117385832881624,
+   "mcc_abs": 0.45563310608932167,
+   "act_iou": 0.2773663262421639,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9633762132387402,
+   "matched_mcc_signed": {
+    "0.6": 0.8420470710724287,
+    "0.7": 0.8647901231803938,
+    "0.8": 0.886614729846971,
+    "0.9": 0.9113213679050564
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8496070591448321,
+    "0.7": 0.866759731814174,
+    "0.8": 0.8874008804842938,
+    "0.9": 0.911969696381172
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.4477362564731795,
+   "mcc_abs": 0.4516465626343151,
+   "act_iou": 0.2759875501994431,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9644652677694873,
+   "matched_mcc_signed": {
+    "0.6": 0.8448913327351015,
+    "0.7": 0.863807699357218,
+    "0.8": 0.8848975631698218,
+    "0.9": 0.911641796843879
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8514896080272842,
+    "0.7": 0.8669443111444717,
+    "0.8": 0.8867886077224357,
+    "0.9": 0.9133200576375257
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 32.0,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.8988520066136368,
+  "recon_r2_std": 0.00010641788594477893,
+  "l1_per_token_mean": 2.746838193317039,
+  "l1_per_token_std": 0.002974892052709484,
+  "input_r2_mean": 0.9637247397995754,
+  "input_r2_std": 0.000641684323693246,
+  "mcc_signed_mean": 0.4475614083183612,
+  "mcc_signed_std": 0.003702971397942647,
+  "mcc_abs_mean": 0.4516521306920797,
+  "mcc_abs_std": 0.003978194290848751,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8425402038983157,
+   "0.7": 0.8639965314569503,
+   "0.8": 0.8849995033397265,
+   "0.9": 0.9111280130257281
+  },
+  "matched_mcc_signed_band": 0.8756660629301802,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.8502723480167895,
+   "0.7": 0.866407420729698,
+   "0.8": 0.8864822105150204,
+   "0.9": 0.9123945778408794
+  },
+  "matched_mcc_abs_band": 0.8788891392755969,
+  "act_iou_mean": 0.2742358547767782,
+  "act_iou_std": 0.004283914213202571,
+  "act_iou_n_pairs": 6144.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_32/dict_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x8/l0_32/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..47fcdf8d9f807177e7626ab519010284631c1b38
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_32/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 32.0,
+   "recon_r2": 0.8986304301359793,
+   "l1_per_token": 2.7485882850160888,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 11,
+   "dense_rate": 0.0017903645833333333,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.2013852596282959,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.2162807732820511
+  },
+  "seed1": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 32.0,
+   "recon_r2": 0.8987353610220457,
+   "l1_per_token": 2.74875414969738,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 11,
+   "dense_rate": 0.0017903645833333333,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.20043718814849854,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.2155430167913437
+  },
+  "seed2": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 32.0,
+   "recon_r2": 0.8988573053112209,
+   "l1_per_token": 2.7434598983303005,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 11,
+   "dense_rate": 0.0017903645833333333,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.20269536972045898,
+   "n_dup_abs": {
+    "0.5": 2,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 1,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 2,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.21712301671504974
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.5232587456703186,
+   "dcs_abs": 0.8227135539054871,
+   "frac_sign_flipped": 0.5109049479166666,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.5232587456703186,
+    "0.6": 0.5232587456703186,
+    "0.7": 0.5232587456703186,
+    "0.8": 0.5232587456703186,
+    "0.9": 0.5232587456703186
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5232587456703186,
+    "0.6": 0.5232587456703186,
+    "0.7": 0.5232587456703186,
+    "0.8": 0.5232587456703186,
+    "0.9": 0.5232587456703186
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4669596354166667,
+    "0.6": 0.435546875,
+    "0.7": 0.400390625,
+    "0.8": 0.3553059895833333,
+    "0.9": 0.2716471354166667
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8227135539054871,
+    "0.6": 0.8227135539054871,
+    "0.7": 0.8227135539054871,
+    "0.8": 0.8227135539054871,
+    "0.9": 0.8227135539054871
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8227135539054871,
+    "0.6": 0.8227135539054871,
+    "0.7": 0.8227135539054871,
+    "0.8": 0.8227135539054871,
+    "0.9": 0.8227135539054871
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8963216145833334,
+    "0.6": 0.86376953125,
+    "0.7": 0.8165690104166666,
+    "0.8": 0.7278645833333334,
+    "0.9": 0.5564778645833334
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.5331408977508545,
+   "dcs_abs": 0.821063756942749,
+   "frac_sign_flipped": 0.4943033854166667,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.5331409573554993,
+    "0.6": 0.5331409573554993,
+    "0.7": 0.5331409573554993,
+    "0.8": 0.5331409573554993,
+    "0.9": 0.5331409573554993
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5331409573554993,
+    "0.6": 0.5331409573554993,
+    "0.7": 0.5331409573554993,
+    "0.8": 0.5331409573554993,
+    "0.9": 0.5331409573554993
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4822591145833333,
+    "0.6": 0.4498697916666667,
+    "0.7": 0.4119466145833333,
+    "0.8": 0.36767578125,
+    "0.9": 0.28564453125
+   },
+   "matched_cos_abs": {
+    "0.5": 0.821121335029602,
+    "0.6": 0.821063756942749,
+    "0.7": 0.821063756942749,
+    "0.8": 0.821063756942749,
+    "0.9": 0.821063756942749
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9999186197916666,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8210545120042904,
+    "0.6": 0.821063756942749,
+    "0.7": 0.821063756942749,
+    "0.8": 0.821063756942749,
+    "0.9": 0.821063756942749
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8916015625,
+    "0.6": 0.8601888020833334,
+    "0.7": 0.80859375,
+    "0.8": 0.7239583333333334,
+    "0.9": 0.5641276041666666
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.5294851660728455,
+   "dcs_abs": 0.823700487613678,
+   "frac_sign_flipped": 0.5047200520833334,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.5294851660728455,
+    "0.6": 0.5294851660728455,
+    "0.7": 0.5294851660728455,
+    "0.8": 0.5294851660728455,
+    "0.9": 0.5294851660728455
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5294851660728455,
+    "0.6": 0.5294851660728455,
+    "0.7": 0.5294851660728455,
+    "0.8": 0.5294851660728455,
+    "0.9": 0.5294851660728455
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4728190104166667,
+    "0.6": 0.44287109375,
+    "0.7": 0.4109700520833333,
+    "0.8": 0.3653971354166667,
+    "0.9": 0.2815755208333333
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8237238526344299,
+    "0.6": 0.823700487613678,
+    "0.7": 0.823700487613678,
+    "0.8": 0.823700487613678,
+    "0.9": 0.823700487613678
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9999186197916666,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8236568178156933,
+    "0.6": 0.823700487613678,
+    "0.7": 0.823700487613678,
+    "0.8": 0.823700487613678,
+    "0.9": 0.823700487613678
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8914388020833334,
+    "0.6": 0.8642578125,
+    "0.7": 0.8160807291666666,
+    "0.8": 0.7303059895833334,
+    "0.9": 0.5633138020833334
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 11,
+  "n_dense_std": 0.0,
+  "l0_mean": 32,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.8987410321564152,
+  "recon_r2_std": null,
+  "l1_per_token_mean": 2.7469341110145895,
+  "l1_per_token_std": 0.0030098989871755154,
+  "babel_mu1_signed_mean": 0.20150593916575113,
+  "babel_mu1_signed_std": 0.0011339173982507498,
+  "n_dup_signed_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_signed_band": 0.0,
+  "n_dup_signed_band_std": 0.0,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_signed_band": 0.0,
+  "n_dup_clusters_signed_band_std": 0.0,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_signed_band": 1.0,
+  "largest_dup_signed_band_std": 0.0,
+  "dcs_signed_mean": 0.5286282698313395,
+  "dcs_signed_std": 0.004996492350863775,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.5286282896995544,
+   "0.6": 0.5286282896995544,
+   "0.7": 0.5286282896995544,
+   "0.8": 0.5286282896995544,
+   "0.9": 0.5286282896995544
+  },
+  "dedup_dcs_signed_band": 0.5286282896995544,
+  "dedup_dcs_signed_band_std": 0.004996519267150742,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.5286282896995544,
+   "0.6": 0.5286282896995544,
+   "0.7": 0.5286282896995544,
+   "0.8": 0.5286282896995544,
+   "0.9": 0.5286282896995544
+  },
+  "matched_cos_signed_band": 0.5286282896995544,
+  "matched_cos_signed_band_std": 0.004996519267150742,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_signed_band": 1.0,
+  "unique_frac_signed_band_std": 0.0,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.4740125868055556,
+   "0.6": 0.4427625868055556,
+   "0.7": 0.40776909722222215,
+   "0.8": 0.36279296875,
+   "0.9": 0.2796223958333333
+  },
+  "matched_frac_signed_band": 0.37323676215277773,
+  "matched_frac_signed_band_std": 0.006749203358906455,
+  "babel_mu1_abs_mean": 0.21631560226281485,
+  "babel_mu1_abs_std": 0.0007905755708049459,
+  "n_dup_abs_per_threshold": {
+   "0.5": 0.6666666666666666,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_abs_band": 0.0,
+  "n_dup_abs_band_std": 0.0,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 0.3333333333333333,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_abs_band": 0.0,
+  "n_dup_clusters_abs_band_std": 0.0,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 1.3333333333333333,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_abs_band": 1.0,
+  "largest_dup_abs_band_std": 0.0,
+  "dcs_abs_mean": 0.8224925994873047,
+  "dcs_abs_std": 0.0013321797172078913,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.8224749612418236,
+   "0.6": 0.8224925994873047,
+   "0.7": 0.8224925994873047,
+   "0.8": 0.8224925994873047,
+   "0.9": 0.8224925994873047
+  },
+  "dedup_dcs_abs_band": 0.8224925994873047,
+  "dedup_dcs_abs_band_std": 0.0013321797172078913,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.822519580523173,
+   "0.6": 0.8224925994873047,
+   "0.7": 0.8224925994873047,
+   "0.8": 0.8224925994873047,
+   "0.9": 0.8224925994873047
+  },
+  "matched_cos_abs_band": 0.8224925994873047,
+  "matched_cos_abs_band_std": 0.0013321797172078913,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.9999457465277777,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_abs_band": 1.0,
+  "unique_frac_abs_band_std": 0.0,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.8931206597222223,
+   "0.6": 0.8627387152777778,
+   "0.7": 0.813747829861111,
+   "0.8": 0.7273763020833334,
+   "0.9": 0.5613064236111112
+  },
+  "matched_frac_abs_band": 0.7412923177083334,
+  "matched_frac_abs_band_std": 0.0021388446616131874,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_32/seed0/config.json b/lm/gpt2-small_layer8/isae/x8/l0_32/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..4f6f74c4605a9dd3c941492c25517f92ab351c74
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_32/seed0/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..91d383284158e121ce4a5186f0125df0d402a591
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 32.0,
+  "l0_per_token_0.001": 32.0,
+  "l0_per_token_0.01": 31.981455322265624,
+  "l0_per_token_0.05": 25.026007080078124,
+  "l0_per_token_0.1": 7.612111376953125,
+  "l0_per_token_0.5": 0.009205908203125,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 2.7485882850160888,
+  "r2_score": 0.8986304301359793,
+  "l2_loss": 2.3773631301592864
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_32/seed1/config.json b/lm/gpt2-small_layer8/isae/x8/l0_32/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..ca107f7c169ca904540ec33a7211206af747fd8e
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_32/seed1/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..1daefcd64082d7dab381538700870603c9432c80
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 32.0,
+  "l0_per_token_0.001": 32.0,
+  "l0_per_token_0.01": 31.98243935546875,
+  "l0_per_token_0.05": 24.954922119140626,
+  "l0_per_token_0.1": 7.541885986328125,
+  "l0_per_token_0.5": 0.01154814453125,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 2.74875414969738,
+  "r2_score": 0.8987353610220457,
+  "l2_loss": 2.374902245496525
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_32/seed2/config.json b/lm/gpt2-small_layer8/isae/x8/l0_32/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..c247844b1ed1ce682ed350650507bce658425e5e
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_32/seed2/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..f1d51b3f8e86801369354b822b9f09e3c669165c
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 32.0,
+  "l0_per_token_0.001": 32.0,
+  "l0_per_token_0.01": 31.980485546875,
+  "l0_per_token_0.05": 24.931205712890623,
+  "l0_per_token_0.1": 7.5236568359375,
+  "l0_per_token_0.5": 0.00775712890625,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 2.7434598983303005,
+  "r2_score": 0.8988573053112209,
+  "l2_loss": 2.372042355123037
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_64/code_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x8/l0_64/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..eda849715ca4af23deb2078fd0dae04f72b92d9d
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_64/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 64.0,
+   "recon_r2": 0.9197204578579211,
+   "l1_per_token": 3.7595075359029178
+  },
+  "seed1": {
+   "l0": 64.0,
+   "recon_r2": 0.9196443753877038,
+   "l1_per_token": 3.89885244588904
+  },
+  "seed2": {
+   "l0": 64.0,
+   "recon_r2": 0.9197733474466683,
+   "l1_per_token": 3.759107554411879
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.46663840292252584,
+   "mcc_abs": 0.47406919645472173,
+   "act_iou": 0.27234833909243744,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.965792968957384,
+   "matched_mcc_signed": {
+    "0.6": 0.83156707947323,
+    "0.7": 0.8565464486973429,
+    "0.8": 0.8819369291201936,
+    "0.9": 0.9091434602730584
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8352835017718443,
+    "0.7": 0.8567544817762152,
+    "0.8": 0.8807919224965118,
+    "0.9": 0.9098355996395787
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.46529813761299427,
+   "mcc_abs": 0.4716236796159366,
+   "act_iou": 0.27142451324003053,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9656623077887693,
+   "matched_mcc_signed": {
+    "0.6": 0.8362555391178856,
+    "0.7": 0.8563776943231686,
+    "0.8": 0.8800279309483572,
+    "0.9": 0.9112945026214664
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8369905500100518,
+    "0.7": 0.855955100902865,
+    "0.8": 0.8793871386659765,
+    "0.9": 0.9103139370961815
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.4719000648883573,
+   "mcc_abs": 0.4766891501354726,
+   "act_iou": 0.2750840869778461,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9655541059612193,
+   "matched_mcc_signed": {
+    "0.6": 0.8356442852840812,
+    "0.7": 0.854931043412528,
+    "0.8": 0.8800674816913087,
+    "0.9": 0.9129431006403166
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8405926756945172,
+    "0.7": 0.8573849212477066,
+    "0.8": 0.8806873439292049,
+    "0.9": 0.9116026115012689
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 64.0,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9197127268974311,
+  "recon_r2_std": 6.483266006386203e-05,
+  "l1_per_token_mean": 3.805822512067946,
+  "l1_per_token_std": 0.08056653422089236,
+  "input_r2_mean": 0.9656697942357909,
+  "input_r2_std": 0.00011960734885846667,
+  "mcc_signed_mean": 0.46794553514129245,
+  "mcc_signed_std": 0.0034896714623727627,
+  "mcc_abs_mean": 0.47412734206871027,
+  "mcc_abs_std": 0.0025332357924977345,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.834488967958399,
+   "0.7": 0.8559517288110131,
+   "0.8": 0.8806774472532864,
+   "0.9": 0.9111270211782804
+  },
+  "matched_mcc_signed_band": 0.8705612913002447,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.8376222424921377,
+   "0.7": 0.8566981679755955,
+   "0.8": 0.8802888016972311,
+   "0.9": 0.910584049412343
+  },
+  "matched_mcc_abs_band": 0.8712983153943269,
+  "act_iou_mean": 0.272952313103438,
+  "act_iou_std": 0.0019030786741751978,
+  "act_iou_n_pairs": 6144.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_64/dict_metrics_step_final.json b/lm/gpt2-small_layer8/isae/x8/l0_64/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..ce967b50d0b541e304351f0e6049a01ee7631088
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_64/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 64.0,
+   "recon_r2": 0.9196428343110338,
+   "l1_per_token": 3.760366927977996,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 50,
+   "dense_rate": 0.008138020833333334,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.16662007570266724,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.1783999651670456
+  },
+  "seed1": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 64.0,
+   "recon_r2": 0.919564820921079,
+   "l1_per_token": 3.899866392185017,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 48,
+   "dense_rate": 0.0078125,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.16627992689609528,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.17774665355682373
+  },
+  "seed2": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 64.0,
+   "recon_r2": 0.9196784937485526,
+   "l1_per_token": 3.7600670645534113,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 48,
+   "dense_rate": 0.0078125,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.16832061111927032,
+   "n_dup_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 0,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_abs": {
+    "0.5": 1,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_abs": 0.17993734776973724
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.5172920227050781,
+   "dcs_abs": 0.8004654049873352,
+   "frac_sign_flipped": 0.4881184895833333,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.5172920227050781,
+    "0.6": 0.5172920227050781,
+    "0.7": 0.5172920227050781,
+    "0.8": 0.5172920227050781,
+    "0.9": 0.5172920227050781
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5172920227050781,
+    "0.6": 0.5172920227050781,
+    "0.7": 0.5172920227050781,
+    "0.8": 0.5172920227050781,
+    "0.9": 0.5172920227050781
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4835611979166667,
+    "0.6": 0.4503580729166667,
+    "0.7": 0.4033203125,
+    "0.8": 0.3408203125,
+    "0.9": 0.24283854166666666
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8004654049873352,
+    "0.6": 0.8004654049873352,
+    "0.7": 0.8004654049873352,
+    "0.8": 0.8004654049873352,
+    "0.9": 0.8004654049873352
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8004654049873352,
+    "0.6": 0.8004654049873352,
+    "0.7": 0.8004654049873352,
+    "0.8": 0.8004654049873352,
+    "0.9": 0.8004654049873352
+   },
+   "matched_frac_abs": {
+    "0.5": 0.89111328125,
+    "0.6": 0.8523763020833334,
+    "0.7": 0.78076171875,
+    "0.8": 0.6671549479166666,
+    "0.9": 0.4713541666666667
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.5149627327919006,
+   "dcs_abs": 0.8042609691619873,
+   "frac_sign_flipped": 0.4988606770833333,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.5149627327919006,
+    "0.6": 0.5149627327919006,
+    "0.7": 0.5149627327919006,
+    "0.8": 0.5149627327919006,
+    "0.9": 0.5149627327919006
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5149627327919006,
+    "0.6": 0.5149627327919006,
+    "0.7": 0.5149627327919006,
+    "0.8": 0.5149627327919006,
+    "0.9": 0.5149627327919006
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4777018229166667,
+    "0.6": 0.4396158854166667,
+    "0.7": 0.40234375,
+    "0.8": 0.34423828125,
+    "0.9": 0.24055989583333334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8042610287666321,
+    "0.6": 0.8042610287666321,
+    "0.7": 0.8042610287666321,
+    "0.8": 0.8042610287666321,
+    "0.9": 0.8042610287666321
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8042610287666321,
+    "0.6": 0.8042610287666321,
+    "0.7": 0.8042610287666321,
+    "0.8": 0.8042610287666321,
+    "0.9": 0.8042610287666321
+   },
+   "matched_frac_abs": {
+    "0.5": 0.89794921875,
+    "0.6": 0.8557942708333334,
+    "0.7": 0.791015625,
+    "0.8": 0.68115234375,
+    "0.9": 0.474609375
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.5202613472938538,
+   "dcs_abs": 0.8031803965568542,
+   "frac_sign_flipped": 0.4873046875,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.5202613472938538,
+    "0.6": 0.5202613472938538,
+    "0.7": 0.5202613472938538,
+    "0.8": 0.5202613472938538,
+    "0.9": 0.5202613472938538
+   },
+   "unique_frac_signed": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5202613472938538,
+    "0.6": 0.5202613472938538,
+    "0.7": 0.5202613472938538,
+    "0.8": 0.5202613472938538,
+    "0.9": 0.5202613472938538
+   },
+   "matched_frac_signed": {
+    "0.5": 0.4855143229166667,
+    "0.6": 0.44970703125,
+    "0.7": 0.4119466145833333,
+    "0.8": 0.3504231770833333,
+    "0.9": 0.240234375
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8031803965568542,
+    "0.6": 0.8031803965568542,
+    "0.7": 0.8031803965568542,
+    "0.8": 0.8031803965568542,
+    "0.9": 0.8031803965568542
+   },
+   "unique_frac_abs": {
+    "0.5": 1.0,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8031803965568542,
+    "0.6": 0.8031803965568542,
+    "0.7": 0.8031803965568542,
+    "0.8": 0.8031803965568542,
+    "0.9": 0.8031803965568542
+   },
+   "matched_frac_abs": {
+    "0.5": 0.89208984375,
+    "0.6": 0.8474934895833334,
+    "0.7": 0.7884114583333334,
+    "0.8": 0.6793619791666666,
+    "0.9": 0.4752604166666667
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 49,
+  "n_dense_std": 1.1547005383792517,
+  "l0_mean": 64,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9196287163268885,
+  "recon_r2_std": null,
+  "l1_per_token_mean": 3.806766794905475,
+  "l1_per_token_std": 0.08062675595260727,
+  "babel_mu1_signed_mean": 0.16707353790601095,
+  "babel_mu1_signed_std": 0.0010933064538704996,
+  "n_dup_signed_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_signed_band": 0.0,
+  "n_dup_signed_band_std": 0.0,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_signed_band": 0.0,
+  "n_dup_clusters_signed_band_std": 0.0,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_signed_band": 1.0,
+  "largest_dup_signed_band_std": 0.0,
+  "dcs_signed_mean": 0.5175053675969442,
+  "dcs_signed_std": 0.00265574206997611,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.5175053675969442,
+   "0.6": 0.5175053675969442,
+   "0.7": 0.5175053675969442,
+   "0.8": 0.5175053675969442,
+   "0.9": 0.5175053675969442
+  },
+  "dedup_dcs_signed_band": 0.5175053675969442,
+  "dedup_dcs_signed_band_std": 0.00265574206997611,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.5175053675969442,
+   "0.6": 0.5175053675969442,
+   "0.7": 0.5175053675969442,
+   "0.8": 0.5175053675969442,
+   "0.9": 0.5175053675969442
+  },
+  "matched_cos_signed_band": 0.5175053675969442,
+  "matched_cos_signed_band_std": 0.00265574206997611,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_signed_band": 1.0,
+  "unique_frac_signed_band_std": 0.0,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.4822591145833333,
+   "0.6": 0.44656032986111116,
+   "0.7": 0.4058702256944444,
+   "0.8": 0.34516059027777773,
+   "0.9": 0.2412109375
+  },
+  "matched_frac_signed_band": 0.3597005208333333,
+  "matched_frac_signed_band_std": 0.003209879300418782,
+  "babel_mu1_abs_mean": 0.17869465549786887,
+  "babel_mu1_abs_std": 0.0011246853235166415,
+  "n_dup_abs_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_abs_band": 0.0,
+  "n_dup_abs_band_std": 0.0,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 0.0,
+   "0.6": 0.0,
+   "0.7": 0.0,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_abs_band": 0.0,
+  "n_dup_clusters_abs_band_std": 0.0,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_abs_band": 1.0,
+  "largest_dup_abs_band_std": 0.0,
+  "dcs_abs_mean": 0.8026355902353922,
+  "dcs_abs_std": 0.0019555529389163394,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.8026356101036072,
+   "0.6": 0.8026356101036072,
+   "0.7": 0.8026356101036072,
+   "0.8": 0.8026356101036072,
+   "0.9": 0.8026356101036072
+  },
+  "dedup_dcs_abs_band": 0.8026356101036072,
+  "dedup_dcs_abs_band_std": 0.0019555777095840794,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.8026356101036072,
+   "0.6": 0.8026356101036072,
+   "0.7": 0.8026356101036072,
+   "0.8": 0.8026356101036072,
+   "0.9": 0.8026356101036072
+  },
+  "matched_cos_abs_band": 0.8026356101036072,
+  "matched_cos_abs_band_std": 0.0019555777095840794,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 1.0,
+   "0.6": 1.0,
+   "0.7": 1.0,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_abs_band": 1.0,
+  "unique_frac_abs_band_std": 0.0,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.8937174479166666,
+   "0.6": 0.8518880208333334,
+   "0.7": 0.7867296006944445,
+   "0.8": 0.6758897569444443,
+   "0.9": 0.4737413194444445
+  },
+  "matched_frac_abs_band": 0.6970621744791667,
+  "matched_frac_abs_band_std": 0.0038969140279339165,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_64/seed0/config.json b/lm/gpt2-small_layer8/isae/x8/l0_64/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..912b65aaadacd5e52a6b294a895f07ff548f9a07
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_64/seed0/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..c95dc826df0bc497b33fa7a191a56213f2af0317
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.0,
+  "l0_per_token_0.001": 64.0,
+  "l0_per_token_0.01": 63.90839033203125,
+  "l0_per_token_0.05": 23.06042939453125,
+  "l0_per_token_0.1": 7.517118896484375,
+  "l0_per_token_0.5": 0.00372294921875,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 3.760366927977996,
+  "r2_score": 0.9196428343110338,
+  "l2_loss": 1.8845711115210582
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_64/seed1/config.json b/lm/gpt2-small_layer8/isae/x8/l0_64/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..884c307e2edf20d0b9bc4dfe9176dc9311b8e1eb
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_64/seed1/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..a8b2c57e15ac779be13904823f5147441fb8f4a1
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.0,
+  "l0_per_token_0.001": 64.0,
+  "l0_per_token_0.01": 63.90637290039062,
+  "l0_per_token_0.05": 23.29100732421875,
+  "l0_per_token_0.1": 7.81364267578125,
+  "l0_per_token_0.5": 0.004352685546875,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 3.899866392185017,
+  "r2_score": 0.919564820921079,
+  "l2_loss": 1.886400715387247
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_64/seed2/config.json b/lm/gpt2-small_layer8/isae/x8/l0_64/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..e0f0b8b967544c630769a9f6b12e5881dec0c6f2
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_64/seed2/config.json
@@ -0,0 +1,20 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "isae",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "use_abstopk": true,
+    "num_encode_steps": 1,
+    "initial_step_size": 0.9,
+    "rip_weight": 0.1,
+    "rip_multiplier": 2,
+    "rip_weighted": false,
+    "l1_weight": 0.01,
+    "input_norm": true,
+    "class_type": "IsaeSAE",
+    "module_name": "consdl.models.isae"
+}
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/isae/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..8536ea7e708f87f0561c1e281cfcfd8ff9e4775b
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.0,
+  "l0_per_token_0.001": 64.0,
+  "l0_per_token_0.01": 63.91211962890625,
+  "l0_per_token_0.05": 22.981458740234373,
+  "l0_per_token_0.1": 7.42931689453125,
+  "l0_per_token_0.5": 0.0044779296875,
+  "l0_per_token_1.0": 0.0,
+  "l1_per_token": 3.7600670645534113,
+  "r2_score": 0.9196784937485526,
+  "l2_loss": 1.8837348109220409
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/isae/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/isae/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/isae/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_128/code_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x16/l0_128/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..6d243acb1ead9de3eac513ece4ade82b2b099434
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_128/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 132.318875,
+   "recon_r2": 0.9284653626609871,
+   "l1_per_token": 251.71471418002963
+  },
+  "seed1": {
+   "l0": 134.133515,
+   "recon_r2": 0.9293312226933406,
+   "l1_per_token": 254.24325088425343
+  },
+  "seed2": {
+   "l0": 131.88106666666667,
+   "recon_r2": 0.9286152778706667,
+   "l1_per_token": 252.05869134148875
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7629224289278098,
+   "mcc_abs": 0.7629242091940114,
+   "act_iou": 0.43090739498778535,
+   "act_iou_n_pairs": 12269,
+   "input_r2": 0.9849076151212589,
+   "matched_mcc_signed": {
+    "0.6": 0.8549075297642076,
+    "0.7": 0.8896918926931576,
+    "0.8": 0.9213459854532546,
+    "0.9": 0.9513888272414164
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.854073145101787,
+    "0.7": 0.889383335387622,
+    "0.8": 0.9211461345274717,
+    "0.9": 0.9513636849712505
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7616383091469113,
+   "mcc_abs": 0.7616386788542294,
+   "act_iou": 0.43398320719440575,
+   "act_iou_n_pairs": 12168,
+   "input_r2": 0.9851181069698298,
+   "matched_mcc_signed": {
+    "0.6": 0.8589421647849268,
+    "0.7": 0.8908433515485337,
+    "0.8": 0.9221185684200619,
+    "0.9": 0.9510073341930781
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8577759108626889,
+    "0.7": 0.8906533000729361,
+    "0.8": 0.9221073058953662,
+    "0.9": 0.9510115842687815
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7633901120194148,
+   "mcc_abs": 0.763392397754945,
+   "act_iou": 0.431304068296149,
+   "act_iou_n_pairs": 12269,
+   "input_r2": 0.9851728501827905,
+   "matched_mcc_signed": {
+    "0.6": 0.8575546556140516,
+    "0.7": 0.8918022708863074,
+    "0.8": 0.9208309486466134,
+    "0.9": 0.9513308365809803
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8566399539485391,
+    "0.7": 0.8917338871568433,
+    "0.8": 0.9206864365243346,
+    "0.9": 0.9513547366209538
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 132.77781888888887,
+  "l0_std": 1.1943002107700378,
+  "recon_r2_mean": 0.9288039544083314,
+  "recon_r2_std": 0.00046273915525450884,
+  "l1_per_token_mean": 252.67221880192392,
+  "l1_per_token_std": 1.3713812105620995,
+  "input_r2_mean": 0.9850661907579598,
+  "input_r2_std": 0.0001400317080436662,
+  "mcc_signed_mean": 0.762650283364712,
+  "mcc_signed_std": 0.000907056079668285,
+  "mcc_abs_mean": 0.7626517619343952,
+  "mcc_abs_std": 0.0009080490777684039,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8571347833877286,
+   "0.7": 0.8907791717093329,
+   "0.8": 0.92143183417331,
+   "0.9": 0.9512423326718249
+  },
+  "matched_mcc_signed_band": 0.9051470304855491,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.8561630033043383,
+   "0.7": 0.8905901742058004,
+   "0.8": 0.9213132923157241,
+   "0.9": 0.9512433352869953
+  },
+  "matched_mcc_abs_band": 0.9048274512782145,
+  "act_iou_mean": 0.43206489015944677,
+  "act_iou_std": 0.0016731086675265945,
+  "act_iou_n_pairs": 12235.333333333334,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_128/dict_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x16/l0_128/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..5f0cedbd7fb26250b2d96ce4472140e2382f1585
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_128/dict_metrics_step_final.json
@@ -0,0 +1,595 @@
+{
+ "schema": 3,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 132,
+   "l0_raw": 132.3279265625,
+   "recon_r2": 0.9283246495916933,
+   "l1_per_token": 251.69036003414433,
+   "stats_source": "stats",
+   "n_dead": 162,
+   "dead_rate": 0.01318359375,
+   "alive_source": "stats",
+   "n_dense": 36,
+   "dense_rate": 0.0029296875,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 3369,
+    "0.6": 1440,
+    "0.7": 461,
+    "0.8": 73,
+    "0.9": 6
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 1399,
+    "0.6": 622,
+    "0.7": 213,
+    "0.8": 36,
+    "0.9": 3
+   },
+   "n_dup_removed_signed": {
+    "0.5": 1970,
+    "0.6": 818,
+    "0.7": 248,
+    "0.8": 37,
+    "0.9": 3
+   },
+   "largest_dup_signed": {
+    "0.5": 13,
+    "0.6": 7,
+    "0.7": 4,
+    "0.8": 3,
+    "0.9": 2
+   },
+   "babel_mu1_signed": 0.46096938848495483,
+   "n_dup_abs": {
+    "0.5": 3475,
+    "0.6": 1498,
+    "0.7": 486,
+    "0.8": 73,
+    "0.9": 6
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 1450,
+    "0.6": 649,
+    "0.7": 226,
+    "0.8": 36,
+    "0.9": 3
+   },
+   "n_dup_removed_abs": {
+    "0.5": 2025,
+    "0.6": 849,
+    "0.7": 260,
+    "0.8": 37,
+    "0.9": 3
+   },
+   "largest_dup_abs": {
+    "0.5": 10,
+    "0.6": 7,
+    "0.7": 4,
+    "0.8": 3,
+    "0.9": 2
+   },
+   "babel_mu1_abs": 0.46819061040878296,
+   "dup_excludes_dead": true
+  },
+  "seed1": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 134,
+   "l0_raw": 134.03701655273437,
+   "recon_r2": 0.9291439052970092,
+   "l1_per_token": 254.00240409847908,
+   "stats_source": "stats",
+   "n_dead": 22,
+   "dead_rate": 0.0017903645833333333,
+   "alive_source": "stats",
+   "n_dense": 40,
+   "dense_rate": 0.0032552083333333335,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 3494,
+    "0.6": 1485,
+    "0.7": 463,
+    "0.8": 73,
+    "0.9": 8
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 1456,
+    "0.6": 638,
+    "0.7": 211,
+    "0.8": 34,
+    "0.9": 3
+   },
+   "n_dup_removed_signed": {
+    "0.5": 2038,
+    "0.6": 847,
+    "0.7": 252,
+    "0.8": 39,
+    "0.9": 5
+   },
+   "largest_dup_signed": {
+    "0.5": 16,
+    "0.6": 9,
+    "0.7": 5,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_signed": 0.46261340379714966,
+   "n_dup_abs": {
+    "0.5": 3595,
+    "0.6": 1538,
+    "0.7": 483,
+    "0.8": 77,
+    "0.9": 8
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 1506,
+    "0.6": 665,
+    "0.7": 222,
+    "0.8": 36,
+    "0.9": 3
+   },
+   "n_dup_removed_abs": {
+    "0.5": 2089,
+    "0.6": 873,
+    "0.7": 261,
+    "0.8": 41,
+    "0.9": 5
+   },
+   "largest_dup_abs": {
+    "0.5": 15,
+    "0.6": 9,
+    "0.7": 5,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_abs": 0.4696386158466339,
+   "dup_excludes_dead": true
+  },
+  "seed2": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 132,
+   "l0_raw": 131.888416015625,
+   "recon_r2": 0.9458630207665515,
+   "l1_per_token": 251.95925920856158,
+   "stats_source": "stats",
+   "n_dead": 174,
+   "dead_rate": 0.01416015625,
+   "alive_source": "stats",
+   "n_dense": 40,
+   "dense_rate": 0.0032552083333333335,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 3339,
+    "0.6": 1428,
+    "0.7": 431,
+    "0.8": 63,
+    "0.9": 2
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 1381,
+    "0.6": 619,
+    "0.7": 198,
+    "0.8": 30,
+    "0.9": 1
+   },
+   "n_dup_removed_signed": {
+    "0.5": 1958,
+    "0.6": 809,
+    "0.7": 233,
+    "0.8": 33,
+    "0.9": 1
+   },
+   "largest_dup_signed": {
+    "0.5": 14,
+    "0.6": 7,
+    "0.7": 5,
+    "0.8": 3,
+    "0.9": 2
+   },
+   "babel_mu1_signed": 0.4603356420993805,
+   "n_dup_abs": {
+    "0.5": 3455,
+    "0.6": 1487,
+    "0.7": 460,
+    "0.8": 73,
+    "0.9": 2
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 1435,
+    "0.6": 651,
+    "0.7": 212,
+    "0.8": 35,
+    "0.9": 1
+   },
+   "n_dup_removed_abs": {
+    "0.5": 2020,
+    "0.6": 836,
+    "0.7": 248,
+    "0.8": 38,
+    "0.9": 1
+   },
+   "largest_dup_abs": {
+    "0.5": 12,
+    "0.6": 6,
+    "0.7": 5,
+    "0.8": 3,
+    "0.9": 2
+   },
+   "babel_mu1_abs": 0.4676814675331116,
+   "dup_excludes_dead": true
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.8383777737617493,
+   "dcs_abs": 0.8386764526367188,
+   "frac_sign_flipped": 0.011311848958333334,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 12126,
+   "n_alive_b": 12266,
+   "matched_cos_signed": {
+    "0.5": 0.8338509202003479,
+    "0.6": 0.8410871028900146,
+    "0.7": 0.8456299304962158,
+    "0.8": 0.8478838205337524,
+    "0.9": 0.8482896685600281
+   },
+   "unique_frac_signed": {
+    "0.5": 0.8294270833333334,
+    "0.6": 0.9247639973958334,
+    "0.7": 0.97216796875,
+    "0.8": 0.9894205729166666,
+    "0.9": 0.9921875
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.6916185366765907,
+    "0.6": 0.7778070714266505,
+    "0.7": 0.8220943318447098,
+    "0.8": 0.8389136954792775,
+    "0.9": 0.8416624055244029
+   },
+   "matched_frac_signed": {
+    "0.5": 0.7294921875,
+    "0.6": 0.7798665364583334,
+    "0.7": 0.7718912760416666,
+    "0.8": 0.7201334635416666,
+    "0.9": 0.6128743489583334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8350489139556885,
+    "0.6": 0.8416901230812073,
+    "0.7": 0.8460467457771301,
+    "0.8": 0.8479912281036377,
+    "0.9": 0.8484496474266052
+   },
+   "unique_frac_abs": {
+    "0.5": 0.8251139322916666,
+    "0.6": 0.9224446614583334,
+    "0.7": 0.9713134765625,
+    "0.8": 0.9893391927083334,
+    "0.9": 0.9921875
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.6890104930498637,
+    "0.6": 0.7764125606384672,
+    "0.7": 0.8217766059751739,
+    "0.8": 0.8389509570358011,
+    "0.9": 0.8418211345560849
+   },
+   "matched_frac_abs": {
+    "0.5": 0.726806640625,
+    "0.6": 0.7784016927083334,
+    "0.7": 0.77197265625,
+    "0.8": 0.7205403645833334,
+    "0.9": 0.6129557291666666
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.8356342911720276,
+   "dcs_abs": 0.8359951376914978,
+   "frac_sign_flipped": 0.01318359375,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 12126,
+   "n_alive_b": 12114,
+   "matched_cos_signed": {
+    "0.5": 0.8328262567520142,
+    "0.6": 0.8397153615951538,
+    "0.7": 0.8432490825653076,
+    "0.8": 0.8456360101699829,
+    "0.9": 0.8462221622467041
+   },
+   "unique_frac_signed": {
+    "0.5": 0.8264973958333334,
+    "0.6": 0.9201253255208334,
+    "0.7": 0.9667561848958334,
+    "0.8": 0.9834798177083334,
+    "0.9": 0.9861653645833334
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.6883287323871627,
+    "0.6": 0.7726433704325852,
+    "0.7": 0.8152162659777484,
+    "0.8": 0.8316659491295771,
+    "0.9": 0.8345149871505176
+   },
+   "matched_frac_signed": {
+    "0.5": 0.7274576822916666,
+    "0.6": 0.7763671875,
+    "0.7": 0.771240234375,
+    "0.8": 0.720947265625,
+    "0.9": 0.6177571614583334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8339460492134094,
+    "0.6": 0.8401309251785278,
+    "0.7": 0.8434145450592041,
+    "0.8": 0.8460039496421814,
+    "0.9": 0.8464446067810059
+   },
+   "unique_frac_abs": {
+    "0.5": 0.8217366536458334,
+    "0.6": 0.9177652994791666,
+    "0.7": 0.9656575520833334,
+    "0.8": 0.9832763671875,
+    "0.9": 0.9861653645833334
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.6852840358017905,
+    "0.6": 0.7710430101481809,
+    "0.7": 0.8144496249733493,
+    "0.8": 0.8318556902304408,
+    "0.9": 0.8347343542457869
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7230631510416666,
+    "0.6": 0.77490234375,
+    "0.7": 0.7704264322916666,
+    "0.8": 0.7207845052083334,
+    "0.9": 0.6175944010416666
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.838294267654419,
+   "dcs_abs": 0.8385542035102844,
+   "frac_sign_flipped": 0.010986328125,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 12266,
+   "n_alive_b": 12114,
+   "matched_cos_signed": {
+    "0.5": 0.8344737887382507,
+    "0.6": 0.8416755199432373,
+    "0.7": 0.845518171787262,
+    "0.8": 0.8483132719993591,
+    "0.9": 0.8489343523979187
+   },
+   "unique_frac_signed": {
+    "0.5": 0.8294270833333334,
+    "0.6": 0.9246419270833334,
+    "0.7": 0.9722900390625,
+    "0.8": 0.9890950520833334,
+    "0.9": 0.9917805989583334
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.6921351607112836,
+    "0.6": 0.7782484747391816,
+    "0.7": 0.8220888962750905,
+    "0.8": 0.8390624599511891,
+    "0.9": 0.8419566204975126
+   },
+   "matched_frac_signed": {
+    "0.5": 0.7311197916666666,
+    "0.6": 0.7793782552083334,
+    "0.7": 0.7703450520833334,
+    "0.8": 0.725341796875,
+    "0.9": 0.6162923177083334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8358864188194275,
+    "0.6": 0.8419227600097656,
+    "0.7": 0.8458235263824463,
+    "0.8": 0.848504900932312,
+    "0.9": 0.8490839004516602
+   },
+   "unique_frac_abs": {
+    "0.5": 0.8248291015625,
+    "0.6": 0.9224853515625,
+    "0.7": 0.9713134765625,
+    "0.8": 0.9888102213541666,
+    "0.9": 0.9917805989583334
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.689463443843124,
+    "0.6": 0.776661413256079,
+    "0.7": 0.8215597899688873,
+    "0.8": 0.8390103189109747,
+    "0.9": 0.8421049393558254
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7275390625,
+    "0.6": 0.7772623697916666,
+    "0.7": 0.769775390625,
+    "0.8": 0.725341796875,
+    "0.9": 0.616455078125
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 119,
+  "n_dead_std": 84.5064100132844,
+  "n_dense_mean": 39,
+  "n_dense_std": 2.309401076758503,
+  "l0_mean": 133,
+  "l0_std": 1.1350948846976243,
+  "recon_r2_mean": 0.9344438585517514,
+  "recon_r2_std": 0.009897764608416048,
+  "l1_per_token_mean": 252.55067444706165,
+  "l1_per_token_std": 1.2644033878094298,
+  "babel_mu1_signed_mean": 0.4613061447938283,
+  "babel_mu1_signed_std": 0.0011756288515644053,
+  "n_dup_signed_per_threshold": {
+   "0.5": 3400.6666666666665,
+   "0.6": 1451.0,
+   "0.7": 451.6666666666667,
+   "0.8": 69.66666666666667,
+   "0.9": 5.333333333333333
+  },
+  "n_dup_signed_band": 494.4166666666667,
+  "n_dup_signed_band_std": 13.134718624063986,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 1412.0,
+   "0.6": 626.3333333333334,
+   "0.7": 207.33333333333334,
+   "0.8": 33.333333333333336,
+   "0.9": 2.3333333333333335
+  },
+  "n_dup_clusters_signed_band": 217.33333333333337,
+  "n_dup_clusters_signed_band_std": 4.856267428111155,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 14.333333333333334,
+   "0.6": 7.666666666666667,
+   "0.7": 4.666666666666667,
+   "0.8": 3.3333333333333335,
+   "0.9": 2.6666666666666665
+  },
+  "largest_dup_signed_band": 4.583333333333334,
+  "largest_dup_signed_band_std": 0.8036375634160796,
+  "dcs_signed_mean": 0.8374354441960653,
+  "dcs_signed_std": 0.0015604029862475984,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.6906941432583457,
+   "0.6": 0.7762329721994724,
+   "0.7": 0.8197998313658497,
+   "0.8": 0.8365473681866812,
+   "0.9": 0.8393780043908111
+  },
+  "dedup_dcs_signed_band": 0.8179895440357036,
+  "dedup_dcs_signed_band_std": 0.003880830469657095,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.8337169885635376,
+   "0.6": 0.8408259948094686,
+   "0.7": 0.8447990616162618,
+   "0.8": 0.8472777009010315,
+   "0.9": 0.8478153944015503
+  },
+  "matched_cos_signed_band": 0.8451795379320781,
+  "matched_cos_signed_band_std": 0.0012910567685389398,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.8284505208333334,
+   "0.6": 0.9231770833333334,
+   "0.7": 0.9704047309027778,
+   "0.8": 0.9873318142361112,
+   "0.9": 0.9900444878472223
+  },
+  "unique_frac_signed_band": 0.9677395290798612,
+  "unique_frac_signed_band_std": 0.003125835898536252,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.7293565538194443,
+   "0.6": 0.778537326388889,
+   "0.7": 0.7711588541666666,
+   "0.8": 0.7221408420138888,
+   "0.9": 0.6156412760416666
+  },
+  "matched_frac_signed_band": 0.7218695746527778,
+  "matched_frac_signed_band_std": 0.0008618076969612166,
+  "babel_mu1_abs_mean": 0.46850356459617615,
+  "babel_mu1_abs_std": 0.0010154125382503567,
+  "n_dup_abs_per_threshold": {
+   "0.5": 3508.3333333333335,
+   "0.6": 1507.6666666666667,
+   "0.7": 476.3333333333333,
+   "0.8": 74.33333333333333,
+   "0.9": 5.333333333333333
+  },
+  "n_dup_abs_band": 515.9166666666667,
+  "n_dup_abs_band_std": 10.500992016630303,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 1463.6666666666667,
+   "0.6": 655.0,
+   "0.7": 220.0,
+   "0.8": 35.666666666666664,
+   "0.9": 2.3333333333333335
+  },
+  "n_dup_clusters_abs_band": 228.25,
+  "n_dup_clusters_abs_band_std": 3.3819373146171707,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 12.333333333333334,
+   "0.6": 7.333333333333333,
+   "0.7": 4.666666666666667,
+   "0.8": 3.3333333333333335,
+   "0.9": 2.6666666666666665
+  },
+  "largest_dup_abs_band": 4.5,
+  "largest_dup_abs_band_std": 0.8660254037844386,
+  "dcs_abs_mean": 0.8377419312795004,
+  "dcs_abs_std": 0.0015140020117369508,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.6879193242315927,
+   "0.6": 0.7747056613475757,
+   "0.7": 0.8192620069724702,
+   "0.8": 0.8366056553924056,
+   "0.9": 0.8395534760525658
+  },
+  "dedup_dcs_abs_band": 0.8175316999412543,
+  "dedup_dcs_abs_band_std": 0.003906948128289929,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.8349604606628418,
+   "0.6": 0.8412479360898336,
+   "0.7": 0.8450949390729269,
+   "0.8": 0.8475000262260437,
+   "0.9": 0.8479927182197571
+  },
+  "matched_cos_abs_band": 0.8454589049021404,
+  "matched_cos_abs_band_std": 0.0012729890281750068,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.8238932291666666,
+   "0.6": 0.9208984375,
+   "0.7": 0.9694281684027778,
+   "0.8": 0.9871419270833334,
+   "0.9": 0.9900444878472223
+  },
+  "unique_frac_abs_band": 0.9668782552083334,
+  "unique_frac_abs_band_std": 0.0031734531539730423,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.7258029513888888,
+   "0.6": 0.77685546875,
+   "0.7": 0.7707248263888888,
+   "0.8": 0.7222222222222223,
+   "0.9": 0.6156684027777778
+  },
+  "matched_frac_abs_band": 0.7213677300347222,
+  "matched_frac_abs_band_std": 0.0007285498477615751,
+  "alive_source": "stats",
+  "frac_denominator": "d_sae",
+  "d_sae": 12288
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_128/seed0/config.json b/lm/gpt2-small_layer8/relu/x16/l0_128/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..5949eb30b783de244a80f491fa25f7af3f2edec1
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_128/seed0/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..a9501d4ac1d2b4d7151c307aaec1c8e7ab6ffa48
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 132.3279265625,
+  "l0_per_token_0.001": 132.22594365234374,
+  "l0_per_token_0.01": 131.3115458496094,
+  "l0_per_token_0.05": 127.33303525390625,
+  "l0_per_token_0.1": 122.54965546875,
+  "l0_per_token_0.5": 90.895825390625,
+  "l0_per_token_1.0": 63.98563916015625,
+  "l1_per_token": 251.69036003414433,
+  "r2_score": 0.9283246495916933,
+  "l2_loss": 1.6809614259894607
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_128/seed1/config.json b/lm/gpt2-small_layer8/relu/x16/l0_128/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..4c15515c994745470445ff31f3f2293062c96289
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_128/seed1/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..5118d2fb34cbed9e366fa844296c1f3d0ef3812a
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 134.03701655273437,
+  "l0_per_token_0.001": 133.93394477539061,
+  "l0_per_token_0.01": 133.01055405273436,
+  "l0_per_token_0.05": 128.99053842773438,
+  "l0_per_token_0.1": 124.1567912109375,
+  "l0_per_token_0.5": 92.13887353515625,
+  "l0_per_token_1.0": 64.861635546875,
+  "l1_per_token": 254.00240409847908,
+  "r2_score": 0.9291439052970092,
+  "l2_loss": 1.6617478856187067
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_128/seed2/config.json b/lm/gpt2-small_layer8/relu/x16/l0_128/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..43913a5027e3e99b474c9dfef9490cb3c811038a
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_128/seed2/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..69e25350019baafc45a66021e3b0587638d9df72
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 131.888416015625,
+  "l0_per_token_0.001": 131.78722841796875,
+  "l0_per_token_0.01": 130.87964873046874,
+  "l0_per_token_0.05": 126.93050698242187,
+  "l0_per_token_0.1": 122.1808310546875,
+  "l0_per_token_0.5": 90.73830991210937,
+  "l0_per_token_1.0": 63.97712749023437,
+  "l1_per_token": 251.95925920856158,
+  "r2_score": 0.9458630207665515,
+  "l2_loss": 1.6782572737197297
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_32/code_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x16/l0_32/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..9e75967cad635a7d6db42ffe2a5e17985e22a191
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_32/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 29.287398333333332,
+   "recon_r2": 0.8625571860584201,
+   "l1_per_token": 151.51710788889537
+  },
+  "seed1": {
+   "l0": 30.701116666666667,
+   "recon_r2": 0.8709821223252308,
+   "l1_per_token": 151.3712687154165
+  },
+  "seed2": {
+   "l0": 28.979218333333332,
+   "recon_r2": 0.8614838299524523,
+   "l1_per_token": 150.59068381950524
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.32071659661113605,
+   "mcc_abs": 0.3207169979154952,
+   "act_iou": 0.34725129807120775,
+   "act_iou_n_pairs": 6619,
+   "input_r2": 0.9804652086639377,
+   "matched_mcc_signed": {
+    "0.6": 0.8809305592109807,
+    "0.7": 0.8954102515109008,
+    "0.8": 0.9098118858439597,
+    "0.9": 0.9373204218005723
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8797950902702741,
+    "0.7": 0.894883177320663,
+    "0.8": 0.9093115541350578,
+    "0.9": 0.9373204218005723
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.30827607228014714,
+   "mcc_abs": 0.30827634929490305,
+   "act_iou": 0.5212747336587108,
+   "act_iou_n_pairs": 4910,
+   "input_r2": 0.986322486966254,
+   "matched_mcc_signed": {
+    "0.6": 0.9113229347930276,
+    "0.7": 0.9287954354032282,
+    "0.8": 0.943940745550279,
+    "0.9": 0.960834248052799
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.9107510918122217,
+    "0.7": 0.9280298577043768,
+    "0.8": 0.9434981699007451,
+    "0.9": 0.960834248052799
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.31435740033249066,
+   "mcc_abs": 0.3143577167673639,
+   "act_iou": 0.34146120713231304,
+   "act_iou_n_pairs": 6518,
+   "input_r2": 0.9800706088269118,
+   "matched_mcc_signed": {
+    "0.6": 0.8756852907149243,
+    "0.7": 0.8925075153531802,
+    "0.8": 0.9063647138145173,
+    "0.9": 0.9348364561518471
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8744313318001831,
+    "0.7": 0.8918371766491703,
+    "0.8": 0.9059206687920093,
+    "0.9": 0.9348364561518472
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 29.65591111111111,
+  "l0_std": 0.9181964485463612,
+  "recon_r2_mean": 0.8650077127787011,
+  "recon_r2_std": 0.005201749744538216,
+  "l1_per_token_mean": 151.15968680793904,
+  "l1_per_token_std": 0.4981370963552013,
+  "input_r2_mean": 0.9822861014857013,
+  "input_r2_std": 0.003501175955098319,
+  "mcc_signed_mean": 0.3144500230745913,
+  "mcc_signed_std": 0.006220779343197077,
+  "mcc_abs_mean": 0.314450354659254,
+  "mcc_abs_std": 0.006220841652021101,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8893129282396443,
+   "0.7": 0.9055710674224363,
+   "0.8": 0.9200391150695854,
+   "0.9": 0.9443303753350728
+  },
+  "matched_mcc_signed_band": 0.9148133715166846,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.888325837960893,
+   "0.7": 0.9049167372247368,
+   "0.8": 0.9195767976092708,
+   "0.9": 0.9443303753350728
+  },
+  "matched_mcc_abs_band": 0.9142874370324933,
+  "act_iou_mean": 0.4033290796207439,
+  "act_iou_std": 0.10218495128003442,
+  "act_iou_n_pairs": 6015.666666666667,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_32/dict_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x16/l0_32/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..9fd97d1ab00cbe68ddaaba5646342dd57590fb58
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_32/dict_metrics_step_final.json
@@ -0,0 +1,595 @@
+{
+ "schema": 3,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 29,
+   "l0_raw": 29.27993671875,
+   "recon_r2": 0.8623482589771835,
+   "l1_per_token": 151.43357586614445,
+   "stats_source": "stats",
+   "n_dead": 7829,
+   "dead_rate": 0.6371256510416666,
+   "alive_source": "stats",
+   "n_dense": 31,
+   "dense_rate": 0.0025227864583333335,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 1778,
+    "0.6": 825,
+    "0.7": 344,
+    "0.8": 61,
+    "0.9": 5
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 752,
+    "0.6": 362,
+    "0.7": 156,
+    "0.8": 28,
+    "0.9": 1
+   },
+   "n_dup_removed_signed": {
+    "0.5": 1026,
+    "0.6": 463,
+    "0.7": 188,
+    "0.8": 33,
+    "0.9": 4
+   },
+   "largest_dup_signed": {
+    "0.5": 8,
+    "0.6": 7,
+    "0.7": 5,
+    "0.8": 5,
+    "0.9": 5
+   },
+   "babel_mu1_signed": 0.511090099811554,
+   "n_dup_abs": {
+    "0.5": 1779,
+    "0.6": 828,
+    "0.7": 348,
+    "0.8": 71,
+    "0.9": 5
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 751,
+    "0.6": 359,
+    "0.7": 157,
+    "0.8": 33,
+    "0.9": 1
+   },
+   "n_dup_removed_abs": {
+    "0.5": 1028,
+    "0.6": 469,
+    "0.7": 191,
+    "0.8": 38,
+    "0.9": 4
+   },
+   "largest_dup_abs": {
+    "0.5": 8,
+    "0.6": 7,
+    "0.7": 6,
+    "0.8": 5,
+    "0.9": 5
+   },
+   "babel_mu1_abs": 0.5118075609207153,
+   "dup_excludes_dead": true
+  },
+  "seed1": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 31,
+   "l0_raw": 30.683714013671874,
+   "recon_r2": 0.8707289851082196,
+   "l1_per_token": 151.1954084528797,
+   "stats_source": "stats",
+   "n_dead": 5990,
+   "dead_rate": 0.4874674479166667,
+   "alive_source": "stats",
+   "n_dense": 26,
+   "dense_rate": 0.0021158854166666665,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 2850,
+    "0.6": 1330,
+    "0.7": 512,
+    "0.8": 127,
+    "0.9": 13
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 1213,
+    "0.6": 574,
+    "0.7": 231,
+    "0.8": 58,
+    "0.9": 3
+   },
+   "n_dup_removed_signed": {
+    "0.5": 1637,
+    "0.6": 756,
+    "0.7": 281,
+    "0.8": 69,
+    "0.9": 10
+   },
+   "largest_dup_signed": {
+    "0.5": 12,
+    "0.6": 11,
+    "0.7": 9,
+    "0.8": 9,
+    "0.9": 9
+   },
+   "babel_mu1_signed": 0.5303625464439392,
+   "n_dup_abs": {
+    "0.5": 2855,
+    "0.6": 1336,
+    "0.7": 522,
+    "0.8": 133,
+    "0.9": 13
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 1212,
+    "0.6": 575,
+    "0.7": 235,
+    "0.8": 61,
+    "0.9": 3
+   },
+   "n_dup_removed_abs": {
+    "0.5": 1643,
+    "0.6": 761,
+    "0.7": 287,
+    "0.8": 72,
+    "0.9": 10
+   },
+   "largest_dup_abs": {
+    "0.5": 13,
+    "0.6": 11,
+    "0.7": 9,
+    "0.8": 9,
+    "0.9": 9
+   },
+   "babel_mu1_abs": 0.5309417247772217,
+   "dup_excludes_dead": true
+  },
+  "seed2": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 29,
+   "l0_raw": 28.973237353515625,
+   "recon_r2": 0.8612391428325397,
+   "l1_per_token": 150.44482073044597,
+   "stats_source": "stats",
+   "n_dead": 7909,
+   "dead_rate": 0.6436360677083334,
+   "alive_source": "stats",
+   "n_dense": 33,
+   "dense_rate": 0.002685546875,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 1721,
+    "0.6": 801,
+    "0.7": 296,
+    "0.8": 74,
+    "0.9": 4
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 733,
+    "0.6": 350,
+    "0.7": 135,
+    "0.8": 35,
+    "0.9": 1
+   },
+   "n_dup_removed_signed": {
+    "0.5": 988,
+    "0.6": 451,
+    "0.7": 161,
+    "0.8": 39,
+    "0.9": 3
+   },
+   "largest_dup_signed": {
+    "0.5": 8,
+    "0.6": 8,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_signed": 0.5097309947013855,
+   "n_dup_abs": {
+    "0.5": 1724,
+    "0.6": 806,
+    "0.7": 305,
+    "0.8": 81,
+    "0.9": 4
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 731,
+    "0.6": 354,
+    "0.7": 138,
+    "0.8": 38,
+    "0.9": 1
+   },
+   "n_dup_removed_abs": {
+    "0.5": 993,
+    "0.6": 452,
+    "0.7": 167,
+    "0.8": 43,
+    "0.9": 3
+   },
+   "largest_dup_abs": {
+    "0.5": 8,
+    "0.6": 8,
+    "0.7": 5,
+    "0.8": 5,
+    "0.9": 4
+   },
+   "babel_mu1_abs": 0.510436475276947,
+   "dup_excludes_dead": true
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.42975470423698425,
+   "dcs_abs": 0.433308869600296,
+   "frac_sign_flipped": 0.2881673177083333,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 4459,
+   "n_alive_b": 6298,
+   "matched_cos_signed": {
+    "0.5": 0.9322665333747864,
+    "0.6": 0.9359939098358154,
+    "0.7": 0.939612627029419,
+    "0.8": 0.9405375719070435,
+    "0.9": 0.9416401386260986
+   },
+   "unique_frac_signed": {
+    "0.5": 0.329345703125,
+    "0.6": 0.3881022135416667,
+    "0.7": 0.4186197916666667,
+    "0.8": 0.4335530598958333,
+    "0.9": 0.4371337890625
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.3070379769342253,
+    "0.6": 0.36326130826879915,
+    "0.7": 0.39334044217442476,
+    "0.8": 0.40777294224729604,
+    "0.9": 0.41162272173096426
+   },
+   "matched_frac_signed": {
+    "0.5": 0.275390625,
+    "0.6": 0.3193359375,
+    "0.7": 0.3380533854166667,
+    "0.8": 0.3385416666666667,
+    "0.9": 0.2996419270833333
+   },
+   "matched_cos_abs": {
+    "0.5": 0.9320581555366516,
+    "0.6": 0.936193585395813,
+    "0.7": 0.9395206570625305,
+    "0.8": 0.9406428337097168,
+    "0.9": 0.941696286201477
+   },
+   "unique_frac_abs": {
+    "0.5": 0.3290201822916667,
+    "0.6": 0.3876546223958333,
+    "0.7": 0.4182535807291667,
+    "0.8": 0.4332275390625,
+    "0.9": 0.4371337890625
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.3066659442411037,
+    "0.6": 0.36291977083601523,
+    "0.7": 0.39295787898542284,
+    "0.8": 0.407512379984837,
+    "0.9": 0.4116472657333361
+   },
+   "matched_frac_abs": {
+    "0.5": 0.2752278645833333,
+    "0.6": 0.319091796875,
+    "0.7": 0.3377278645833333,
+    "0.8": 0.3382975260416667,
+    "0.9": 0.2996419270833333
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.4131103456020355,
+   "dcs_abs": 0.4168144464492798,
+   "frac_sign_flipped": 0.306640625,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 4459,
+   "n_alive_b": 4379,
+   "matched_cos_signed": {
+    "0.5": 0.9092767238616943,
+    "0.6": 0.9134637117385864,
+    "0.7": 0.9131386280059814,
+    "0.8": 0.9161136746406555,
+    "0.9": 0.9158425331115723
+   },
+   "unique_frac_signed": {
+    "0.5": 0.2776692708333333,
+    "0.6": 0.3224283854166667,
+    "0.7": 0.3454182942708333,
+    "0.8": 0.356689453125,
+    "0.9": 0.3593343098958333
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.25247820490039885,
+    "0.6": 0.29452662971258786,
+    "0.7": 0.3154147873186351,
+    "0.8": 0.3267680856079096,
+    "0.9": 0.32909364460889867
+   },
+   "matched_frac_signed": {
+    "0.5": 0.258544921875,
+    "0.6": 0.2958984375,
+    "0.7": 0.3104654947916667,
+    "0.8": 0.3098958333333333,
+    "0.9": 0.2903645833333333
+   },
+   "matched_cos_abs": {
+    "0.5": 0.9094918966293335,
+    "0.6": 0.9129810929298401,
+    "0.7": 0.9132170677185059,
+    "0.8": 0.9161067008972168,
+    "0.9": 0.915854275226593
+   },
+   "unique_frac_abs": {
+    "0.5": 0.2773844401041667,
+    "0.6": 0.3221435546875,
+    "0.7": 0.3450520833333333,
+    "0.8": 0.3563232421875,
+    "0.9": 0.3593343098958333
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.2522789005258043,
+    "0.6": 0.29411097463889746,
+    "0.7": 0.3151074517518282,
+    "0.8": 0.3264301098533906,
+    "0.9": 0.3290978639536964
+   },
+   "matched_frac_abs": {
+    "0.5": 0.2582194010416667,
+    "0.6": 0.2955729166666667,
+    "0.7": 0.3101399739583333,
+    "0.8": 0.3096516927083333,
+    "0.9": 0.2903645833333333
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.4248100817203522,
+   "dcs_abs": 0.42828086018562317,
+   "frac_sign_flipped": 0.2799479166666667,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 6298,
+   "n_alive_b": 4379,
+   "matched_cos_signed": {
+    "0.5": 0.9332655072212219,
+    "0.6": 0.9362735748291016,
+    "0.7": 0.9399933218955994,
+    "0.8": 0.9413590431213379,
+    "0.9": 0.9418835043907166
+   },
+   "unique_frac_signed": {
+    "0.5": 0.32763671875,
+    "0.6": 0.3853352864583333,
+    "0.7": 0.4164632161458333,
+    "0.8": 0.4300537109375,
+    "0.9": 0.4339192708333333
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.3057720485085156,
+    "0.6": 0.3607792461601396,
+    "0.7": 0.39147264199224685,
+    "0.8": 0.40483494981890544,
+    "0.9": 0.4087014034351644
+   },
+   "matched_frac_signed": {
+    "0.5": 0.2730305989583333,
+    "0.6": 0.3154296875,
+    "0.7": 0.3347981770833333,
+    "0.8": 0.3330891927083333,
+    "0.9": 0.2920735677083333
+   },
+   "matched_cos_abs": {
+    "0.5": 0.933249831199646,
+    "0.6": 0.9360746145248413,
+    "0.7": 0.940065860748291,
+    "0.8": 0.9414230585098267,
+    "0.9": 0.9418949484825134
+   },
+   "unique_frac_abs": {
+    "0.5": 0.3271891276041667,
+    "0.6": 0.3850911458333333,
+    "0.7": 0.4159749348958333,
+    "0.8": 0.4297688802083333,
+    "0.9": 0.4339192708333333
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.305349198106948,
+    "0.6": 0.3604740458928669,
+    "0.7": 0.39104383522256586,
+    "0.8": 0.40459433365807246,
+    "0.9": 0.4087063692471323
+   },
+   "matched_frac_abs": {
+    "0.5": 0.2726236979166667,
+    "0.6": 0.3153483072916667,
+    "0.7": 0.3343912760416667,
+    "0.8": 0.3329264322916667,
+    "0.9": 0.2920735677083333
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 7243,
+  "n_dead_std": 1085.5783404864585,
+  "n_dense_mean": 30,
+  "n_dense_std": 3.605551275463989,
+  "l0_mean": 30,
+  "l0_std": 0.9119928361663648,
+  "recon_r2_mean": 0.8647721289726477,
+  "recon_r2_std": 0.0051885099904022515,
+  "l1_per_token_mean": 151.0246016831567,
+  "l1_per_token_std": 0.5160333264721693,
+  "babel_mu1_signed_mean": 0.5170612136522929,
+  "babel_mu1_signed_std": 0.011539318966070562,
+  "n_dup_signed_per_threshold": {
+   "0.5": 2116.3333333333335,
+   "0.6": 985.3333333333334,
+   "0.7": 384.0,
+   "0.8": 87.33333333333333,
+   "0.9": 7.333333333333333
+  },
+  "n_dup_signed_band": 366.0,
+  "n_dup_signed_band_std": 112.40078958797397,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 899.3333333333334,
+   "0.6": 428.6666666666667,
+   "0.7": 174.0,
+   "0.8": 40.333333333333336,
+   "0.9": 1.6666666666666667
+  },
+  "n_dup_clusters_signed_band": 161.16666666666669,
+  "n_dup_clusters_signed_band_std": 48.030155458142474,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 9.333333333333334,
+   "0.6": 8.666666666666666,
+   "0.7": 6.0,
+   "0.8": 6.0,
+   "0.9": 6.0
+  },
+  "largest_dup_signed_band": 6.666666666666666,
+  "largest_dup_signed_band_std": 2.466441431158124,
+  "dcs_signed_mean": 0.42255837718645733,
+  "dcs_signed_std": 0.008547590220241614,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.2884294101143799,
+   "0.6": 0.3395223947138422,
+   "0.7": 0.36674262382843564,
+   "0.8": 0.37979199255803703,
+   "0.9": 0.3831392565916758
+  },
+  "dedup_dcs_signed_band": 0.3672990669229977,
+  "dedup_dcs_signed_band_std": 0.04405438959851286,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.9249362548192342,
+   "0.6": 0.9285770654678345,
+   "0.7": 0.9309148589769999,
+   "0.8": 0.9326700965563456,
+   "0.9": 0.9331220587094625
+  },
+  "matched_cos_signed_band": 0.9313210199276606,
+  "matched_cos_signed_band_std": 0.014448110955564015,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.3115505642361111,
+   "0.6": 0.36528862847222227,
+   "0.7": 0.39350043402777773,
+   "0.8": 0.4067654079861111,
+   "0.9": 0.41012912326388884
+  },
+  "unique_frac_signed_band": 0.3939208984375,
+  "unique_frac_signed_band_std": 0.04155423472577503,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.26898871527777773,
+   "0.6": 0.3102213541666667,
+   "0.7": 0.3277723524305556,
+   "0.8": 0.3271755642361111,
+   "0.9": 0.2940266927083333
+  },
+  "matched_frac_signed_band": 0.3147989908854167,
+  "matched_frac_signed_band_std": 0.01165831843350895,
+  "babel_mu1_abs_mean": 0.517728586991628,
+  "babel_mu1_abs_std": 0.011463429965174002,
+  "n_dup_abs_per_threshold": {
+   "0.5": 2119.3333333333335,
+   "0.6": 990.0,
+   "0.7": 391.6666666666667,
+   "0.8": 95.0,
+   "0.9": 7.333333333333333
+  },
+  "n_dup_abs_band": 371.0,
+  "n_dup_abs_band_std": 112.8007092176286,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 898.0,
+   "0.6": 429.3333333333333,
+   "0.7": 176.66666666666666,
+   "0.8": 44.0,
+   "0.9": 1.6666666666666667
+  },
+  "n_dup_clusters_abs_band": 162.91666666666666,
+  "n_dup_clusters_abs_band_std": 48.195132880129435,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 9.666666666666666,
+   "0.6": 8.666666666666666,
+   "0.7": 6.666666666666667,
+   "0.8": 6.333333333333333,
+   "0.9": 6.0
+  },
+  "largest_dup_abs_band": 6.916666666666666,
+  "largest_dup_abs_band_std": 2.2407216099581255,
+  "dcs_abs_mean": 0.426134725411733,
+  "dcs_abs_std": 0.008454047528963306,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.2880980142912854,
+   "0.6": 0.33916826378925985,
+   "0.7": 0.3663697219866056,
+   "0.8": 0.3795122744987667,
+   "0.9": 0.38315049964472153
+  },
+  "dedup_dcs_abs_band": 0.3670501899798384,
+  "dedup_dcs_abs_band_std": 0.044067677272827836,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.9249332944552103,
+   "0.6": 0.9284164309501648,
+   "0.7": 0.9309345285097758,
+   "0.8": 0.9327241977055868,
+   "0.9": 0.9331485033035278
+  },
+  "matched_cos_abs_band": 0.9313059151172638,
+  "matched_cos_abs_band_std": 0.014520957579692092,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.3111979166666667,
+   "0.6": 0.36496310763888884,
+   "0.7": 0.3930935329861111,
+   "0.8": 0.40643988715277773,
+   "0.9": 0.41012912326388884
+  },
+  "unique_frac_abs_band": 0.39365641276041663,
+  "unique_frac_abs_band_std": 0.041544898988204966,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.2686903211805556,
+   "0.6": 0.31000434027777785,
+   "0.7": 0.3274197048611111,
+   "0.8": 0.32695855034722227,
+   "0.9": 0.2940266927083333
+  },
+  "matched_frac_abs_band": 0.3146023220486111,
+  "matched_frac_abs_band_std": 0.011676878358008168,
+  "alive_source": "stats",
+  "frac_denominator": "d_sae",
+  "d_sae": 12288
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_32/seed0/config.json b/lm/gpt2-small_layer8/relu/x16/l0_32/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..5949eb30b783de244a80f491fa25f7af3f2edec1
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_32/seed0/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..0673a46a2a78eaf7f3374b22fe27a90edb211426
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 29.27993671875,
+  "l0_per_token_0.001": 29.27174404296875,
+  "l0_per_token_0.01": 29.198035009765626,
+  "l0_per_token_0.05": 28.873639794921875,
+  "l0_per_token_0.1": 28.4748041015625,
+  "l0_per_token_0.5": 25.53410654296875,
+  "l0_per_token_1.0": 22.393262939453123,
+  "l1_per_token": 151.43357586614445,
+  "r2_score": 0.8623482589771835,
+  "l2_loss": 3.2282683486657415
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_32/seed1/config.json b/lm/gpt2-small_layer8/relu/x16/l0_32/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..4c15515c994745470445ff31f3f2293062c96289
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_32/seed1/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..3472dabbc92ca8cc37bada4f6f3f7e4ceaa06a48
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 30.683714013671874,
+  "l0_per_token_0.001": 30.674758447265624,
+  "l0_per_token_0.01": 30.594301416015625,
+  "l0_per_token_0.05": 30.239803271484377,
+  "l0_per_token_0.1": 29.804076513671873,
+  "l0_per_token_0.5": 26.59156826171875,
+  "l0_per_token_1.0": 23.16467080078125,
+  "l1_per_token": 151.1954084528797,
+  "r2_score": 0.8707289851082196,
+  "l2_loss": 3.0317199235849768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_32/seed2/config.json b/lm/gpt2-small_layer8/relu/x16/l0_32/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..43913a5027e3e99b474c9dfef9490cb3c811038a
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_32/seed2/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..adcf07a5839564f985d6fe39419e8eb89c879c27
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 28.973237353515625,
+  "l0_per_token_0.001": 28.965130908203125,
+  "l0_per_token_0.01": 28.892264306640627,
+  "l0_per_token_0.05": 28.571207763671875,
+  "l0_per_token_0.1": 28.176348779296877,
+  "l0_per_token_0.5": 25.264233837890625,
+  "l0_per_token_1.0": 22.152956884765626,
+  "l1_per_token": 150.44482073044597,
+  "r2_score": 0.8612391428325397,
+  "l2_loss": 3.254279821663778
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_64/code_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x16/l0_64/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..78672b48c7ef958f0e6faf2e49e2562966e7ab73
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_64/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 64.00778,
+   "recon_r2": 0.9070009159364358,
+   "l1_per_token": 194.09144617253745
+  },
+  "seed1": {
+   "l0": 64.22994833333334,
+   "recon_r2": 0.9085648980479948,
+   "l1_per_token": 193.70289880431503
+  },
+  "seed2": {
+   "l0": 62.634231666666665,
+   "recon_r2": 0.9062843167738401,
+   "l1_per_token": 192.92770097972328
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7413015921112199,
+   "mcc_abs": 0.741302008876759,
+   "act_iou": 0.45311222926985867,
+   "act_iou_n_pairs": 11833,
+   "input_r2": 0.9853854030835436,
+   "matched_mcc_signed": {
+    "0.6": 0.8701732202817195,
+    "0.7": 0.897374955686745,
+    "0.8": 0.9227657960702103,
+    "0.9": 0.9498078845148318
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8695792226543788,
+    "0.7": 0.8972678111811346,
+    "0.8": 0.9225927833408925,
+    "0.9": 0.9498146668469215
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7173826499393762,
+   "mcc_abs": 0.7173828103623144,
+   "act_iou": 0.4842617249165674,
+   "act_iou_n_pairs": 10916,
+   "input_r2": 0.9858247099661006,
+   "matched_mcc_signed": {
+    "0.6": 0.8791904521435905,
+    "0.7": 0.9032050032370305,
+    "0.8": 0.9267740401060183,
+    "0.9": 0.9521196678665709
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8781009972282977,
+    "0.7": 0.9027396227249372,
+    "0.8": 0.9264890490222824,
+    "0.9": 0.9520694263403088
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7377152320786591,
+   "mcc_abs": 0.7377154672194087,
+   "act_iou": 0.45196611297763295,
+   "act_iou_n_pairs": 11822,
+   "input_r2": 0.985596256563419,
+   "matched_mcc_signed": {
+    "0.6": 0.8743551372256735,
+    "0.7": 0.8984131120729897,
+    "0.8": 0.9226123681639881,
+    "0.9": 0.9504539364897098
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8733831023861053,
+    "0.7": 0.8983648577580452,
+    "0.8": 0.9224730531952181,
+    "0.9": 0.9503424265790484
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 63.62398666666667,
+  "l0_std": 0.8643210700328798,
+  "recon_r2_mean": 0.9072833769194236,
+  "recon_r2_std": 0.001166233635335996,
+  "l1_per_token_mean": 193.5740153188586,
+  "l1_per_token_std": 0.5924811668458906,
+  "input_r2_mean": 0.9856021232043544,
+  "input_r2_std": 0.0002197121921341333,
+  "mcc_signed_mean": 0.7321331580430851,
+  "mcc_signed_std": 0.012899558580004732,
+  "mcc_abs_mean": 0.732133428819494,
+  "mcc_abs_std": 0.01289966584516604,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8745729365503278,
+   "0.7": 0.8996643569989217,
+   "0.8": 0.9240507347800722,
+   "0.9": 0.9507938296237041
+  },
+  "matched_mcc_signed_band": 0.9122704644882564,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.873687774089594,
+   "0.7": 0.8994574305547056,
+   "0.8": 0.9238516285194643,
+   "0.9": 0.9507421732554263
+  },
+  "matched_mcc_abs_band": 0.9119347516047976,
+  "act_iou_mean": 0.463113355721353,
+  "act_iou_std": 0.01832398797619246,
+  "act_iou_n_pairs": 11523.666666666666,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_64/dict_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x16/l0_64/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..586f11a8bb6f8db6e8d7fd6f154555e17dfa8503
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_64/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 64.02285834960938,
+   "recon_r2": 0.9068383111568028,
+   "l1_per_token": 194.0389601917935,
+   "stats_source": "stats",
+   "n_dead": 1481,
+   "dead_rate": 0.12052408854166667,
+   "alive_source": "stats",
+   "n_dense": 17,
+   "dense_rate": 0.0013834635416666667,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 4364,
+    "0.6": 2118,
+    "0.7": 803,
+    "0.8": 164,
+    "0.9": 8
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 1805,
+    "0.6": 916,
+    "0.7": 368,
+    "0.8": 78,
+    "0.9": 3
+   },
+   "largest_dup_signed": {
+    "0.5": 14,
+    "0.6": 7,
+    "0.7": 5,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_signed": 0.4702420234680176,
+   "n_dup_abs": {
+    "0.5": 4415,
+    "0.6": 2136,
+    "0.7": 818,
+    "0.8": 170,
+    "0.9": 8
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 1832,
+    "0.6": 926,
+    "0.7": 375,
+    "0.8": 81,
+    "0.9": 3
+   },
+   "largest_dup_abs": {
+    "0.5": 12,
+    "0.6": 7,
+    "0.7": 5,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_abs": 0.4728945791721344
+  },
+  "seed1": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 64.216864453125,
+   "recon_r2": 0.9083323175941816,
+   "l1_per_token": 193.50990386987058,
+   "stats_source": "stats",
+   "n_dead": 486,
+   "dead_rate": 0.03955078125,
+   "alive_source": "stats",
+   "n_dense": 15,
+   "dense_rate": 0.001220703125,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 4910,
+    "0.6": 2304,
+    "0.7": 917,
+    "0.8": 165,
+    "0.9": 8
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 2034,
+    "0.6": 985,
+    "0.7": 420,
+    "0.8": 79,
+    "0.9": 2
+   },
+   "largest_dup_signed": {
+    "0.5": 12,
+    "0.6": 10,
+    "0.7": 6,
+    "0.8": 6,
+    "0.9": 6
+   },
+   "babel_mu1_signed": 0.5022979974746704,
+   "n_dup_abs": {
+    "0.5": 4949,
+    "0.6": 2333,
+    "0.7": 938,
+    "0.8": 179,
+    "0.9": 8
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 2062,
+    "0.6": 999,
+    "0.7": 430,
+    "0.8": 86,
+    "0.9": 2
+   },
+   "largest_dup_abs": {
+    "0.5": 12,
+    "0.6": 10,
+    "0.7": 6,
+    "0.8": 6,
+    "0.9": 6
+   },
+   "babel_mu1_abs": 0.5045491456985474
+  },
+  "seed2": {
+   "d_sae": 12288,
+   "d_in": 768,
+   "l0": 63,
+   "l0_raw": 62.65385610351562,
+   "recon_r2": 0.9060908196811601,
+   "l1_per_token": 192.81446914775037,
+   "stats_source": "stats",
+   "n_dead": 1618,
+   "dead_rate": 0.13167317708333334,
+   "alive_source": "stats",
+   "n_dense": 17,
+   "dense_rate": 0.0013834635416666667,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 4287,
+    "0.6": 1989,
+    "0.7": 733,
+    "0.8": 137,
+    "0.9": 5
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 1776,
+    "0.6": 853,
+    "0.7": 337,
+    "0.8": 65,
+    "0.9": 2
+   },
+   "largest_dup_signed": {
+    "0.5": 11,
+    "0.6": 9,
+    "0.7": 6,
+    "0.8": 4,
+    "0.9": 3
+   },
+   "babel_mu1_signed": 0.4635713994503021,
+   "n_dup_abs": {
+    "0.5": 4322,
+    "0.6": 2001,
+    "0.7": 758,
+    "0.8": 141,
+    "0.9": 5
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 1795,
+    "0.6": 858,
+    "0.7": 349,
+    "0.8": 67,
+    "0.9": 2
+   },
+   "largest_dup_abs": {
+    "0.5": 10,
+    "0.6": 7,
+    "0.7": 6,
+    "0.8": 4,
+    "0.9": 3
+   },
+   "babel_mu1_abs": 0.4661963880062103
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.8120684623718262,
+   "dcs_abs": 0.8129291534423828,
+   "frac_sign_flipped": 0.056559244791666664,
+   "alive_source": "stats",
+   "n_alive_a": 10807,
+   "n_alive_b": 11802,
+   "matched_cos_signed": {
+    "0.5": 0.8954442143440247,
+    "0.6": 0.9013888239860535,
+    "0.7": 0.9040122628211975,
+    "0.8": 0.9062672257423401,
+    "0.9": 0.9069611430168152
+   },
+   "unique_frac_signed": {
+    "0.5": 0.7596974656110399,
+    "0.6": 0.8884957317882259,
+    "0.7": 0.9587774779954885,
+    "0.8": 0.9923924101021717,
+    "0.9": 0.9995134680879296
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.6802667002332243,
+    "0.6": 0.800880122793217,
+    "0.7": 0.8667465974247025,
+    "0.8": 0.8993727163510498,
+    "0.9": 0.9065198774777297
+   },
+   "matched_frac_signed": {
+    "0.5": 0.693263744526516,
+    "0.6": 0.7916316511123889,
+    "0.7": 0.8240081383519837,
+    "0.8": 0.8018930514396921,
+    "0.9": 0.7139634658764209
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8962576389312744,
+    "0.6": 0.9012293219566345,
+    "0.7": 0.9040499329566956,
+    "0.8": 0.906204879283905,
+    "0.9": 0.9069899916648865
+   },
+   "unique_frac_abs": {
+    "0.5": 0.7581494095271795,
+    "0.6": 0.8874784377902605,
+    "0.7": 0.9579371046928215,
+    "0.8": 0.9919501083639258,
+    "0.9": 0.9995134680879296
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.6794971997399697,
+    "0.6": 0.7998215907408497,
+    "0.7": 0.8660229752742763,
+    "0.8": 0.898910028205588,
+    "0.9": 0.9065487120900131
+   },
+   "matched_frac_abs": {
+    "0.5": 0.6920252996594276,
+    "0.6": 0.7904816665929497,
+    "0.7": 0.8233004555707903,
+    "0.8": 0.8015392100490955,
+    "0.9": 0.7139634658764209
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.7891480326652527,
+   "dcs_abs": 0.7901210784912109,
+   "frac_sign_flipped": 0.064697265625,
+   "alive_source": "stats",
+   "n_alive_a": 10807,
+   "n_alive_b": 10670,
+   "matched_cos_signed": {
+    "0.5": 0.8798726797103882,
+    "0.6": 0.8831486701965332,
+    "0.7": 0.8870993256568909,
+    "0.8": 0.8895934224128723,
+    "0.9": 0.8907148838043213
+   },
+   "unique_frac_signed": {
+    "0.5": 0.7640266331424315,
+    "0.6": 0.8911393583833869,
+    "0.7": 0.9613540066117242,
+    "0.8": 0.9926898542626996,
+    "0.9": 0.9996275084974624
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.6722461610731368,
+    "0.6": 0.78700853931608,
+    "0.7": 0.8528164909828108,
+    "0.8": 0.8830903648480904,
+    "0.9": 0.8903831000789204
+   },
+   "matched_frac_signed": {
+    "0.5": 0.7005633933975881,
+    "0.6": 0.7977836755599013,
+    "0.7": 0.833822228430414,
+    "0.8": 0.8152907761791685,
+    "0.9": 0.7312008194813056
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8798260688781738,
+    "0.6": 0.8831661343574524,
+    "0.7": 0.8873321413993835,
+    "0.8": 0.8896431922912598,
+    "0.9": 0.8907793760299683
+   },
+   "unique_frac_abs": {
+    "0.5": 0.7621641756297435,
+    "0.6": 0.8904409368161289,
+    "0.7": 0.960376216417563,
+    "0.8": 0.9924570470736136,
+    "0.9": 0.9996275084974624
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.6705719104840913,
+    "0.6": 0.786407280041529,
+    "0.7": 0.852172684662834,
+    "0.8": 0.8829326555705267,
+    "0.9": 0.8904475682817613
+   },
+   "matched_frac_abs": {
+    "0.5": 0.6989803045118034,
+    "0.6": 0.7975043069329981,
+    "0.7": 0.8331703683009731,
+    "0.8": 0.815197653303534,
+    "0.9": 0.7313870652325744
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.8067523837089539,
+   "dcs_abs": 0.8077052235603333,
+   "frac_sign_flipped": 0.061767578125,
+   "alive_source": "stats",
+   "n_alive_a": 11802,
+   "n_alive_b": 10670,
+   "matched_cos_signed": {
+    "0.5": 0.8993087410926819,
+    "0.6": 0.9042173624038696,
+    "0.7": 0.9068920016288757,
+    "0.8": 0.9099218249320984,
+    "0.9": 0.9108036756515503
+   },
+   "unique_frac_signed": {
+    "0.5": 0.7603684585261659,
+    "0.6": 0.8907529369882521,
+    "0.7": 0.9603061587753648,
+    "0.8": 0.9930135279458883,
+    "0.9": 0.9995995016019936
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.6838060012037493,
+    "0.6": 0.8054342712370176,
+    "0.7": 0.8708939745083276,
+    "0.8": 0.903564681530784,
+    "0.9": 0.9104389002385536
+   },
+   "matched_frac_signed": {
+    "0.5": 0.6894802420790317,
+    "0.6": 0.7935208259166964,
+    "0.7": 0.8277856888572446,
+    "0.8": 0.8092737629049483,
+    "0.9": 0.7168031327874689
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8997672200202942,
+    "0.6": 0.9041493535041809,
+    "0.7": 0.9070651531219482,
+    "0.8": 0.9098820090293884,
+    "0.9": 0.9108443260192871
+   },
+   "unique_frac_abs": {
+    "0.5": 0.7591669633321466,
+    "0.6": 0.8897739409042363,
+    "0.7": 0.9592381630473478,
+    "0.8": 0.9926130295478818,
+    "0.9": 0.9995995016019936
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.6830735481286142,
+    "0.6": 0.8044885334334325,
+    "0.7": 0.8700915112449589,
+    "0.8": 0.9031607375137743,
+    "0.9": 0.9104795343258831
+   },
+   "matched_frac_abs": {
+    "0.5": 0.6885902456390175,
+    "0.6": 0.7928978284086864,
+    "0.7": 0.826628693485226,
+    "0.8": 0.8090067639729441,
+    "0.9": 0.7168921324314703
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 1195,
+  "n_dead_std": 617.8211715375251,
+  "n_dense_mean": 16,
+  "n_dense_std": 1.1547005383792515,
+  "l0_mean": 64,
+  "l0_std": 0.8519390254941654,
+  "recon_r2_mean": 0.9070871494773814,
+  "recon_r2_std": 0.0008634067003435642,
+  "l1_per_token_mean": 193.45444440313815,
+  "l1_per_token_std": 0.6141265289605597,
+  "babel_mu1_signed_mean": 0.4787038067976634,
+  "babel_mu1_signed_std": 0.020703591028458205,
+  "n_dup_signed_per_threshold": {
+   "0.5": 4520.333333333333,
+   "0.6": 2137.0,
+   "0.7": 817.6666666666666,
+   "0.8": 155.33333333333334,
+   "0.9": 7.0
+  },
+  "n_dup_signed_band": 779.25,
+  "n_dup_signed_band_std": 66.45346115891932,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 1871.6666666666667,
+   "0.6": 918.0,
+   "0.7": 375.0,
+   "0.8": 74.0,
+   "0.9": 2.3333333333333335
+  },
+  "n_dup_clusters_signed_band": 342.3333333333333,
+  "n_dup_clusters_signed_band_std": 28.640370691269577,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 12.333333333333334,
+   "0.6": 8.666666666666666,
+   "0.7": 5.666666666666667,
+   "0.8": 4.666666666666667,
+   "0.9": 4.333333333333333
+  },
+  "largest_dup_signed_band": 5.833333333333333,
+  "largest_dup_signed_band_std": 1.0408329997330663,
+  "dcs_signed_mean": 0.8026562929153442,
+  "dcs_signed_std": 0.011996665967076612,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.6787729541700368,
+   "0.6": 0.7977743111154382,
+   "0.7": 0.8634856876386136,
+   "0.8": 0.8953425875766414,
+   "0.9": 0.9024472925984012
+  },
+  "dedup_dcs_signed_band": 0.8647624697322736,
+  "dedup_dcs_signed_band_std": 0.010125947441305728,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.8915418783823649,
+   "0.6": 0.8962516188621521,
+   "0.7": 0.8993345300356547,
+   "0.8": 0.9019274910291036,
+   "0.9": 0.9028265674908956
+  },
+  "matched_cos_signed_band": 0.9000850518544515,
+  "matched_cos_signed_band_std": 0.01090419538170323,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.761364185759879,
+   "0.6": 0.8901293423866216,
+   "0.7": 0.9601458811275259,
+   "0.8": 0.9926985974369199,
+   "0.9": 0.9995801593957951
+  },
+  "unique_frac_signed_band": 0.9606384950867156,
+  "unique_frac_signed_band_std": 0.0007444178840912974,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.6944357933343785,
+   "0.6": 0.7943120508629956,
+   "0.7": 0.8285386852132142,
+   "0.8": 0.8088191968412697,
+   "0.9": 0.7206558060483985
+  },
+  "matched_frac_signed_band": 0.7880814347414695,
+  "matched_frac_signed_band_std": 0.0059226142479571545,
+  "babel_mu1_abs_mean": 0.48121337095896405,
+  "babel_mu1_abs_std": 0.020485000079234744,
+  "n_dup_abs_per_threshold": {
+   "0.5": 4562.0,
+   "0.6": 2156.6666666666665,
+   "0.7": 838.0,
+   "0.8": 163.33333333333334,
+   "0.9": 7.0
+  },
+  "n_dup_abs_band": 791.25,
+  "n_dup_abs_band_std": 69.49325506838775,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 1896.3333333333333,
+   "0.6": 927.6666666666666,
+   "0.7": 384.6666666666667,
+   "0.8": 78.0,
+   "0.9": 2.3333333333333335
+  },
+  "n_dup_clusters_abs_band": 348.16666666666663,
+  "n_dup_clusters_abs_band_std": 30.170694942830426,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 11.333333333333334,
+   "0.6": 8.0,
+   "0.7": 5.666666666666667,
+   "0.8": 4.666666666666667,
+   "0.9": 4.333333333333333
+  },
+  "largest_dup_abs_band": 5.666666666666667,
+  "largest_dup_abs_band_std": 1.1547005383792517,
+  "dcs_abs_mean": 0.803585151831309,
+  "dcs_abs_std": 0.01194919721294268,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.6777142194508917,
+   "0.6": 0.7969058014052705,
+   "0.7": 0.862762390394023,
+   "0.8": 0.8950011404299629,
+   "0.9": 0.9024919382325525
+  },
+  "dedup_dcs_abs_band": 0.8642903176154522,
+  "dedup_dcs_abs_band_std": 0.01001217902930059,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.8919503092765808,
+   "0.6": 0.896181603272756,
+   "0.7": 0.8994824091593424,
+   "0.8": 0.9019100268681844,
+   "0.9": 0.9028712312380472
+  },
+  "matched_cos_abs_band": 0.9001113176345825,
+  "matched_cos_abs_band_std": 0.010853685206125332,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.7598268494963566,
+   "0.6": 0.8892311051702085,
+   "0.7": 0.9591838280525774,
+   "0.8": 0.9923400616618071,
+   "0.9": 0.9995801593957951
+  },
+  "unique_frac_abs_band": 0.960083788570097,
+  "unique_frac_abs_band_std": 0.0007770649620260262,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.6931986166034162,
+   "0.6": 0.7936279339782114,
+   "0.7": 0.8276998391189965,
+   "0.8": 0.8085812091085245,
+   "0.9": 0.7207475545134886
+  },
+  "matched_frac_abs_band": 0.7876641341798053,
+  "matched_frac_abs_band_std": 0.006102836981860767,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_64/seed0/config.json b/lm/gpt2-small_layer8/relu/x16/l0_64/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..5949eb30b783de244a80f491fa25f7af3f2edec1
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_64/seed0/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..f176d473b05b1ca254610cbac8327d40f3f9dbab
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.02285834960938,
+  "l0_per_token_0.001": 63.992184521484376,
+  "l0_per_token_0.01": 63.716652490234374,
+  "l0_per_token_0.05": 62.50953608398437,
+  "l0_per_token_0.1": 61.038549365234374,
+  "l0_per_token_0.5": 50.63392524414063,
+  "l0_per_token_1.0": 40.44646479492187,
+  "l1_per_token": 194.0389601917935,
+  "r2_score": 0.9068383111568028,
+  "l2_loss": 2.1848683435895557
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_64/seed1/config.json b/lm/gpt2-small_layer8/relu/x16/l0_64/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..4c15515c994745470445ff31f3f2293062c96289
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_64/seed1/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..4f98342c2a1126d2228190f70189546492bda969
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.216864453125,
+  "l0_per_token_0.001": 64.18581586914063,
+  "l0_per_token_0.01": 63.90680434570312,
+  "l0_per_token_0.05": 62.68320131835937,
+  "l0_per_token_0.1": 61.19266044921875,
+  "l0_per_token_0.5": 50.6585111328125,
+  "l0_per_token_1.0": 40.365172900390625,
+  "l1_per_token": 193.50990386987058,
+  "r2_score": 0.9083323175941816,
+  "l2_loss": 2.1498302564672613
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_64/seed2/config.json b/lm/gpt2-small_layer8/relu/x16/l0_64/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..43913a5027e3e99b474c9dfef9490cb3c811038a
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_64/seed2/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..19a4950bbd760778d0ac2b143da05aef24c30a27
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 62.65385610351562,
+  "l0_per_token_0.001": 62.6241310546875,
+  "l0_per_token_0.01": 62.3569228515625,
+  "l0_per_token_0.05": 61.1860392578125,
+  "l0_per_token_0.1": 59.75935390625,
+  "l0_per_token_0.5": 49.66794755859375,
+  "l0_per_token_1.0": 39.78678276367187,
+  "l1_per_token": 192.81446914775037,
+  "r2_score": 0.9060908196811601,
+  "l2_loss": 2.2023988379645947
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_128/code_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x4/l0_128/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..fbb744895bc9ec3f21420c213c10377480d5a921
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_128/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 133.16802833333333,
+   "recon_r2": 0.9133065951988699,
+   "l1_per_token": 306.1630905285732
+  },
+  "seed1": {
+   "l0": 134.53975666666668,
+   "recon_r2": 0.913789340776904,
+   "l1_per_token": 308.21883865772713
+  },
+  "seed2": {
+   "l0": 133.45174666666668,
+   "recon_r2": 0.9134445223015636,
+   "l1_per_token": 306.64460396675514
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.8345115611911792,
+   "mcc_abs": 0.8345331122044525,
+   "act_iou": 0.5311803065336508,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9852496075998013,
+   "matched_mcc_signed": {
+    "0.6": 0.9056615132316653,
+    "0.7": 0.9228323539510618,
+    "0.8": 0.9399366422557245,
+    "0.9": 0.9610319604219176
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.9023046481598266,
+    "0.7": 0.9195717607405761,
+    "0.8": 0.9392254421771719,
+    "0.9": 0.9610527635883465
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.8422055023896461,
+   "mcc_abs": 0.8422055023896461,
+   "act_iou": 0.5415594008908371,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9857822186645041,
+   "matched_mcc_signed": {
+    "0.6": 0.9120157537863872,
+    "0.7": 0.9279242202938907,
+    "0.8": 0.9432220770115621,
+    "0.9": 0.9624774091938259
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.910050987775504,
+    "0.7": 0.9262865478543265,
+    "0.8": 0.9422185836926738,
+    "0.9": 0.962407970786599
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.8392304246742458,
+   "mcc_abs": 0.8392308072283855,
+   "act_iou": 0.536867755792291,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9856121413892439,
+   "matched_mcc_signed": {
+    "0.6": 0.9118035895462198,
+    "0.7": 0.9230261145207099,
+    "0.8": 0.9415464066875887,
+    "0.9": 0.9615243245128982
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.9096337727774431,
+    "0.7": 0.9209208208052296,
+    "0.8": 0.9404706570074107,
+    "0.9": 0.9614281717685015
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 133.7198438888889,
+  "l0_std": 0.7240971934345719,
+  "recon_r2_mean": 0.9135134860924459,
+  "recon_r2_std": 0.000248652019130149,
+  "l1_per_token_mean": 307.0088443843518,
+  "l1_per_token_std": 1.0751876133904366,
+  "input_r2_mean": 0.9855479892178498,
+  "input_r2_std": 0.0002720390824530623,
+  "mcc_signed_mean": 0.8386491627516904,
+  "mcc_signed_std": 0.0038797656963180356,
+  "mcc_abs_mean": 0.8386564739408281,
+  "mcc_abs_std": 0.0038683054212873694,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.9098269521880908,
+   "0.7": 0.9245942295885542,
+   "0.8": 0.9415683753182917,
+   "0.9": 0.9616778980428805
+  },
+  "matched_mcc_signed_band": 0.9344168637844543,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.9073298029042579,
+   "0.7": 0.9222597098000441,
+   "0.8": 0.940638227625752,
+   "0.9": 0.961629635381149
+  },
+  "matched_mcc_abs_band": 0.9329643439278007,
+  "act_iou_mean": 0.5365358210722596,
+  "act_iou_std": 0.0051975028054456985,
+  "act_iou_n_pairs": 3072.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_128/dict_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x4/l0_128/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..7cb37d2bf6b4dfa52d748717f06fda7575e0e60c
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_128/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 133,
+   "l0_raw": 133.22763266601564,
+   "recon_r2": 0.9131796719309067,
+   "l1_per_token": 306.134578111358,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 159,
+   "dense_rate": 0.0517578125,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 321,
+    "0.6": 137,
+    "0.7": 47,
+    "0.8": 6,
+    "0.9": 4
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 131,
+    "0.6": 58,
+    "0.7": 22,
+    "0.8": 2,
+    "0.9": 1
+   },
+   "largest_dup_signed": {
+    "0.5": 6,
+    "0.6": 5,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_signed": 0.3476353585720062,
+   "n_dup_abs": {
+    "0.5": 495,
+    "0.6": 280,
+    "0.7": 136,
+    "0.8": 62,
+    "0.9": 16
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 217,
+    "0.6": 129,
+    "0.7": 66,
+    "0.8": 30,
+    "0.9": 7
+   },
+   "largest_dup_abs": {
+    "0.5": 6,
+    "0.6": 5,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_abs": 0.38312283158302307
+  },
+  "seed1": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 135,
+   "l0_raw": 134.5747197265625,
+   "recon_r2": 0.9136504611419225,
+   "l1_per_token": 308.1257509002589,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 163,
+   "dense_rate": 0.053059895833333336,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 316,
+    "0.6": 135,
+    "0.7": 43,
+    "0.8": 8,
+    "0.9": 4
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 131,
+    "0.6": 59,
+    "0.7": 20,
+    "0.8": 3,
+    "0.9": 1
+   },
+   "largest_dup_signed": {
+    "0.5": 7,
+    "0.6": 5,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_signed": 0.3464624881744385,
+   "n_dup_abs": {
+    "0.5": 504,
+    "0.6": 284,
+    "0.7": 156,
+    "0.8": 72,
+    "0.9": 22
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 220,
+    "0.6": 131,
+    "0.7": 76,
+    "0.8": 35,
+    "0.9": 10
+   },
+   "largest_dup_abs": {
+    "0.5": 7,
+    "0.6": 6,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_abs": 0.38377127051353455
+  },
+  "seed2": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 134,
+   "l0_raw": 133.52997338867186,
+   "recon_r2": 0.9133321926177452,
+   "l1_per_token": 306.66340677926985,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 154,
+   "dense_rate": 0.050130208333333336,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 313,
+    "0.6": 131,
+    "0.7": 50,
+    "0.8": 6,
+    "0.9": 4
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 127,
+    "0.6": 57,
+    "0.7": 23,
+    "0.8": 2,
+    "0.9": 1
+   },
+   "largest_dup_signed": {
+    "0.5": 7,
+    "0.6": 5,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_signed": 0.34607362747192383,
+   "n_dup_abs": {
+    "0.5": 488,
+    "0.6": 276,
+    "0.7": 150,
+    "0.8": 58,
+    "0.9": 18
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 210,
+    "0.6": 128,
+    "0.7": 73,
+    "0.8": 28,
+    "0.9": 8
+   },
+   "largest_dup_abs": {
+    "0.5": 7,
+    "0.6": 5,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_abs": 0.38221311569213867
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.8671703338623047,
+   "dcs_abs": 0.8676190972328186,
+   "frac_sign_flipped": 0.009765625,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.8655097484588623,
+    "0.6": 0.8661993741989136,
+    "0.7": 0.8663203716278076,
+    "0.8": 0.8672842979431152,
+    "0.9": 0.8670409917831421
+   },
+   "unique_frac_signed": {
+    "0.5": 0.93896484375,
+    "0.6": 0.9747721354166666,
+    "0.7": 0.9921875,
+    "0.8": 0.99853515625,
+    "0.9": 0.9990234375
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.8126832257257774,
+    "0.6": 0.8443470136844553,
+    "0.7": 0.8595522437244654,
+    "0.8": 0.8660138619598001,
+    "0.9": 0.8661942720646039
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8564453125,
+    "0.6": 0.8645833333333334,
+    "0.7": 0.84375,
+    "0.8": 0.7958984375,
+    "0.9": 0.6920572916666666
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8723536729812622,
+    "0.6": 0.8704424500465393,
+    "0.7": 0.8712881803512573,
+    "0.8": 0.8702430129051208,
+    "0.9": 0.8680124878883362
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9085286458333334,
+    "0.6": 0.9505208333333334,
+    "0.7": 0.9755859375,
+    "0.8": 0.98876953125,
+    "0.9": 0.99658203125
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7925583012014007,
+    "0.6": 0.8273736829869449,
+    "0.7": 0.8500164962606505,
+    "0.8": 0.860469775943784,
+    "0.9": 0.8650456483301241
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8349609375,
+    "0.6": 0.8473307291666666,
+    "0.7": 0.8372395833333334,
+    "0.8": 0.7926432291666666,
+    "0.9": 0.6920572916666666
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.873558521270752,
+   "dcs_abs": 0.8739964365959167,
+   "frac_sign_flipped": 0.008138020833333334,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.8712841868400574,
+    "0.6": 0.8729472160339355,
+    "0.7": 0.8731625080108643,
+    "0.8": 0.8733981251716614,
+    "0.9": 0.873435378074646
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9388020833333334,
+    "0.6": 0.97509765625,
+    "0.7": 0.9915364583333334,
+    "0.8": 0.9986979166666666,
+    "0.9": 0.9990234375
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.8179634097808351,
+    "0.6": 0.851208784384653,
+    "0.7": 0.8657724607425432,
+    "0.8": 0.8722608880295107,
+    "0.9": 0.872582413838245
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8603515625,
+    "0.6": 0.8665364583333334,
+    "0.7": 0.8483072916666666,
+    "0.8": 0.806640625,
+    "0.9": 0.7125651041666666
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8767913579940796,
+    "0.6": 0.8779126405715942,
+    "0.7": 0.8777647018432617,
+    "0.8": 0.8762246966362,
+    "0.9": 0.8745048642158508
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9095052083333334,
+    "0.6": 0.9513346354166666,
+    "0.7": 0.97607421875,
+    "0.8": 0.9899088541666666,
+    "0.9": 0.9969075520833334
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7974463067172717,
+    "0.6": 0.8351887018458607,
+    "0.7": 0.8567634955979884,
+    "0.8": 0.8673825854396757,
+    "0.9": 0.8718005034703917
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8375651041666666,
+    "0.6": 0.8518880208333334,
+    "0.7": 0.841796875,
+    "0.8": 0.8046875,
+    "0.9": 0.7122395833333334
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.869877815246582,
+   "dcs_abs": 0.8705165982246399,
+   "frac_sign_flipped": 0.0126953125,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.8655919432640076,
+    "0.6": 0.8684747815132141,
+    "0.7": 0.8698001503944397,
+    "0.8": 0.8699907660484314,
+    "0.9": 0.8697513937950134
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9396158854166666,
+    "0.6": 0.9755859375,
+    "0.7": 0.9918619791666666,
+    "0.8": 0.99853515625,
+    "0.9": 0.9990234375
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.8133239401795436,
+    "0.6": 0.8472717839176767,
+    "0.7": 0.8627216986496933,
+    "0.8": 0.8687163655122276,
+    "0.9": 0.8689020271995105
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8512369791666666,
+    "0.6": 0.8639322916666666,
+    "0.7": 0.853515625,
+    "0.8": 0.8046875,
+    "0.9": 0.7027994791666666
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8728061318397522,
+    "0.6": 0.8734996318817139,
+    "0.7": 0.8732595443725586,
+    "0.8": 0.8731129765510559,
+    "0.9": 0.8710984587669373
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9085286458333334,
+    "0.6": 0.9510091145833334,
+    "0.7": 0.9744466145833334,
+    "0.8": 0.9890950520833334,
+    "0.9": 0.9964192708333334
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7929693730353999,
+    "0.6": 0.8307061115046963,
+    "0.7": 0.8509448066664239,
+    "0.8": 0.8635917250164009,
+    "0.9": 0.8679792911085922
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8268229166666666,
+    "0.6": 0.84765625,
+    "0.7": 0.8447265625,
+    "0.8": 0.8017578125,
+    "0.9": 0.7024739583333334
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 159,
+  "n_dense_std": 4.509249752822894,
+  "l0_mean": 134,
+  "l0_std": 0.7068177949803144,
+  "recon_r2_mean": 0.913387441896858,
+  "recon_r2_std": 0.0001817235619613366,
+  "l1_per_token_mean": 306.97457859696226,
+  "l1_per_token_std": 1.0314131991709323,
+  "babel_mu1_signed_mean": 0.3467238247394562,
+  "babel_mu1_signed_std": 0.0008130028335759856,
+  "n_dup_signed_per_threshold": {
+   "0.5": 316.6666666666667,
+   "0.6": 134.33333333333334,
+   "0.7": 46.666666666666664,
+   "0.8": 6.666666666666667,
+   "0.9": 4.0
+  },
+  "n_dup_signed_band": 47.916666666666664,
+  "n_dup_signed_band_std": 0.5204164998665333,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 129.66666666666666,
+   "0.6": 58.0,
+   "0.7": 21.666666666666668,
+   "0.8": 2.3333333333333335,
+   "0.9": 1.0
+  },
+  "n_dup_clusters_signed_band": 20.75,
+  "n_dup_clusters_signed_band_std": 0.0,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 6.666666666666667,
+   "0.6": 5.0,
+   "0.7": 4.0,
+   "0.8": 4.0,
+   "0.9": 4.0
+  },
+  "largest_dup_signed_band": 4.25,
+  "largest_dup_signed_band_std": 0.0,
+  "dcs_signed_mean": 0.8702022234598795,
+  "dcs_signed_std": 0.0032064255968293136,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.814656858562052,
+   "0.6": 0.847609193995595,
+   "0.7": 0.8626821343722341,
+   "0.8": 0.8689970385005128,
+   "0.9": 0.8692262377007864
+  },
+  "dedup_dcs_signed_band": 0.8621286511422821,
+  "dedup_dcs_signed_band_std": 0.0032205804278259607,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.8674619595209757,
+   "0.6": 0.8692071239153544,
+   "0.7": 0.8697610100110372,
+   "0.8": 0.870224396387736,
+   "0.9": 0.8700759212176005
+  },
+  "matched_cos_signed_band": 0.869817112882932,
+  "matched_cos_signed_band_std": 0.0032735047061164487,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.9391276041666666,
+   "0.6": 0.9751519097222222,
+   "0.7": 0.9918619791666666,
+   "0.8": 0.9985894097222222,
+   "0.9": 0.9990234375
+  },
+  "unique_frac_signed_band": 0.9911566840277778,
+  "unique_frac_signed_band_std": 8.470320635847041e-05,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.8560112847222222,
+   "0.6": 0.8650173611111112,
+   "0.7": 0.8485243055555555,
+   "0.8": 0.8024088541666666,
+   "0.9": 0.7024739583333334
+  },
+  "matched_frac_signed_band": 0.8046061197916666,
+  "matched_frac_signed_band_std": 0.0049260240724028235,
+  "babel_mu1_abs_mean": 0.38303573926289874,
+  "babel_mu1_abs_std": 0.0007827198835012934,
+  "n_dup_abs_per_threshold": {
+   "0.5": 495.6666666666667,
+   "0.6": 280.0,
+   "0.7": 147.33333333333334,
+   "0.8": 64.0,
+   "0.9": 18.666666666666668
+  },
+  "n_dup_abs_band": 127.50000000000001,
+  "n_dup_abs_band_std": 5.291502622129181,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 215.66666666666666,
+   "0.6": 129.33333333333334,
+   "0.7": 71.66666666666667,
+   "0.8": 31.0,
+   "0.9": 8.333333333333334
+  },
+  "n_dup_clusters_abs_band": 60.083333333333336,
+  "n_dup_clusters_abs_band_std": 2.6020824993326657,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 6.666666666666667,
+   "0.6": 5.333333333333333,
+   "0.7": 4.0,
+   "0.8": 4.0,
+   "0.9": 4.0
+  },
+  "largest_dup_abs_band": 4.333333333333333,
+  "largest_dup_abs_band_std": 0.14433756729740646,
+  "dcs_abs_mean": 0.8707107106844584,
+  "dcs_abs_std": 0.003193097880322941,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.7943246603180242,
+   "0.6": 0.8310894987791673,
+   "0.7": 0.8525749328416876,
+   "0.8": 0.8638146954666203,
+   "0.9": 0.8682751476363694
+  },
+  "dedup_dcs_abs_band": 0.8539385686809612,
+  "dedup_dcs_abs_band_std": 0.003571049471409843,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.8739837209383646,
+   "0.6": 0.8739515741666158,
+   "0.7": 0.8741041421890259,
+   "0.8": 0.8731935620307922,
+   "0.9": 0.8712052702903748
+  },
+  "matched_cos_abs_band": 0.8731136371692022,
+  "matched_cos_abs_band_std": 0.003318187111078899,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.9088541666666666,
+   "0.6": 0.9509548611111112,
+   "0.7": 0.9753689236111112,
+   "0.8": 0.9892578125,
+   "0.9": 0.9966362847222223
+  },
+  "unique_frac_abs_band": 0.9780544704861112,
+  "unique_frac_abs_band_std": 0.0004388750457726203,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.8331163194444443,
+   "0.6": 0.8489583333333334,
+   "0.7": 0.8412543402777778,
+   "0.8": 0.7996961805555555,
+   "0.9": 0.7022569444444445
+  },
+  "matched_frac_abs_band": 0.7980414496527778,
+  "matched_frac_abs_band_std": 0.005256640733584964,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_128/seed0/config.json b/lm/gpt2-small_layer8/relu/x4/l0_128/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..920481756aae71a20da4a01e54beb062205c045a
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_128/seed0/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..842b44715d0031d9c9b9e9f60a80fc3a9e924982
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 133.22763266601564,
+  "l0_per_token_0.001": 133.14381337890626,
+  "l0_per_token_0.01": 132.39021098632813,
+  "l0_per_token_0.05": 129.1008873046875,
+  "l0_per_token_0.1": 125.12221918945312,
+  "l0_per_token_0.5": 98.03541508789063,
+  "l0_per_token_1.0": 73.58695122070313,
+  "l1_per_token": 306.134578111358,
+  "r2_score": 0.9131796719309067,
+  "l2_loss": 2.0361479996084615
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_128/seed1/config.json b/lm/gpt2-small_layer8/relu/x4/l0_128/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..2b729114fa7a39fd6d4550e7ebd1e771051ed38a
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_128/seed1/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..76a7acac4ac8a2af4256a9627f81e2fed3190cb2
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 134.5747197265625,
+  "l0_per_token_0.001": 134.48992202148438,
+  "l0_per_token_0.01": 133.72779145507812,
+  "l0_per_token_0.05": 130.40123681640625,
+  "l0_per_token_0.1": 126.37799912109375,
+  "l0_per_token_0.5": 98.97656630859375,
+  "l0_per_token_1.0": 74.22664829101562,
+  "l1_per_token": 308.1257509002589,
+  "r2_score": 0.9136504611419225,
+  "l2_loss": 2.02510684678669
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_128/seed2/config.json b/lm/gpt2-small_layer8/relu/x4/l0_128/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..d2ac614f301c676dc8383e32f921621c96254746
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_128/seed2/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..15b2baaec39633c8edf9cd6e99fbdfe92442117a
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 133.52997338867186,
+  "l0_per_token_0.001": 133.4454296875,
+  "l0_per_token_0.01": 132.687690234375,
+  "l0_per_token_0.05": 129.37985991210937,
+  "l0_per_token_0.1": 125.37955751953125,
+  "l0_per_token_0.5": 98.15329677734375,
+  "l0_per_token_1.0": 73.600826953125,
+  "l1_per_token": 306.66340677926985,
+  "r2_score": 0.9133321926177452,
+  "l2_loss": 2.0325710182918484
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_32/code_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x4/l0_32/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..a5b70c0eba3b3076374252fa7868cd2a85673f3c
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_32/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 30.777418333333333,
+   "recon_r2": 0.8576081711569581,
+   "l1_per_token": 159.0001499288626
+  },
+  "seed1": {
+   "l0": 30.650196666666666,
+   "recon_r2": 0.8575394643015779,
+   "l1_per_token": 158.877434295291
+  },
+  "seed2": {
+   "l0": 30.421895,
+   "recon_r2": 0.8572033107833146,
+   "l1_per_token": 158.237112538271
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.8654386932565398,
+   "mcc_abs": 0.8654454800008732,
+   "act_iou": 0.5995016795260429,
+   "act_iou_n_pairs": 3069,
+   "input_r2": 0.9862883786692969,
+   "matched_mcc_signed": {
+    "0.6": 0.9271416533819563,
+    "0.7": 0.9363930918136538,
+    "0.8": 0.9485881321928074,
+    "0.9": 0.9638700779937296
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.9272186675248668,
+    "0.7": 0.9367115703611046,
+    "0.8": 0.9479330888879861,
+    "0.9": 0.9639542867114299
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.8600208672454513,
+   "mcc_abs": 0.8600288736203424,
+   "act_iou": 0.5933268458312826,
+   "act_iou_n_pairs": 3068,
+   "input_r2": 0.9859553207342857,
+   "matched_mcc_signed": {
+    "0.6": 0.9261328555503763,
+    "0.7": 0.9361424606290005,
+    "0.8": 0.9469193613210625,
+    "0.9": 0.9637184685051642
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.9248310571078391,
+    "0.7": 0.9354426946996001,
+    "0.8": 0.9466264543058788,
+    "0.9": 0.963574333719074
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.8663848108300408,
+   "mcc_abs": 0.8663943941883131,
+   "act_iou": 0.6001136170721506,
+   "act_iou_n_pairs": 3070,
+   "input_r2": 0.986509344613452,
+   "matched_mcc_signed": {
+    "0.6": 0.9287942348508964,
+    "0.7": 0.9394074065113508,
+    "0.8": 0.9490347977848638,
+    "0.9": 0.964295600895391
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.9278989420272552,
+    "0.7": 0.9382064112636884,
+    "0.8": 0.9490894005070657,
+    "0.9": 0.9643777642062104
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 30.616503333333338,
+  "l0_std": 0.18014061360349745,
+  "recon_r2_mean": 0.8574503154139501,
+  "recon_r2_std": 0.00021665323367223277,
+  "l1_per_token_mean": 158.7048989208082,
+  "l1_per_token_std": 0.40973510521574624,
+  "input_r2_mean": 0.9862510146723449,
+  "input_r2_std": 0.00027889543896015716,
+  "mcc_signed_mean": 0.863948123777344,
+  "mcc_signed_std": 0.003433845150293164,
+  "mcc_abs_mean": 0.8639562492698429,
+  "mcc_abs_std": 0.003434140234905703,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.927356247927743,
+   "0.7": 0.937314319651335,
+   "0.8": 0.9481807637662446,
+   "0.9": 0.9639613824647616
+  },
+  "matched_mcc_signed_band": 0.944203178452521,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.9266495555533204,
+   "0.7": 0.936786892108131,
+   "0.8": 0.9478829812336436,
+   "0.9": 0.9639687948789047
+  },
+  "matched_mcc_abs_band": 0.9438220559435,
+  "act_iou_mean": 0.5976473808098254,
+  "act_iou_std": 0.0037541821700195735,
+  "act_iou_n_pairs": 3069.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_32/dict_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x4/l0_32/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..ff8e111b0f7015f7e501537bba522347d5922e5e
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_32/dict_metrics_step_final.json
@@ -0,0 +1,595 @@
+{
+ "schema": 3,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 31,
+   "l0_raw": 30.7731021484375,
+   "recon_r2": 0.857375615987999,
+   "l1_per_token": 158.86725765566513,
+   "stats_source": "stats",
+   "n_dead": 9,
+   "dead_rate": 0.0029296875,
+   "alive_source": "stats",
+   "n_dense": 38,
+   "dense_rate": 0.012369791666666666,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 959,
+    "0.6": 452,
+    "0.7": 159,
+    "0.8": 34,
+    "0.9": 6
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 418,
+    "0.6": 200,
+    "0.7": 74,
+    "0.8": 16,
+    "0.9": 2
+   },
+   "n_dup_removed_signed": {
+    "0.5": 541,
+    "0.6": 252,
+    "0.7": 85,
+    "0.8": 18,
+    "0.9": 4
+   },
+   "largest_dup_signed": {
+    "0.5": 7,
+    "0.6": 5,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_signed": 0.4792686104774475,
+   "n_dup_abs": {
+    "0.5": 964,
+    "0.6": 455,
+    "0.7": 164,
+    "0.8": 44,
+    "0.9": 6
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 416,
+    "0.6": 201,
+    "0.7": 75,
+    "0.8": 21,
+    "0.9": 2
+   },
+   "n_dup_removed_abs": {
+    "0.5": 548,
+    "0.6": 254,
+    "0.7": 89,
+    "0.8": 23,
+    "0.9": 4
+   },
+   "largest_dup_abs": {
+    "0.5": 8,
+    "0.6": 5,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_abs": 0.4804518222808838,
+   "dup_excludes_dead": true
+  },
+  "seed1": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 31,
+   "l0_raw": 30.641238916015624,
+   "recon_r2": 0.8573009543411525,
+   "l1_per_token": 158.69405532050325,
+   "stats_source": "stats",
+   "n_dead": 3,
+   "dead_rate": 0.0009765625,
+   "alive_source": "stats",
+   "n_dense": 39,
+   "dense_rate": 0.0126953125,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 953,
+    "0.6": 468,
+    "0.7": 153,
+    "0.8": 44,
+    "0.9": 7
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 415,
+    "0.6": 212,
+    "0.7": 70,
+    "0.8": 20,
+    "0.9": 2
+   },
+   "n_dup_removed_signed": {
+    "0.5": 538,
+    "0.6": 256,
+    "0.7": 83,
+    "0.8": 24,
+    "0.9": 5
+   },
+   "largest_dup_signed": {
+    "0.5": 9,
+    "0.6": 8,
+    "0.7": 5,
+    "0.8": 5,
+    "0.9": 5
+   },
+   "babel_mu1_signed": 0.48020607233047485,
+   "n_dup_abs": {
+    "0.5": 955,
+    "0.6": 474,
+    "0.7": 164,
+    "0.8": 52,
+    "0.9": 7
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 412,
+    "0.6": 210,
+    "0.7": 73,
+    "0.8": 24,
+    "0.9": 2
+   },
+   "n_dup_removed_abs": {
+    "0.5": 543,
+    "0.6": 264,
+    "0.7": 91,
+    "0.8": 28,
+    "0.9": 5
+   },
+   "largest_dup_abs": {
+    "0.5": 10,
+    "0.6": 6,
+    "0.7": 5,
+    "0.8": 5,
+    "0.9": 5
+   },
+   "babel_mu1_abs": 0.4813156723976135,
+   "dup_excludes_dead": true
+  },
+  "seed2": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 30,
+   "l0_raw": 30.427043701171876,
+   "recon_r2": 0.8917929536866043,
+   "l1_per_token": 158.1019702360834,
+   "stats_source": "stats",
+   "n_dead": 6,
+   "dead_rate": 0.001953125,
+   "alive_source": "stats",
+   "n_dense": 34,
+   "dense_rate": 0.011067708333333334,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 931,
+    "0.6": 429,
+    "0.7": 145,
+    "0.8": 27,
+    "0.9": 7
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 407,
+    "0.6": 195,
+    "0.7": 68,
+    "0.8": 13,
+    "0.9": 3
+   },
+   "n_dup_removed_signed": {
+    "0.5": 524,
+    "0.6": 234,
+    "0.7": 77,
+    "0.8": 14,
+    "0.9": 4
+   },
+   "largest_dup_signed": {
+    "0.5": 8,
+    "0.6": 5,
+    "0.7": 4,
+    "0.8": 3,
+    "0.9": 3
+   },
+   "babel_mu1_signed": 0.4782335162162781,
+   "n_dup_abs": {
+    "0.5": 931,
+    "0.6": 430,
+    "0.7": 152,
+    "0.8": 33,
+    "0.9": 7
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 404,
+    "0.6": 196,
+    "0.7": 69,
+    "0.8": 16,
+    "0.9": 3
+   },
+   "n_dup_removed_abs": {
+    "0.5": 527,
+    "0.6": 234,
+    "0.7": 83,
+    "0.8": 17,
+    "0.9": 4
+   },
+   "largest_dup_abs": {
+    "0.5": 9,
+    "0.6": 5,
+    "0.7": 4,
+    "0.8": 3,
+    "0.9": 3
+   },
+   "babel_mu1_abs": 0.47892436385154724,
+   "dup_excludes_dead": true
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.9098032116889954,
+   "dcs_abs": 0.9098515510559082,
+   "frac_sign_flipped": 0.0032552083333333335,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 3063,
+   "n_alive_b": 3069,
+   "matched_cos_signed": {
+    "0.5": 0.9109748005867004,
+    "0.6": 0.9112560749053955,
+    "0.7": 0.911565363407135,
+    "0.8": 0.9106254577636719,
+    "0.9": 0.9115678668022156
+   },
+   "unique_frac_signed": {
+    "0.5": 0.8224283854166666,
+    "0.6": 0.9153645833333334,
+    "0.7": 0.970703125,
+    "0.8": 0.9912109375,
+    "0.9": 0.99658203125
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7492115344017899,
+    "0.6": 0.8341315373157462,
+    "0.7": 0.8848593469010666,
+    "0.8": 0.9026219137012959,
+    "0.9": 0.9084521563199814
+   },
+   "matched_frac_signed": {
+    "0.5": 0.7682291666666666,
+    "0.6": 0.8404947916666666,
+    "0.7": 0.880859375,
+    "0.8": 0.8714192708333334,
+    "0.9": 0.8170572916666666
+   },
+   "matched_cos_abs": {
+    "0.5": 0.9114334583282471,
+    "0.6": 0.9116860032081604,
+    "0.7": 0.9114117622375488,
+    "0.8": 0.9107311964035034,
+    "0.9": 0.9115778207778931
+   },
+   "unique_frac_abs": {
+    "0.5": 0.8204752604166666,
+    "0.6": 0.9137369791666666,
+    "0.7": 0.96875,
+    "0.8": 0.98974609375,
+    "0.9": 0.99658203125
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7478086040743316,
+    "0.6": 0.8330412145199564,
+    "0.7": 0.8829301446676254,
+    "0.8": 0.9013926440966316,
+    "0.9": 0.9084620762732811
+   },
+   "matched_frac_abs": {
+    "0.5": 0.765625,
+    "0.6": 0.8395182291666666,
+    "0.7": 0.8792317708333334,
+    "0.8": 0.8707682291666666,
+    "0.9": 0.8167317708333334
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.9066429138183594,
+   "dcs_abs": 0.9067557454109192,
+   "frac_sign_flipped": 0.0026041666666666665,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 3063,
+   "n_alive_b": 3066,
+   "matched_cos_signed": {
+    "0.5": 0.9063559174537659,
+    "0.6": 0.9108986854553223,
+    "0.7": 0.9098514318466187,
+    "0.8": 0.9085838794708252,
+    "0.9": 0.9081938862800598
+   },
+   "unique_frac_signed": {
+    "0.5": 0.82421875,
+    "0.6": 0.91845703125,
+    "0.7": 0.97119140625,
+    "0.8": 0.9923502604166666,
+    "0.9": 0.9962565104166666
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7470355413388461,
+    "0.6": 0.8366213024128228,
+    "0.7": 0.8836398915736936,
+    "0.8": 0.9016334494032586,
+    "0.9": 0.9047940719271234
+   },
+   "matched_frac_signed": {
+    "0.5": 0.7610677083333334,
+    "0.6": 0.8404947916666666,
+    "0.7": 0.8743489583333334,
+    "0.8": 0.8642578125,
+    "0.9": 0.8059895833333334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.9071646332740784,
+    "0.6": 0.911187469959259,
+    "0.7": 0.9095990061759949,
+    "0.8": 0.9088952541351318,
+    "0.9": 0.9082985520362854
+   },
+   "unique_frac_abs": {
+    "0.5": 0.8225911458333334,
+    "0.6": 0.9181315104166666,
+    "0.7": 0.9695638020833334,
+    "0.8": 0.9910481770833334,
+    "0.9": 0.9962565104166666
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7462255951443998,
+    "0.6": 0.8365899280664355,
+    "0.7": 0.881914270799219,
+    "0.8": 0.9007589847703155,
+    "0.9": 0.9048983458681807
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7594401041666666,
+    "0.6": 0.8401692708333334,
+    "0.7": 0.8727213541666666,
+    "0.8": 0.8639322916666666,
+    "0.9": 0.8063151041666666
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.9099404215812683,
+   "dcs_abs": 0.9100392460823059,
+   "frac_sign_flipped": 0.0035807291666666665,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 3069,
+   "n_alive_b": 3066,
+   "matched_cos_signed": {
+    "0.5": 0.9094085693359375,
+    "0.6": 0.9146425127983093,
+    "0.7": 0.9114592671394348,
+    "0.8": 0.9121108055114746,
+    "0.9": 0.9109505414962769
+   },
+   "unique_frac_signed": {
+    "0.5": 0.82568359375,
+    "0.6": 0.9187825520833334,
+    "0.7": 0.9724934895833334,
+    "0.8": 0.9923502604166666,
+    "0.9": 0.9970703125
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7508837357163429,
+    "0.6": 0.8403575821527436,
+    "0.7": 0.8863882033134967,
+    "0.8": 0.9051333953781674,
+    "0.9": 0.908281741081737
+   },
+   "matched_frac_signed": {
+    "0.5": 0.7662760416666666,
+    "0.6": 0.8453776041666666,
+    "0.7": 0.8805338541666666,
+    "0.8": 0.875,
+    "0.9": 0.8193359375
+   },
+   "matched_cos_abs": {
+    "0.5": 0.9098859429359436,
+    "0.6": 0.915928065776825,
+    "0.7": 0.9119309782981873,
+    "0.8": 0.9122180938720703,
+    "0.9": 0.9110379219055176
+   },
+   "unique_frac_abs": {
+    "0.5": 0.8243815104166666,
+    "0.6": 0.91748046875,
+    "0.7": 0.97021484375,
+    "0.8": 0.9912109375,
+    "0.9": 0.9970703125
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7500931479444262,
+    "0.6": 0.8403461111302022,
+    "0.7": 0.8847689716203604,
+    "0.8": 0.9042005520313978,
+    "0.9": 0.908368865493685
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7652994791666666,
+    "0.6": 0.8447265625,
+    "0.7": 0.87890625,
+    "0.8": 0.8727213541666666,
+    "0.9": 0.8186848958333334
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 6,
+  "n_dead_std": 3.0,
+  "n_dense_mean": 37,
+  "n_dense_std": 2.6457513110645907,
+  "l0_mean": 31,
+  "l0_std": 0.1746539202015416,
+  "recon_r2_mean": 0.8688231746719186,
+  "recon_r2_std": 0.019892447174191464,
+  "l1_per_token_mean": 158.55442773741726,
+  "l1_per_token_std": 0.40129553338075763,
+  "babel_mu1_signed_mean": 0.47923606634140015,
+  "babel_mu1_signed_std": 0.000986680670990911,
+  "n_dup_signed_per_threshold": {
+   "0.5": 947.6666666666666,
+   "0.6": 449.6666666666667,
+   "0.7": 152.33333333333334,
+   "0.8": 35.0,
+   "0.9": 6.666666666666667
+  },
+  "n_dup_signed_band": 160.91666666666666,
+  "n_dup_signed_band_std": 8.156030488744713,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 413.3333333333333,
+   "0.6": 202.33333333333334,
+   "0.7": 70.66666666666667,
+   "0.8": 16.333333333333332,
+   "0.9": 2.3333333333333335
+  },
+  "n_dup_clusters_signed_band": 72.91666666666666,
+  "n_dup_clusters_signed_band_std": 3.125833222251842,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 8.0,
+   "0.6": 6.0,
+   "0.7": 4.333333333333333,
+   "0.8": 4.0,
+   "0.9": 4.0
+  },
+  "largest_dup_signed_band": 4.583333333333333,
+  "largest_dup_signed_band_std": 1.0408329997330665,
+  "dcs_signed_mean": 0.9087955156962076,
+  "dcs_signed_std": 0.001865469852884638,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.7490436038189929,
+   "0.6": 0.8370368072937708,
+   "0.7": 0.8849624805960857,
+   "0.8": 0.9031295861609072,
+   "0.9": 0.9071759897762806
+  },
+  "dedup_dcs_signed_band": 0.8830762159567611,
+  "dedup_dcs_signed_band_std": 0.0017524622669898523,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.9089130957921346,
+   "0.6": 0.9122657577196757,
+   "0.7": 0.9109586874643961,
+   "0.8": 0.9104400475819906,
+   "0.9": 0.9102374315261841
+  },
+  "matched_cos_signed_band": 0.9109754810730616,
+  "matched_cos_signed_band_std": 0.001474227178889647,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.8241102430555555,
+   "0.6": 0.9175347222222223,
+   "0.7": 0.9714626736111112,
+   "0.8": 0.991970486111111,
+   "0.9": 0.9966362847222222
+  },
+  "unique_frac_signed_band": 0.9694010416666667,
+  "unique_frac_signed_band_std": 0.0008660399029049685,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.7651909722222222,
+   "0.6": 0.8421223958333334,
+   "0.7": 0.8785807291666666,
+   "0.8": 0.8702256944444445,
+   "0.9": 0.8141276041666666
+  },
+  "matched_frac_signed_band": 0.8512641059027778,
+  "matched_frac_signed_band_std": 0.00451446268722102,
+  "babel_mu1_abs_mean": 0.48023061951001483,
+  "babel_mu1_abs_std": 0.0012109034395865275,
+  "n_dup_abs_per_threshold": {
+   "0.5": 950.0,
+   "0.6": 453.0,
+   "0.7": 160.0,
+   "0.8": 43.0,
+   "0.9": 6.666666666666667
+  },
+  "n_dup_abs_band": 165.66666666666666,
+  "n_dup_abs_band_std": 9.47474713822661,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 410.6666666666667,
+   "0.6": 202.33333333333334,
+   "0.7": 72.33333333333333,
+   "0.8": 20.333333333333332,
+   "0.9": 2.3333333333333335
+  },
+  "n_dup_clusters_abs_band": 74.33333333333333,
+  "n_dup_clusters_abs_band_std": 3.145764348029479,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 9.0,
+   "0.6": 5.333333333333333,
+   "0.7": 4.333333333333333,
+   "0.8": 4.0,
+   "0.9": 4.0
+  },
+  "largest_dup_abs_band": 4.416666666666666,
+  "largest_dup_abs_band_std": 0.7637626158259734,
+  "dcs_abs_mean": 0.9088821808497111,
+  "dcs_abs_std": 0.0018439368514800657,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.7480424490543859,
+   "0.6": 0.8366590845721981,
+   "0.7": 0.8832044623624017,
+   "0.8": 0.9021173936327817,
+   "0.9": 0.9072430958783823
+  },
+  "dedup_dcs_abs_band": 0.882306009111441,
+  "dedup_dcs_abs_band_std": 0.001843523594278421,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.909494678179423,
+   "0.6": 0.9129338463147482,
+   "0.7": 0.9109805822372437,
+   "0.8": 0.9106148481369019,
+   "0.9": 0.9103047649065653
+  },
+  "matched_cos_abs_band": 0.9112085103988647,
+  "matched_cos_abs_band_std": 0.0016465232216753482,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.8224826388888888,
+   "0.6": 0.9164496527777777,
+   "0.7": 0.9695095486111112,
+   "0.8": 0.9906684027777778,
+   "0.9": 0.9966362847222222
+  },
+  "unique_frac_abs_band": 0.9683159722222222,
+  "unique_frac_abs_band_std": 0.0009708946488809501,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.763454861111111,
+   "0.6": 0.8414713541666666,
+   "0.7": 0.876953125,
+   "0.8": 0.869140625,
+   "0.9": 0.8139105902777778
+  },
+  "matched_frac_abs_band": 0.850368923611111,
+  "matched_frac_abs_band_std": 0.004119425096478593,
+  "alive_source": "stats",
+  "frac_denominator": "d_sae",
+  "d_sae": 3072
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_32/seed0/config.json b/lm/gpt2-small_layer8/relu/x4/l0_32/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..920481756aae71a20da4a01e54beb062205c045a
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_32/seed0/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..a3a613dcd18ca63c4dfc2ce3761e6b1a3bc3f110
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 30.7731021484375,
+  "l0_per_token_0.001": 30.764328662109374,
+  "l0_per_token_0.01": 30.68584609375,
+  "l0_per_token_0.05": 30.340010888671873,
+  "l0_per_token_0.1": 29.91510498046875,
+  "l0_per_token_0.5": 26.795005712890624,
+  "l0_per_token_1.0": 23.48800966796875,
+  "l1_per_token": 158.86725765566513,
+  "r2_score": 0.857375615987999,
+  "l2_loss": 3.3448889550443894
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_32/seed1/config.json b/lm/gpt2-small_layer8/relu/x4/l0_32/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..2b729114fa7a39fd6d4550e7ebd1e771051ed38a
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_32/seed1/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..8e8c9cb2c242487ba7569febf70fa7908fd2e137
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 30.641238916015624,
+  "l0_per_token_0.001": 30.6325423828125,
+  "l0_per_token_0.01": 30.554298388671874,
+  "l0_per_token_0.05": 30.209945849609376,
+  "l0_per_token_0.1": 29.787095654296873,
+  "l0_per_token_0.5": 26.678293310546874,
+  "l0_per_token_1.0": 23.382165087890623,
+  "l1_per_token": 158.69405532050325,
+  "r2_score": 0.8573009543411525,
+  "l2_loss": 3.346639952390548
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_32/seed2/config.json b/lm/gpt2-small_layer8/relu/x4/l0_32/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..d2ac614f301c676dc8383e32f921621c96254746
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_32/seed2/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..45f9ded400cc5c793de78d97fc41a798b0aad457
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 30.427043701171876,
+  "l0_per_token_0.001": 30.41833974609375,
+  "l0_per_token_0.01": 30.340196533203127,
+  "l0_per_token_0.05": 29.996419677734377,
+  "l0_per_token_0.1": 29.574092236328124,
+  "l0_per_token_0.5": 26.473829541015625,
+  "l0_per_token_1.0": 23.195003076171876,
+  "l1_per_token": 158.1019702360834,
+  "r2_score": 0.8917929536866043,
+  "l2_loss": 3.354440257187146
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_64/code_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x4/l0_64/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..33d03d02a3cb9442050506a245f894f65bf5adbd
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_64/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 67.79718,
+   "recon_r2": 0.8895117454916638,
+   "l1_per_token": 224.0494992746434
+  },
+  "seed1": {
+   "l0": 67.44265,
+   "recon_r2": 0.8898536064032856,
+   "l1_per_token": 225.0048554069657
+  },
+  "seed2": {
+   "l0": 67.42118,
+   "recon_r2": 0.8893652495221115,
+   "l1_per_token": 223.5672629712749
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.8564358562276394,
+   "mcc_abs": 0.8564371454334504,
+   "act_iou": 0.5669568116511763,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9860857025992825,
+   "matched_mcc_signed": {
+    "0.6": 0.9181662050299778,
+    "0.7": 0.9309742028993138,
+    "0.8": 0.9454268562380749,
+    "0.9": 0.9635285781509857
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.9154107702948895,
+    "0.7": 0.929416516356726,
+    "0.8": 0.9444129591842337,
+    "0.9": 0.9634085085535484
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.8566618975009904,
+   "mcc_abs": 0.8566698632940591,
+   "act_iou": 0.5654278196018431,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9861346117062605,
+   "matched_mcc_signed": {
+    "0.6": 0.9166171548842256,
+    "0.7": 0.9305953563640375,
+    "0.8": 0.9440754725668381,
+    "0.9": 0.9628593323672413
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.9143256126612352,
+    "0.7": 0.9276585431061486,
+    "0.8": 0.9427065257734183,
+    "0.9": 0.962679487044736
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.8536072720168884,
+   "mcc_abs": 0.8536167970536105,
+   "act_iou": 0.5614501302638525,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9860095382714625,
+   "matched_mcc_signed": {
+    "0.6": 0.9133090950972399,
+    "0.7": 0.9250341473980924,
+    "0.8": 0.9424556661728537,
+    "0.9": 0.962042976897987
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.9127573366676844,
+    "0.7": 0.9245496296167824,
+    "0.8": 0.9420081432477986,
+    "0.9": 0.9619196973942117
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 67.55367,
+  "l0_std": 0.21115889822595307,
+  "recon_r2_mean": 0.8895768671390204,
+  "recon_r2_std": 0.0002506067288868006,
+  "l1_per_token_mean": 224.20720588429467,
+  "l1_per_token_std": 0.7316567049064856,
+  "input_r2_mean": 0.9860766175256686,
+  "input_r2_std": 6.302971477777955e-05,
+  "mcc_signed_mean": 0.8555683419151728,
+  "mcc_signed_std": 0.0017020928364170826,
+  "mcc_abs_mean": 0.85557460192704,
+  "mcc_abs_std": 0.0016994967909589784,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.9160308183371478,
+   "0.7": 0.9288679022204812,
+   "0.8": 0.9439859983259221,
+   "0.9": 0.9628102958054047
+  },
+  "matched_mcc_signed_band": 0.937923753672239,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.9141645732079363,
+   "0.7": 0.927208229693219,
+   "0.8": 0.9430425427351502,
+   "0.9": 0.9626692309974988
+  },
+  "matched_mcc_abs_band": 0.9367711441584511,
+  "act_iou_mean": 0.5646115871722905,
+  "act_iou_std": 0.002842632848199784,
+  "act_iou_n_pairs": 3072.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_64/dict_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x4/l0_64/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..90126f21024e7b9f32ac1a03fde7a262b3a74002
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_64/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 68,
+   "l0_raw": 67.81546708984375,
+   "recon_r2": 0.8893530105065302,
+   "l1_per_token": 224.03872149346174,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 78,
+   "dense_rate": 0.025390625,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 576,
+    "0.6": 247,
+    "0.7": 73,
+    "0.8": 15,
+    "0.9": 3
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 247,
+    "0.6": 109,
+    "0.7": 35,
+    "0.8": 7,
+    "0.9": 1
+   },
+   "largest_dup_signed": {
+    "0.5": 8,
+    "0.6": 5,
+    "0.7": 3,
+    "0.8": 3,
+    "0.9": 3
+   },
+   "babel_mu1_signed": 0.41098713874816895,
+   "n_dup_abs": {
+    "0.5": 610,
+    "0.6": 277,
+    "0.7": 93,
+    "0.8": 21,
+    "0.9": 3
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 260,
+    "0.6": 123,
+    "0.7": 45,
+    "0.8": 10,
+    "0.9": 1
+   },
+   "largest_dup_abs": {
+    "0.5": 8,
+    "0.6": 6,
+    "0.7": 3,
+    "0.8": 3,
+    "0.9": 3
+   },
+   "babel_mu1_abs": 0.41815099120140076
+  },
+  "seed1": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 67,
+   "l0_raw": 67.44947939453125,
+   "recon_r2": 0.8896708348547783,
+   "l1_per_token": 224.84895001402577,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 75,
+   "dense_rate": 0.0244140625,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 544,
+    "0.6": 240,
+    "0.7": 66,
+    "0.8": 17,
+    "0.9": 3
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 236,
+    "0.6": 107,
+    "0.7": 31,
+    "0.8": 8,
+    "0.9": 1
+   },
+   "largest_dup_signed": {
+    "0.5": 5,
+    "0.6": 4,
+    "0.7": 3,
+    "0.8": 3,
+    "0.9": 3
+   },
+   "babel_mu1_signed": 0.4083280563354492,
+   "n_dup_abs": {
+    "0.5": 575,
+    "0.6": 270,
+    "0.7": 80,
+    "0.8": 27,
+    "0.9": 3
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 244,
+    "0.6": 120,
+    "0.7": 38,
+    "0.8": 13,
+    "0.9": 1
+   },
+   "largest_dup_abs": {
+    "0.5": 7,
+    "0.6": 5,
+    "0.7": 3,
+    "0.8": 3,
+    "0.9": 3
+   },
+   "babel_mu1_abs": 0.41518664360046387
+  },
+  "seed2": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 67,
+   "l0_raw": 67.45180219726562,
+   "recon_r2": 0.8892171446981043,
+   "l1_per_token": 223.5389451400725,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 76,
+   "dense_rate": 0.024739583333333332,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 573,
+    "0.6": 236,
+    "0.7": 73,
+    "0.8": 12,
+    "0.9": 2
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 247,
+    "0.6": 105,
+    "0.7": 36,
+    "0.8": 6,
+    "0.9": 1
+   },
+   "largest_dup_signed": {
+    "0.5": 7,
+    "0.6": 5,
+    "0.7": 3,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "babel_mu1_signed": 0.40990838408470154,
+   "n_dup_abs": {
+    "0.5": 603,
+    "0.6": 265,
+    "0.7": 105,
+    "0.8": 26,
+    "0.9": 2
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 259,
+    "0.6": 118,
+    "0.7": 51,
+    "0.8": 13,
+    "0.9": 1
+   },
+   "largest_dup_abs": {
+    "0.5": 9,
+    "0.6": 5,
+    "0.7": 3,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "babel_mu1_abs": 0.41793885827064514
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.8935964703559875,
+   "dcs_abs": 0.893878161907196,
+   "frac_sign_flipped": 0.006184895833333333,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.8913441896438599,
+    "0.6": 0.8931550979614258,
+    "0.7": 0.893197238445282,
+    "0.8": 0.8934041857719421,
+    "0.9": 0.8935273885726929
+   },
+   "unique_frac_signed": {
+    "0.5": 0.8963216145833334,
+    "0.6": 0.9558919270833334,
+    "0.7": 0.9881184895833334,
+    "0.8": 0.9972330729166666,
+    "0.9": 0.9993489583333334
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7989310632110573,
+    "0.6": 0.8537597477746507,
+    "0.7": 0.8825847061525565,
+    "0.8": 0.8909322015339664,
+    "0.9": 0.8929456650124242
+   },
+   "matched_frac_signed": {
+    "0.5": 0.828125,
+    "0.6": 0.8694661458333334,
+    "0.7": 0.8753255208333334,
+    "0.8": 0.8447265625,
+    "0.9": 0.7646484375
+   },
+   "matched_cos_abs": {
+    "0.5": 0.892730712890625,
+    "0.6": 0.8939826488494873,
+    "0.7": 0.8944237232208252,
+    "0.8": 0.8937780857086182,
+    "0.9": 0.8938090205192566
+   },
+   "unique_frac_abs": {
+    "0.5": 0.88916015625,
+    "0.6": 0.9505208333333334,
+    "0.7": 0.9853515625,
+    "0.8": 0.9959309895833334,
+    "0.9": 0.9993489583333334
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.793780580163002,
+    "0.6": 0.8497491323699554,
+    "0.7": 0.8813218132127076,
+    "0.8": 0.8901412933676814,
+    "0.9": 0.8932271136048561
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8219401041666666,
+    "0.6": 0.8655598958333334,
+    "0.7": 0.8727213541666666,
+    "0.8": 0.8434244791666666,
+    "0.9": 0.7652994791666666
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.8937034606933594,
+   "dcs_abs": 0.8944053053855896,
+   "frac_sign_flipped": 0.010416666666666666,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.8886062502861023,
+    "0.6": 0.8929953575134277,
+    "0.7": 0.8933100700378418,
+    "0.8": 0.8938999176025391,
+    "0.9": 0.8939065337181091
+   },
+   "unique_frac_signed": {
+    "0.5": 0.8933919270833334,
+    "0.6": 0.9562174479166666,
+    "0.7": 0.98779296875,
+    "0.8": 0.9977213541666666,
+    "0.9": 0.99951171875
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7938736503613958,
+    "0.6": 0.8538977417629212,
+    "0.7": 0.8824054060969502,
+    "0.8": 0.891863036279877,
+    "0.9": 0.8934700559184421
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8258463541666666,
+    "0.6": 0.8688151041666666,
+    "0.7": 0.8733723958333334,
+    "0.8": 0.8453776041666666,
+    "0.9": 0.7613932291666666
+   },
+   "matched_cos_abs": {
+    "0.5": 0.890869140625,
+    "0.6": 0.894310712814331,
+    "0.7": 0.8948605060577393,
+    "0.8": 0.8945366740226746,
+    "0.9": 0.8946104645729065
+   },
+   "unique_frac_abs": {
+    "0.5": 0.8870442708333334,
+    "0.6": 0.9510091145833334,
+    "0.7": 0.9833984375,
+    "0.8": 0.99609375,
+    "0.9": 0.99951171875
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7902403672536215,
+    "0.6": 0.8504976391559467,
+    "0.7": 0.8800044234376401,
+    "0.8": 0.8910423901397735,
+    "0.9": 0.8941736430570018
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8203125,
+    "0.6": 0.8655598958333334,
+    "0.7": 0.8723958333333334,
+    "0.8": 0.8463541666666666,
+    "0.9": 0.7626953125
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.8918560147285461,
+   "dcs_abs": 0.8921594619750977,
+   "frac_sign_flipped": 0.007486979166666667,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.8901619911193848,
+    "0.6": 0.8911995887756348,
+    "0.7": 0.8916789293289185,
+    "0.8": 0.8919728994369507,
+    "0.9": 0.8920238614082336
+   },
+   "unique_frac_signed": {
+    "0.5": 0.8968098958333334,
+    "0.6": 0.95703125,
+    "0.7": 0.98828125,
+    "0.8": 0.99755859375,
+    "0.9": 0.99951171875
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7983060825305681,
+    "0.6": 0.8529058564454317,
+    "0.7": 0.8812295668758452,
+    "0.8": 0.8897952312254347,
+    "0.9": 0.8915883028821554
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8307291666666666,
+    "0.6": 0.8723958333333334,
+    "0.7": 0.8815104166666666,
+    "0.8": 0.83984375,
+    "0.9": 0.7542317708333334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8906879425048828,
+    "0.6": 0.8920943737030029,
+    "0.7": 0.8935738801956177,
+    "0.8": 0.892180860042572,
+    "0.9": 0.8922937512397766
+   },
+   "unique_frac_abs": {
+    "0.5": 0.89013671875,
+    "0.6": 0.95166015625,
+    "0.7": 0.984375,
+    "0.8": 0.99560546875,
+    "0.9": 0.99951171875
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.792834042571485,
+    "0.6": 0.8489706710679457,
+    "0.7": 0.8796117883175611,
+    "0.8": 0.8882601433724631,
+    "0.9": 0.8918580609315541
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8238932291666666,
+    "0.6": 0.8662109375,
+    "0.7": 0.8766276041666666,
+    "0.8": 0.8388671875,
+    "0.9": 0.7548828125
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 76,
+  "n_dense_std": 1.5275252316519468,
+  "l0_mean": 67,
+  "l0_std": 0.21063576090277994,
+  "recon_r2_mean": 0.8894136633531377,
+  "recon_r2_std": 0.00017615468877441097,
+  "l1_per_token_mean": 224.14220554918668,
+  "l1_per_token_std": 0.6611050712956091,
+  "babel_mu1_signed_mean": 0.40974119305610657,
+  "babel_mu1_signed_std": 0.0013374021270511065,
+  "n_dup_signed_per_threshold": {
+   "0.5": 564.3333333333334,
+   "0.6": 241.0,
+   "0.7": 70.66666666666667,
+   "0.8": 14.666666666666666,
+   "0.9": 2.6666666666666665
+  },
+  "n_dup_signed_band": 82.25000000000001,
+  "n_dup_signed_band_std": 1.984313483298443,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 243.33333333333334,
+   "0.6": 107.0,
+   "0.7": 34.0,
+   "0.8": 7.0,
+   "0.9": 1.0
+  },
+  "n_dup_clusters_signed_band": 37.25,
+  "n_dup_clusters_signed_band_std": 0.6614378277661477,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 6.666666666666667,
+   "0.6": 4.666666666666667,
+   "0.7": 3.0,
+   "0.8": 2.6666666666666665,
+   "0.9": 2.6666666666666665
+  },
+  "largest_dup_signed_band": 3.25,
+  "largest_dup_signed_band_std": 0.25,
+  "dcs_signed_mean": 0.8930519819259644,
+  "dcs_signed_std": 0.0010371185496595918,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.7970369320343403,
+   "0.6": 0.853521115327668,
+   "0.7": 0.8820732263751173,
+   "0.8": 0.8908634896797594,
+   "0.9": 0.892668007937674
+  },
+  "dedup_dcs_signed_band": 0.8797814598300546,
+  "dedup_dcs_signed_band_std": 0.0008006633921919912,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.890037477016449,
+   "0.6": 0.8924500147501627,
+   "0.7": 0.8927287459373474,
+   "0.8": 0.8930923342704773,
+   "0.9": 0.8931525945663452
+  },
+  "matched_cos_signed_band": 0.8928559223810831,
+  "matched_cos_signed_band_std": 0.0009901834384499397,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.8955078125,
+   "0.6": 0.9563802083333334,
+   "0.7": 0.9880642361111112,
+   "0.8": 0.9975043402777777,
+   "0.9": 0.9994574652777778
+  },
+  "unique_frac_signed_band": 0.9853515625,
+  "unique_frac_signed_band_std": 0.00022655291189898147,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.8282335069444443,
+   "0.6": 0.8702256944444445,
+   "0.7": 0.8767361111111112,
+   "0.8": 0.8433159722222222,
+   "0.9": 0.7600911458333334
+  },
+  "matched_frac_signed_band": 0.8375922309027779,
+  "matched_frac_signed_band_std": 0.0008312474742863165,
+  "babel_mu1_abs_mean": 0.41709216435750324,
+  "babel_mu1_abs_std": 0.0016536345157587614,
+  "n_dup_abs_per_threshold": {
+   "0.5": 596.0,
+   "0.6": 270.6666666666667,
+   "0.7": 92.66666666666667,
+   "0.8": 24.666666666666668,
+   "0.9": 2.6666666666666665
+  },
+  "n_dup_abs_band": 97.66666666666669,
+  "n_dup_abs_band_std": 2.3629078131263044,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 254.33333333333334,
+   "0.6": 120.33333333333333,
+   "0.7": 44.666666666666664,
+   "0.8": 12.0,
+   "0.9": 1.0
+  },
+  "n_dup_clusters_abs_band": 44.5,
+  "n_dup_clusters_abs_band_std": 1.3919410907075054,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 8.0,
+   "0.6": 5.333333333333333,
+   "0.7": 3.0,
+   "0.8": 2.6666666666666665,
+   "0.9": 2.6666666666666665
+  },
+  "largest_dup_abs_band": 3.416666666666666,
+  "largest_dup_abs_band_std": 0.3818813079129867,
+  "dcs_abs_mean": 0.8934809764226278,
+  "dcs_abs_std": 0.0011744234278897557,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.7922849966627029,
+   "0.6": 0.8497391475312827,
+   "0.7": 0.880312674989303,
+   "0.8": 0.8898146089599727,
+   "0.9": 0.8930862725311374
+  },
+  "dedup_dcs_abs_band": 0.878238176002924,
+  "dedup_dcs_abs_band_std": 0.0009343674745983875,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.8914292653401693,
+   "0.6": 0.8934625784556071,
+   "0.7": 0.894286036491394,
+   "0.8": 0.8934985399246216,
+   "0.9": 0.8935710787773132
+  },
+  "matched_cos_abs_band": 0.893704558412234,
+  "matched_cos_abs_band_std": 0.0010531372335074354,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.8887803819444445,
+   "0.6": 0.9510633680555557,
+   "0.7": 0.984375,
+   "0.8": 0.9958767361111112,
+   "0.9": 0.9994574652777778
+  },
+  "unique_frac_abs_band": 0.9826931423611112,
+  "unique_frac_abs_band_std": 0.00016444709815786307,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.822048611111111,
+   "0.6": 0.8657769097222223,
+   "0.7": 0.8739149305555555,
+   "0.8": 0.8428819444444443,
+   "0.9": 0.7609592013888888
+  },
+  "matched_frac_abs_band": 0.8358832465277777,
+  "matched_frac_abs_band_std": 0.0015035163260146611,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_64/seed0/config.json b/lm/gpt2-small_layer8/relu/x4/l0_64/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..920481756aae71a20da4a01e54beb062205c045a
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_64/seed0/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..0483c8cb53270347635565a9ed5eb398e4cbc609
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 67.81546708984375,
+  "l0_per_token_0.001": 67.78537099609375,
+  "l0_per_token_0.01": 67.51494555664063,
+  "l0_per_token_0.05": 66.32953032226563,
+  "l0_per_token_0.1": 64.88450747070313,
+  "l0_per_token_0.5": 54.663884326171875,
+  "l0_per_token_1.0": 44.6233837890625,
+  "l1_per_token": 224.03872149346174,
+  "r2_score": 0.8893530105065302,
+  "l2_loss": 2.59494120018222
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_64/seed1/config.json b/lm/gpt2-small_layer8/relu/x4/l0_64/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..2b729114fa7a39fd6d4550e7ebd1e771051ed38a
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_64/seed1/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..24b250f8a90f63075a4b37a60a623e9de17782df
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 67.44947939453125,
+  "l0_per_token_0.001": 67.419630859375,
+  "l0_per_token_0.01": 67.151688671875,
+  "l0_per_token_0.05": 65.97722612304688,
+  "l0_per_token_0.1": 64.54579858398438,
+  "l0_per_token_0.5": 54.424803125,
+  "l0_per_token_1.0": 44.49315532226562,
+  "l1_per_token": 224.84895001402577,
+  "r2_score": 0.8896708348547783,
+  "l2_loss": 2.5874874456836503
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_64/seed2/config.json b/lm/gpt2-small_layer8/relu/x4/l0_64/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..d2ac614f301c676dc8383e32f921621c96254746
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_64/seed2/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..10ae297501e079f2a16bcb7e899ed5c3cc5c9c5c
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 67.45180219726562,
+  "l0_per_token_0.001": 67.42184458007813,
+  "l0_per_token_0.01": 67.15256127929688,
+  "l0_per_token_0.05": 65.97238212890625,
+  "l0_per_token_0.1": 64.53467993164062,
+  "l0_per_token_0.5": 54.36780102539063,
+  "l0_per_token_1.0": 44.39230356445312,
+  "l1_per_token": 223.5389451400725,
+  "r2_score": 0.8892171446981043,
+  "l2_loss": 2.598127584064818
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x8/l0_128/code_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x8/l0_128/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..478d0e0beb14bc54a11e2c82f220b67dd60dc883
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x8/l0_128/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 123.180675,
+   "recon_r2": 0.9198015861091443,
+   "l1_per_token": 267.1061286605835
+  },
+  "seed1": {
+   "l0": 124.46754666666666,
+   "recon_r2": 0.9204577705967866,
+   "l1_per_token": 269.6733626999381
+  },
+  "seed2": {
+   "l0": 123.86249833333333,
+   "recon_r2": 0.9199110441691133,
+   "l1_per_token": 267.40143723092876
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.8146245869321138,
+   "mcc_abs": 0.8146284806293408,
+   "act_iou": 0.49510302052855537,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9859093646519318,
+   "matched_mcc_signed": {
+    "0.6": 0.8901229367553425,
+    "0.7": 0.9114154565567896,
+    "0.8": 0.9340749511052155,
+    "0.9": 0.956276356717694
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8898561647899446,
+    "0.7": 0.9107094543075205,
+    "0.8": 0.9336756308049269,
+    "0.9": 0.956284744054315
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.8121166300193418,
+   "mcc_abs": 0.8121197415041888,
+   "act_iou": 0.4900390632699783,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9857723680595961,
+   "matched_mcc_signed": {
+    "0.6": 0.8875214333673219,
+    "0.7": 0.9099811623630916,
+    "0.8": 0.9311875674560166,
+    "0.9": 0.9551311027399433
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8857489186135841,
+    "0.7": 0.9090468937729539,
+    "0.8": 0.9308536766164816,
+    "0.9": 0.9551205661123896
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.8174337076633588,
+   "mcc_abs": 0.817441250097371,
+   "act_iou": 0.49735520545118644,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.986108127824478,
+   "matched_mcc_signed": {
+    "0.6": 0.8902167443356372,
+    "0.7": 0.9129030820712484,
+    "0.8": 0.9340012784815442,
+    "0.9": 0.9568960469441931
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8890099371581174,
+    "0.7": 0.9114804604234269,
+    "0.8": 0.9335296653534034,
+    "0.9": 0.9569189028074054
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 123.83690666666666,
+  "l0_std": 0.6438174210670652,
+  "recon_r2_mean": 0.9200568002916815,
+  "recon_r2_std": 0.00035153684857373136,
+  "l1_per_token_mean": 268.0603095304835,
+  "l1_per_token_std": 1.4047267291347267,
+  "input_r2_mean": 0.985929953512002,
+  "input_r2_std": 0.00016882411201491366,
+  "mcc_signed_mean": 0.8147249748716048,
+  "mcc_signed_std": 0.0026599599568269444,
+  "mcc_abs_mean": 0.8147298240769668,
+  "mcc_abs_std": 0.0026622014006477237,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8892870381527672,
+   "0.7": 0.9114332336637099,
+   "0.8": 0.9330879323475921,
+   "0.9": 0.9561011688006101
+  },
+  "matched_mcc_signed_band": 0.9224773432411698,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.888205006853882,
+   "0.7": 0.9104122695013004,
+   "0.8": 0.9326863242582707,
+   "0.9": 0.95610807099137
+  },
+  "matched_mcc_abs_band": 0.9218529179012058,
+  "act_iou_mean": 0.49416576308324006,
+  "act_iou_std": 0.003747041865660494,
+  "act_iou_n_pairs": 6144.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x8/l0_128/dict_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x8/l0_128/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..b9f48d58e77ef7e732ae51d6b7ba5ac958cbfe36
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x8/l0_128/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 123,
+   "l0_raw": 123.205973828125,
+   "recon_r2": 0.9196522611171748,
+   "l1_per_token": 267.0475748861746,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 86,
+   "dense_rate": 0.013997395833333334,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 1195,
+    "0.6": 515,
+    "0.7": 155,
+    "0.8": 33,
+    "0.9": 6
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 496,
+    "0.6": 227,
+    "0.7": 72,
+    "0.8": 15,
+    "0.9": 2
+   },
+   "largest_dup_signed": {
+    "0.5": 8,
+    "0.6": 6,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_signed": 0.4181537330150604,
+   "n_dup_abs": {
+    "0.5": 1341,
+    "0.6": 608,
+    "0.7": 229,
+    "0.8": 83,
+    "0.9": 6
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 567,
+    "0.6": 271,
+    "0.7": 108,
+    "0.8": 40,
+    "0.9": 2
+   },
+   "largest_dup_abs": {
+    "0.5": 13,
+    "0.6": 6,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_abs": 0.4338560998439789
+  },
+  "seed1": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 124,
+   "l0_raw": 124.420975,
+   "recon_r2": 0.9202786477365921,
+   "l1_per_token": 269.44816389988114,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 92,
+   "dense_rate": 0.014973958333333334,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 1161,
+    "0.6": 492,
+    "0.7": 137,
+    "0.8": 22,
+    "0.9": 4
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 479,
+    "0.6": 219,
+    "0.7": 63,
+    "0.8": 10,
+    "0.9": 1
+   },
+   "largest_dup_signed": {
+    "0.5": 12,
+    "0.6": 6,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_signed": 0.4172501862049103,
+   "n_dup_abs": {
+    "0.5": 1308,
+    "0.6": 601,
+    "0.7": 209,
+    "0.8": 68,
+    "0.9": 4
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 553,
+    "0.6": 272,
+    "0.7": 98,
+    "0.8": 33,
+    "0.9": 1
+   },
+   "largest_dup_abs": {
+    "0.5": 10,
+    "0.6": 5,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_abs": 0.4321509599685669
+  },
+  "seed2": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 124,
+   "l0_raw": 123.91108662109374,
+   "recon_r2": 0.9197673443730127,
+   "l1_per_token": 267.3776820182874,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 80,
+   "dense_rate": 0.013020833333333334,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 1163,
+    "0.6": 506,
+    "0.7": 152,
+    "0.8": 28,
+    "0.9": 5
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 483,
+    "0.6": 224,
+    "0.7": 69,
+    "0.8": 12,
+    "0.9": 1
+   },
+   "largest_dup_signed": {
+    "0.5": 9,
+    "0.6": 6,
+    "0.7": 5,
+    "0.8": 5,
+    "0.9": 5
+   },
+   "babel_mu1_signed": 0.4180673062801361,
+   "n_dup_abs": {
+    "0.5": 1296,
+    "0.6": 609,
+    "0.7": 237,
+    "0.8": 84,
+    "0.9": 9
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 546,
+    "0.6": 273,
+    "0.7": 112,
+    "0.8": 40,
+    "0.9": 3
+   },
+   "largest_dup_abs": {
+    "0.5": 9,
+    "0.6": 6,
+    "0.7": 5,
+    "0.8": 5,
+    "0.9": 5
+   },
+   "babel_mu1_abs": 0.4331169128417969
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.8661861419677734,
+   "dcs_abs": 0.866517961025238,
+   "frac_sign_flipped": 0.006510416666666667,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.8608965873718262,
+    "0.6": 0.8651069402694702,
+    "0.7": 0.8663362860679626,
+    "0.8": 0.8664904832839966,
+    "0.9": 0.8662425875663757
+   },
+   "unique_frac_signed": {
+    "0.5": 0.8876139322916666,
+    "0.6": 0.954345703125,
+    "0.7": 0.9872233072916666,
+    "0.8": 0.99755859375,
+    "0.9": 0.9994303385416666
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7641438052135829,
+    "0.6": 0.8256110911897849,
+    "0.7": 0.8552673735587935,
+    "0.8": 0.8643750280025415,
+    "0.9": 0.8657491225506722
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8040364583333334,
+    "0.6": 0.83642578125,
+    "0.7": 0.8284505208333334,
+    "0.8": 0.77783203125,
+    "0.9": 0.6858723958333334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8636515736579895,
+    "0.6": 0.866134762763977,
+    "0.7": 0.86763596534729,
+    "0.8": 0.8675652742385864,
+    "0.9": 0.8665667176246643
+   },
+   "unique_frac_abs": {
+    "0.5": 0.8755696614583334,
+    "0.6": 0.94580078125,
+    "0.7": 0.9811197916666666,
+    "0.8": 0.99365234375,
+    "0.9": 0.9994303385416666
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7561871159656828,
+    "0.6": 0.8191909352899529,
+    "0.7": 0.8512548175640404,
+    "0.8": 0.8620582681032829,
+    "0.9": 0.8660730679645591
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7957356770833334,
+    "0.6": 0.82958984375,
+    "0.7": 0.8248697916666666,
+    "0.8": 0.77685546875,
+    "0.9": 0.6857096354166666
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.8642866015434265,
+   "dcs_abs": 0.8645877838134766,
+   "frac_sign_flipped": 0.006510416666666667,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.8601382374763489,
+    "0.6": 0.8621219396591187,
+    "0.7": 0.8635818362236023,
+    "0.8": 0.864368736743927,
+    "0.9": 0.8642542958259583
+   },
+   "unique_frac_signed": {
+    "0.5": 0.8877766927083334,
+    "0.6": 0.95361328125,
+    "0.7": 0.9864908854166666,
+    "0.8": 0.9972330729166666,
+    "0.9": 0.9993489583333334
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7636106797387281,
+    "0.6": 0.8221309317159466,
+    "0.7": 0.8519156102459722,
+    "0.8": 0.8619770914762436,
+    "0.9": 0.8636916302687799
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8033854166666666,
+    "0.6": 0.8343098958333334,
+    "0.7": 0.8230794270833334,
+    "0.8": 0.7791341145833334,
+    "0.9": 0.6774088541666666
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8627469539642334,
+    "0.6": 0.8634780049324036,
+    "0.7": 0.8653819561004639,
+    "0.8": 0.8653475642204285,
+    "0.9": 0.8647215366363525
+   },
+   "unique_frac_abs": {
+    "0.5": 0.8759765625,
+    "0.6": 0.9452311197916666,
+    "0.7": 0.97998046875,
+    "0.8": 0.992919921875,
+    "0.9": 0.9991861979166666
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7557461110409349,
+    "0.6": 0.8161862815177301,
+    "0.7": 0.8480574149871245,
+    "0.8": 0.8592208358604694,
+    "0.9": 0.8640178244483346
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7931315104166666,
+    "0.6": 0.8297526041666666,
+    "0.7": 0.8194986979166666,
+    "0.8": 0.7771809895833334,
+    "0.9": 0.6774088541666666
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.8681418299674988,
+   "dcs_abs": 0.8685083985328674,
+   "frac_sign_flipped": 0.006998697916666667,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.8618559837341309,
+    "0.6": 0.8656688332557678,
+    "0.7": 0.8676130175590515,
+    "0.8": 0.868464469909668,
+    "0.9": 0.868198037147522
+   },
+   "unique_frac_signed": {
+    "0.5": 0.88916015625,
+    "0.6": 0.954833984375,
+    "0.7": 0.9872233072916666,
+    "0.8": 0.9977213541666666,
+    "0.9": 0.9994303385416666
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7663280011620373,
+    "0.6": 0.8265700212068623,
+    "0.7": 0.8565277926439496,
+    "0.8": 0.8664855469639102,
+    "0.9": 0.8677034581875583
+   },
+   "matched_frac_signed": {
+    "0.5": 0.81103515625,
+    "0.6": 0.8372395833333334,
+    "0.7": 0.8268229166666666,
+    "0.8": 0.7849934895833334,
+    "0.9": 0.6873372395833334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.86452317237854,
+    "0.6": 0.8672987818717957,
+    "0.7": 0.869307816028595,
+    "0.8": 0.8698415756225586,
+    "0.9": 0.8687153458595276
+   },
+   "unique_frac_abs": {
+    "0.5": 0.8775227864583334,
+    "0.6": 0.9458821614583334,
+    "0.7": 0.9807942708333334,
+    "0.8": 0.9935709635416666,
+    "0.9": 0.999267578125
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7586387831834145,
+    "0.6": 0.8203624464270737,
+    "0.7": 0.8526121255514834,
+    "0.8": 0.864249332419907,
+    "0.9": 0.8680790797370719
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8019205729166666,
+    "0.6": 0.8312174479166666,
+    "0.7": 0.8250325520833334,
+    "0.8": 0.7841796875,
+    "0.9": 0.6871744791666666
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 86,
+  "n_dense_std": 6.0,
+  "l0_mean": 124,
+  "l0_std": 0.6101090144951509,
+  "recon_r2_mean": 0.9198994177422598,
+  "recon_r2_std": 0.00025224484475272366,
+  "l1_per_token_mean": 267.957806934781,
+  "l1_per_token_std": 1.301197777607632,
+  "babel_mu1_signed_mean": 0.41782374183336896,
+  "babel_mu1_signed_std": 0.0004985899510919517,
+  "n_dup_signed_per_threshold": {
+   "0.5": 1173.0,
+   "0.6": 504.3333333333333,
+   "0.7": 148.0,
+   "0.8": 27.666666666666668,
+   "0.9": 5.0
+  },
+  "n_dup_signed_band": 171.24999999999997,
+  "n_dup_signed_band_std": 6.87386354243376,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 486.0,
+   "0.6": 223.33333333333334,
+   "0.7": 68.0,
+   "0.8": 12.333333333333334,
+   "0.9": 1.3333333333333333
+  },
+  "n_dup_clusters_signed_band": 76.25,
+  "n_dup_clusters_signed_band_std": 2.883140648667699,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 9.666666666666666,
+   "0.6": 6.0,
+   "0.7": 4.333333333333333,
+   "0.8": 4.333333333333333,
+   "0.9": 4.333333333333333
+  },
+  "largest_dup_signed_band": 4.749999999999999,
+  "largest_dup_signed_band_std": 0.4330127018922193,
+  "dcs_signed_mean": 0.8662048578262329,
+  "dcs_signed_std": 0.0019276823553070375,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.7646941620381161,
+   "0.6": 0.8247706813708646,
+   "0.7": 0.8545702588162385,
+   "0.8": 0.864279222147565,
+   "0.9": 0.8657147370023367
+  },
+  "dedup_dcs_signed_band": 0.8523337248342512,
+  "dedup_dcs_signed_band_std": 0.002225924615169546,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.8609636028607687,
+   "0.6": 0.8642992377281189,
+   "0.7": 0.8658437132835388,
+   "0.8": 0.8664412299791971,
+   "0.9": 0.866231640179952
+  },
+  "matched_cos_signed_band": 0.8657039552927017,
+  "matched_cos_signed_band_std": 0.0019742899627381073,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.88818359375,
+   "0.6": 0.9542643229166666,
+   "0.7": 0.9869791666666666,
+   "0.8": 0.9975043402777777,
+   "0.9": 0.9994032118055555
+  },
+  "unique_frac_signed_band": 0.9845377604166665,
+  "unique_frac_signed_band_std": 0.00032742262653461364,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.80615234375,
+   "0.6": 0.8359917534722223,
+   "0.7": 0.8261176215277778,
+   "0.8": 0.7806532118055557,
+   "0.9": 0.6835394965277778
+  },
+  "matched_frac_signed_band": 0.7815755208333334,
+  "matched_frac_signed_band_std": 0.0028506314922098572,
+  "babel_mu1_abs_mean": 0.4330413242181142,
+  "babel_mu1_abs_std": 0.0008550793698263809,
+  "n_dup_abs_per_threshold": {
+   "0.5": 1315.0,
+   "0.6": 606.0,
+   "0.7": 225.0,
+   "0.8": 78.33333333333333,
+   "0.9": 6.333333333333333
+  },
+  "n_dup_abs_band": 228.91666666666669,
+  "n_dup_abs_band_std": 7.467987234411513,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 555.3333333333334,
+   "0.6": 272.0,
+   "0.7": 106.0,
+   "0.8": 37.666666666666664,
+   "0.9": 2.0
+  },
+  "n_dup_clusters_abs_band": 104.41666666666667,
+  "n_dup_clusters_abs_band_std": 3.0855847635956026,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 10.666666666666666,
+   "0.6": 5.666666666666667,
+   "0.7": 4.333333333333333,
+   "0.8": 4.333333333333333,
+   "0.9": 4.333333333333333
+  },
+  "largest_dup_abs_band": 4.666666666666666,
+  "largest_dup_abs_band_std": 0.5204164998665333,
+  "dcs_abs_mean": 0.8665380477905273,
+  "dcs_abs_std": 0.0019603845421449656,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.7568573367300108,
+   "0.6": 0.818579887744919,
+   "0.7": 0.8506414527008829,
+   "0.8": 0.8618428121278864,
+   "0.9": 0.8660566573833218
+  },
+  "dedup_dcs_abs_band": 0.8492802024892525,
+  "dedup_dcs_abs_band_std": 0.002249781251225502,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.863640566666921,
+   "0.6": 0.8656371831893921,
+   "0.7": 0.8674419124921163,
+   "0.8": 0.8675848046938578,
+   "0.9": 0.8666678667068481
+  },
+  "matched_cos_abs_band": 0.8668329417705536,
+  "matched_cos_abs_band_std": 0.0020330686923579195,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.8763563368055557,
+   "0.6": 0.9456380208333334,
+   "0.7": 0.9806315104166666,
+   "0.8": 0.9933810763888888,
+   "0.9": 0.999294704861111
+  },
+  "unique_frac_abs_band": 0.979736328125,
+  "unique_frac_abs_band_std": 0.0003576333787281908,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.7969292534722222,
+   "0.6": 0.8301866319444443,
+   "0.7": 0.8231336805555555,
+   "0.8": 0.7794053819444445,
+   "0.9": 0.6834309895833334
+  },
+  "matched_frac_abs_band": 0.7790391710069444,
+  "matched_frac_abs_band_std": 0.002976317251623728,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x8/l0_32/code_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x8/l0_32/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..60533062fad48e76ac9188c6dbed89d6e755b1c7
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x8/l0_32/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 33.18100833333333,
+   "recon_r2": 0.8710936071271368,
+   "l1_per_token": 159.00141390715112
+  },
+  "seed1": {
+   "l0": 33.66731166666667,
+   "recon_r2": 0.8734548263822811,
+   "l1_per_token": 158.1527280203
+  },
+  "seed2": {
+   "l0": 33.13527833333333,
+   "recon_r2": 0.8704472558761559,
+   "l1_per_token": 157.94706212068493
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7340208913715975,
+   "mcc_abs": 0.7340213646026396,
+   "act_iou": 0.5066001822569854,
+   "act_iou_n_pairs": 5768,
+   "input_r2": 0.9853335978433151,
+   "matched_mcc_signed": {
+    "0.6": 0.9056522824071249,
+    "0.7": 0.9199839680095443,
+    "0.8": 0.935361113167485,
+    "0.9": 0.9549192325607102
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.9049211083498672,
+    "0.7": 0.91938895214407,
+    "0.8": 0.9352517609350761,
+    "0.9": 0.9548806897405201
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7100345550254943,
+   "mcc_abs": 0.7100395691082406,
+   "act_iou": 0.5558440272756437,
+   "act_iou_n_pairs": 5226,
+   "input_r2": 0.9864763906808702,
+   "matched_mcc_signed": {
+    "0.6": 0.9140679051534363,
+    "0.7": 0.9268814360788445,
+    "0.8": 0.9412475559290769,
+    "0.9": 0.9601380893987292
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.9133451086030592,
+    "0.7": 0.9258532549224415,
+    "0.8": 0.9412898087709165,
+    "0.9": 0.9600408694882144
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7342319438458106,
+   "mcc_abs": 0.7342323347015561,
+   "act_iou": 0.5085693976512834,
+   "act_iou_n_pairs": 5753,
+   "input_r2": 0.9855624676803498,
+   "matched_mcc_signed": {
+    "0.6": 0.9086645249276531,
+    "0.7": 0.9221123449445788,
+    "0.8": 0.9351129499415165,
+    "0.9": 0.9552017585915236
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.9072204411479154,
+    "0.7": 0.9215307840567548,
+    "0.8": 0.9349831544431315,
+    "0.9": 0.9552017585915236
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 33.32786611111111,
+  "l0_std": 0.2948563584439321,
+  "recon_r2_mean": 0.8716652297951913,
+  "recon_r2_std": 0.0015831722721497666,
+  "l1_per_token_mean": 158.36706801604535,
+  "l1_per_token_std": 0.5589012862198439,
+  "input_r2_mean": 0.985790818734845,
+  "input_r2_std": 0.0006046503290817276,
+  "mcc_signed_mean": 0.7260957967476341,
+  "mcc_signed_std": 0.0139098436378513,
+  "mcc_abs_mean": 0.7260977561374787,
+  "mcc_abs_std": 0.013907197959875077,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.9094615708294048,
+   "0.7": 0.9229925830109892,
+   "0.8": 0.9372405396793595,
+   "0.9": 0.956753026850321
+  },
+  "matched_mcc_signed_band": 0.9316119300925186,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.9084955527002805,
+   "0.7": 0.9222576637077555,
+   "0.8": 0.937174908049708,
+   "0.9": 0.9567077726067527
+  },
+  "matched_mcc_abs_band": 0.9311589742661242,
+  "act_iou_mean": 0.5236712023946376,
+  "act_iou_std": 0.02787987532137123,
+  "act_iou_n_pairs": 5582.333333333333,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x8/l0_32/dict_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x8/l0_32/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..87ab0bb5abe0cbfae6562413e7b1ca2256d2b6de
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x8/l0_32/dict_metrics_step_final.json
@@ -0,0 +1,595 @@
+{
+ "schema": 3,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 33,
+   "l0_raw": 33.19051518554687,
+   "recon_r2": 0.870863341941793,
+   "l1_per_token": 158.90994628863464,
+   "stats_source": "stats",
+   "n_dead": 1048,
+   "dead_rate": 0.17057291666666666,
+   "alive_source": "stats",
+   "n_dense": 35,
+   "dense_rate": 0.005696614583333333,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 1955,
+    "0.6": 899,
+    "0.7": 364,
+    "0.8": 76,
+    "0.9": 12
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 834,
+    "0.6": 392,
+    "0.7": 165,
+    "0.8": 34,
+    "0.9": 3
+   },
+   "n_dup_removed_signed": {
+    "0.5": 1121,
+    "0.6": 507,
+    "0.7": 199,
+    "0.8": 42,
+    "0.9": 9
+   },
+   "largest_dup_signed": {
+    "0.5": 11,
+    "0.6": 10,
+    "0.7": 8,
+    "0.8": 8,
+    "0.9": 8
+   },
+   "babel_mu1_signed": 0.5091859102249146,
+   "n_dup_abs": {
+    "0.5": 1957,
+    "0.6": 904,
+    "0.7": 375,
+    "0.8": 88,
+    "0.9": 12
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 834,
+    "0.6": 390,
+    "0.7": 168,
+    "0.8": 40,
+    "0.9": 3
+   },
+   "n_dup_removed_abs": {
+    "0.5": 1123,
+    "0.6": 514,
+    "0.7": 207,
+    "0.8": 48,
+    "0.9": 9
+   },
+   "largest_dup_abs": {
+    "0.5": 9,
+    "0.6": 9,
+    "0.7": 9,
+    "0.8": 8,
+    "0.9": 8
+   },
+   "babel_mu1_abs": 0.5099501609802246,
+   "dup_excludes_dead": true
+  },
+  "seed1": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 34,
+   "l0_raw": 33.65359697265625,
+   "recon_r2": 0.8731963945779327,
+   "l1_per_token": 158.003167719045,
+   "stats_source": "stats",
+   "n_dead": 423,
+   "dead_rate": 0.06884765625,
+   "alive_source": "stats",
+   "n_dense": 36,
+   "dense_rate": 0.005859375,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 2300,
+    "0.6": 1089,
+    "0.7": 437,
+    "0.8": 78,
+    "0.9": 10
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 964,
+    "0.6": 471,
+    "0.7": 195,
+    "0.8": 36,
+    "0.9": 3
+   },
+   "n_dup_removed_signed": {
+    "0.5": 1336,
+    "0.6": 618,
+    "0.7": 242,
+    "0.8": 42,
+    "0.9": 7
+   },
+   "largest_dup_signed": {
+    "0.5": 11,
+    "0.6": 8,
+    "0.7": 6,
+    "0.8": 6,
+    "0.9": 6
+   },
+   "babel_mu1_signed": 0.5157234072685242,
+   "n_dup_abs": {
+    "0.5": 2304,
+    "0.6": 1091,
+    "0.7": 449,
+    "0.8": 88,
+    "0.9": 10
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 963,
+    "0.6": 469,
+    "0.7": 200,
+    "0.8": 41,
+    "0.9": 3
+   },
+   "n_dup_removed_abs": {
+    "0.5": 1341,
+    "0.6": 622,
+    "0.7": 249,
+    "0.8": 47,
+    "0.9": 7
+   },
+   "largest_dup_abs": {
+    "0.5": 11,
+    "0.6": 8,
+    "0.7": 7,
+    "0.8": 6,
+    "0.9": 6
+   },
+   "babel_mu1_abs": 0.516669511795044,
+   "dup_excludes_dead": true
+  },
+  "seed2": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 33,
+   "l0_raw": 33.137987158203124,
+   "recon_r2": 0.8702214027804207,
+   "l1_per_token": 157.82196205586865,
+   "stats_source": "stats",
+   "n_dead": 1063,
+   "dead_rate": 0.17301432291666666,
+   "alive_source": "stats",
+   "n_dense": 35,
+   "dense_rate": 0.005696614583333333,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 1954,
+    "0.6": 889,
+    "0.7": 351,
+    "0.8": 74,
+    "0.9": 7
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 822,
+    "0.6": 382,
+    "0.7": 161,
+    "0.8": 34,
+    "0.9": 2
+   },
+   "n_dup_removed_signed": {
+    "0.5": 1132,
+    "0.6": 507,
+    "0.7": 190,
+    "0.8": 40,
+    "0.9": 5
+   },
+   "largest_dup_signed": {
+    "0.5": 8,
+    "0.6": 8,
+    "0.7": 5,
+    "0.8": 5,
+    "0.9": 5
+   },
+   "babel_mu1_signed": 0.5079862475395203,
+   "n_dup_abs": {
+    "0.5": 1961,
+    "0.6": 896,
+    "0.7": 360,
+    "0.8": 80,
+    "0.9": 7
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 821,
+    "0.6": 387,
+    "0.7": 166,
+    "0.8": 37,
+    "0.9": 2
+   },
+   "n_dup_removed_abs": {
+    "0.5": 1140,
+    "0.6": 509,
+    "0.7": 194,
+    "0.8": 43,
+    "0.9": 5
+   },
+   "largest_dup_abs": {
+    "0.5": 8,
+    "0.6": 8,
+    "0.7": 6,
+    "0.8": 5,
+    "0.9": 5
+   },
+   "babel_mu1_abs": 0.5088593363761902,
+   "dup_excludes_dead": true
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.7964083552360535,
+   "dcs_abs": 0.7974554896354675,
+   "frac_sign_flipped": 0.06966145833333333,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 5096,
+   "n_alive_b": 5721,
+   "matched_cos_signed": {
+    "0.5": 0.9245112538337708,
+    "0.6": 0.9304208159446716,
+    "0.7": 0.9323312044143677,
+    "0.8": 0.9338839650154114,
+    "0.9": 0.9343469738960266
+   },
+   "unique_frac_signed": {
+    "0.5": 0.6803385416666666,
+    "0.6": 0.7887369791666666,
+    "0.7": 0.8444010416666666,
+    "0.8": 0.8734537760416666,
+    "0.9": 0.8789876302083334
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.628980638187689,
+    "0.6": 0.7338573037219854,
+    "0.7": 0.78726144018583,
+    "0.8": 0.8157044756274747,
+    "0.9": 0.8212794323771959
+   },
+   "matched_frac_signed": {
+    "0.5": 0.62255859375,
+    "0.6": 0.71484375,
+    "0.7": 0.7490234375,
+    "0.8": 0.7452799479166666,
+    "0.9": 0.6896158854166666
+   },
+   "matched_cos_abs": {
+    "0.5": 0.9246366620063782,
+    "0.6": 0.9306555390357971,
+    "0.7": 0.9326257705688477,
+    "0.8": 0.9339529871940613,
+    "0.9": 0.9343586564064026
+   },
+   "unique_frac_abs": {
+    "0.5": 0.6797688802083334,
+    "0.6": 0.787841796875,
+    "0.7": 0.8431803385416666,
+    "0.8": 0.87255859375,
+    "0.9": 0.8789876302083334
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.6285392283316469,
+    "0.6": 0.7332093321456341,
+    "0.7": 0.7863717129609237,
+    "0.8": 0.8149287051346619,
+    "0.9": 0.8212897011593062
+   },
+   "matched_frac_abs": {
+    "0.5": 0.6220703125,
+    "0.6": 0.7140299479166666,
+    "0.7": 0.748046875,
+    "0.8": 0.7444661458333334,
+    "0.9": 0.6897786458333334
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.772930920124054,
+   "dcs_abs": 0.7739806175231934,
+   "frac_sign_flipped": 0.0732421875,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 5096,
+   "n_alive_b": 5081,
+   "matched_cos_signed": {
+    "0.5": 0.9057742357254028,
+    "0.6": 0.9083776473999023,
+    "0.7": 0.907649040222168,
+    "0.8": 0.9088093638420105,
+    "0.9": 0.9092308878898621
+   },
+   "unique_frac_signed": {
+    "0.5": 0.6448567708333334,
+    "0.6": 0.7456868489583334,
+    "0.7": 0.7965494791666666,
+    "0.8": 0.821533203125,
+    "0.9": 0.8270670572916666
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.5840946487539138,
+    "0.6": 0.6773652655538172,
+    "0.7": 0.7229873702550927,
+    "0.8": 0.7466170677071204,
+    "0.9": 0.7519949148457574
+   },
+   "matched_frac_signed": {
+    "0.5": 0.5986328125,
+    "0.6": 0.6853841145833334,
+    "0.7": 0.71826171875,
+    "0.8": 0.7171223958333334,
+    "0.9": 0.66552734375
+   },
+   "matched_cos_abs": {
+    "0.5": 0.9064953327178955,
+    "0.6": 0.9076434969902039,
+    "0.7": 0.9073496460914612,
+    "0.8": 0.9086093902587891,
+    "0.9": 0.9092737436294556
+   },
+   "unique_frac_abs": {
+    "0.5": 0.64404296875,
+    "0.6": 0.7449544270833334,
+    "0.7": 0.7955729166666666,
+    "0.8": 0.82080078125,
+    "0.9": 0.8270670572916666
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.5838219452416524,
+    "0.6": 0.6761530412962505,
+    "0.7": 0.7218628043774515,
+    "0.8": 0.7457872973755002,
+    "0.9": 0.7520303594161911
+   },
+   "matched_frac_abs": {
+    "0.5": 0.59765625,
+    "0.6": 0.68408203125,
+    "0.7": 0.71728515625,
+    "0.8": 0.7164713541666666,
+    "0.9": 0.6658528645833334
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.79576176404953,
+   "dcs_abs": 0.7968890070915222,
+   "frac_sign_flipped": 0.06917317708333333,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 5721,
+   "n_alive_b": 5081,
+   "matched_cos_signed": {
+    "0.5": 0.9272977113723755,
+    "0.6": 0.9314013719558716,
+    "0.7": 0.9333760142326355,
+    "0.8": 0.9352584481239319,
+    "0.9": 0.935468852519989
+   },
+   "unique_frac_signed": {
+    "0.5": 0.67822265625,
+    "0.6": 0.7875162760416666,
+    "0.7": 0.8439127604166666,
+    "0.8": 0.8723958333333334,
+    "0.9": 0.8780924479166666
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.6289143169415183,
+    "0.6": 0.7334937399427872,
+    "0.7": 0.7876879286777694,
+    "0.8": 0.8159155732331177,
+    "0.9": 0.8214281346590724
+   },
+   "matched_frac_signed": {
+    "0.5": 0.619140625,
+    "0.6": 0.7107747395833334,
+    "0.7": 0.7478841145833334,
+    "0.8": 0.7509765625,
+    "0.9": 0.6922200520833334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.9276136159896851,
+    "0.6": 0.931235134601593,
+    "0.7": 0.9334214925765991,
+    "0.8": 0.9352922439575195,
+    "0.9": 0.9354735612869263
+   },
+   "unique_frac_abs": {
+    "0.5": 0.6771647135416666,
+    "0.6": 0.7870279947916666,
+    "0.7": 0.843017578125,
+    "0.8": 0.8717447916666666,
+    "0.9": 0.8780924479166666
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.6281472085490046,
+    "0.6": 0.7329081206650395,
+    "0.7": 0.7868907260417473,
+    "0.8": 0.815336142356197,
+    "0.9": 0.821432269391759
+   },
+   "matched_frac_abs": {
+    "0.5": 0.6181640625,
+    "0.6": 0.71044921875,
+    "0.7": 0.7473958333333334,
+    "0.8": 0.75048828125,
+    "0.9": 0.6922200520833334
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 845,
+  "n_dead_std": 365.25105521179995,
+  "n_dense_mean": 35,
+  "n_dense_std": 0.5773502691896258,
+  "l0_mean": 33,
+  "l0_std": 0.28374208240625653,
+  "recon_r2_mean": 0.8714270464333822,
+  "recon_r2_std": 0.0015655561676906672,
+  "l1_per_token_mean": 158.2450253545161,
+  "l1_per_token_std": 0.5829226016922009,
+  "babel_mu1_signed_mean": 0.5109651883443197,
+  "babel_mu1_signed_std": 0.004164166559790241,
+  "n_dup_signed_per_threshold": {
+   "0.5": 2069.6666666666665,
+   "0.6": 959.0,
+   "0.7": 384.0,
+   "0.8": 76.0,
+   "0.9": 9.666666666666666
+  },
+  "n_dup_signed_band": 357.1666666666667,
+  "n_dup_signed_band_std": 40.30069271530371,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 873.3333333333334,
+   "0.6": 415.0,
+   "0.7": 173.66666666666666,
+   "0.8": 34.666666666666664,
+   "0.9": 2.6666666666666665
+  },
+  "n_dup_clusters_signed_band": 156.49999999999997,
+  "n_dup_clusters_signed_band_std": 17.206466807569765,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 10.0,
+   "0.6": 8.666666666666666,
+   "0.7": 6.333333333333333,
+   "0.8": 6.333333333333333,
+   "0.9": 6.333333333333333
+  },
+  "largest_dup_signed_band": 6.916666666666666,
+  "largest_dup_signed_band_std": 1.4215601757693317,
+  "dcs_signed_mean": 0.7883670131365458,
+  "dcs_signed_std": 0.013371957435556976,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.6139965346277071,
+   "0.6": 0.7149054364061965,
+   "0.7": 0.765978913039564,
+   "0.8": 0.792745705522571,
+   "0.9": 0.7982341606273419
+  },
+  "dedup_dcs_signed_band": 0.7679660538989184,
+  "dedup_dcs_signed_band_std": 0.03743389817126972,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.9191944003105164,
+   "0.6": 0.9233999451001486,
+   "0.7": 0.9244520862897238,
+   "0.8": 0.9259839256604513,
+   "0.9": 0.9263489047686259
+  },
+  "matched_cos_signed_band": 0.9250462154547373,
+  "matched_cos_signed_band_std": 0.014326104360468571,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.6678059895833334,
+   "0.6": 0.7739800347222222,
+   "0.7": 0.8282877604166666,
+   "0.8": 0.8557942708333334,
+   "0.9": 0.8613823784722222
+  },
+  "unique_frac_signed_band": 0.8298611111111112,
+  "unique_frac_signed_band_std": 0.027848180159573337,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.6134440104166666,
+   "0.6": 0.7036675347222223,
+   "0.7": 0.7383897569444445,
+   "0.8": 0.73779296875,
+   "0.9": 0.6824544270833334
+  },
+  "matched_frac_signed_band": 0.7155761718750001,
+  "matched_frac_signed_band_std": 0.016460995438718214,
+  "babel_mu1_abs_mean": 0.5118263363838196,
+  "babel_mu1_abs_std": 0.004229625944112309,
+  "n_dup_abs_per_threshold": {
+   "0.5": 2074.0,
+   "0.6": 963.6666666666666,
+   "0.7": 394.6666666666667,
+   "0.8": 85.33333333333333,
+   "0.9": 9.666666666666666
+  },
+  "n_dup_abs_band": 363.3333333333333,
+  "n_dup_abs_band_std": 40.23395125181385,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 872.6666666666666,
+   "0.6": 415.3333333333333,
+   "0.7": 178.0,
+   "0.8": 39.333333333333336,
+   "0.9": 2.6666666666666665
+  },
+  "n_dup_clusters_abs_band": 158.83333333333331,
+  "n_dup_clusters_abs_band_std": 16.852917650464363,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 9.333333333333334,
+   "0.6": 8.333333333333334,
+   "0.7": 7.333333333333333,
+   "0.8": 6.333333333333333,
+   "0.9": 6.333333333333333
+  },
+  "largest_dup_abs_band": 7.083333333333333,
+  "largest_dup_abs_band_std": 1.282900359861721,
+  "dcs_abs_mean": 0.789441704750061,
+  "dcs_abs_std": 0.013392689770757657,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.6135027940407679,
+   "0.6": 0.714090164702308,
+   "0.7": 0.7650417477933741,
+   "0.8": 0.7920173816221197,
+   "0.9": 0.7982507766557521
+  },
+  "dedup_dcs_abs_band": 0.7673500176933884,
+  "dedup_dcs_abs_band_std": 0.037578386912909816,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.9195818702379862,
+   "0.6": 0.9231780568758646,
+   "0.7": 0.9244656364123026,
+   "0.8": 0.9259515404701233,
+   "0.9": 0.9263686537742615
+  },
+  "matched_cos_abs_band": 0.924990971883138,
+  "matched_cos_abs_band_std": 0.01453277942708945,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.6669921875,
+   "0.6": 0.7732747395833334,
+   "0.7": 0.8272569444444443,
+   "0.8": 0.8550347222222222,
+   "0.9": 0.8613823784722222
+  },
+  "unique_frac_abs_band": 0.8292371961805556,
+  "unique_frac_abs_band_std": 0.02783469570956975,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.6126302083333334,
+   "0.6": 0.7028537326388888,
+   "0.7": 0.7375759548611112,
+   "0.8": 0.7371419270833334,
+   "0.9": 0.6826171875
+  },
+  "matched_frac_abs_band": 0.7150472005208334,
+  "matched_frac_abs_band_std": 0.016570617158199297,
+  "alive_source": "stats",
+  "frac_denominator": "d_sae",
+  "d_sae": 6144
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x8/l0_64/code_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x8/l0_64/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..ea27abd49b65b100a3135e08528ce69d61b64870
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x8/l0_64/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 67.07978166666666,
+   "recon_r2": 0.9007928260927786,
+   "l1_per_token": 207.56294368619183
+  },
+  "seed1": {
+   "l0": 67.32586333333333,
+   "recon_r2": 0.9010973484326059,
+   "l1_per_token": 208.35368363583228
+  },
+  "seed2": {
+   "l0": 66.89205,
+   "recon_r2": 0.9004301949762551,
+   "l1_per_token": 206.5687951346374
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.8357420690903994,
+   "mcc_abs": 0.8357453816743017,
+   "act_iou": 0.5267911721315092,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.986135886258377,
+   "matched_mcc_signed": {
+    "0.6": 0.894364347449518,
+    "0.7": 0.9150689179129058,
+    "0.8": 0.9338353385226013,
+    "0.9": 0.9569653778346705
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8929147024633536,
+    "0.7": 0.91401995044166,
+    "0.8": 0.9327024527160549,
+    "0.9": 0.9569884466822652
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.838598835127209,
+   "mcc_abs": 0.8386028905688351,
+   "act_iou": 0.5284520329511887,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9861917363932784,
+   "matched_mcc_signed": {
+    "0.6": 0.8982961385300279,
+    "0.7": 0.9163212835862944,
+    "0.8": 0.9333764649973272,
+    "0.9": 0.9571993105077691
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8972900347866817,
+    "0.7": 0.9153689288617797,
+    "0.8": 0.9336187884871553,
+    "0.9": 0.9572008434044076
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.8370507941731377,
+   "mcc_abs": 0.8370577443076211,
+   "act_iou": 0.5276884558823997,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9862036006577758,
+   "matched_mcc_signed": {
+    "0.6": 0.8963613437872693,
+    "0.7": 0.9143875150561375,
+    "0.8": 0.9357280307811799,
+    "0.9": 0.958058302414559
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8946383374162487,
+    "0.7": 0.9141294782897743,
+    "0.8": 0.935446569349231,
+    "0.9": 0.9580332155147856
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 67.09923166666665,
+  "l0_std": 0.21755971345689085,
+  "recon_r2_mean": 0.9007734565005466,
+  "recon_r2_std": 0.0003339982326414893,
+  "l1_per_token_mean": 207.4951408188872,
+  "l1_per_token_std": 0.8943738938695984,
+  "input_r2_mean": 0.9861770744364771,
+  "input_r2_std": 3.6159918479135776e-05,
+  "mcc_signed_mean": 0.8371305661302486,
+  "mcc_signed_std": 0.001430052698722404,
+  "mcc_abs_mean": 0.8371353388502526,
+  "mcc_abs_std": 0.0014303338615054239,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8963406099222717,
+   "0.7": 0.9152592388517792,
+   "0.8": 0.9343132781003695,
+   "0.9": 0.9574076635856662
+  },
+  "matched_mcc_signed_band": 0.9258301976150216,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.8949476915554281,
+   "0.7": 0.914506119197738,
+   "0.8": 0.9339226035174804,
+   "0.9": 0.9574075018671527
+  },
+  "matched_mcc_abs_band": 0.9251959790344498,
+  "act_iou_mean": 0.5276438869883658,
+  "act_iou_std": 0.0008313269244546277,
+  "act_iou_n_pairs": 6144.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x8/l0_64/dict_metrics_step_final.json b/lm/gpt2-small_layer8/relu/x8/l0_64/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..6d8bcd307961c4fc56b18f54c2f05d0cfaa8cac6
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x8/l0_64/dict_metrics_step_final.json
@@ -0,0 +1,595 @@
+{
+ "schema": 3,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 67,
+   "l0_raw": 67.09376083984375,
+   "recon_r2": 0.9006148371493142,
+   "l1_per_token": 207.49205339636524,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 41,
+   "dense_rate": 0.006673177083333333,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 1797,
+    "0.6": 814,
+    "0.7": 300,
+    "0.8": 53,
+    "0.9": 11
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 740,
+    "0.6": 353,
+    "0.7": 135,
+    "0.8": 23,
+    "0.9": 4
+   },
+   "n_dup_removed_signed": {
+    "0.5": 1057,
+    "0.6": 461,
+    "0.7": 165,
+    "0.8": 30,
+    "0.9": 7
+   },
+   "largest_dup_signed": {
+    "0.5": 11,
+    "0.6": 6,
+    "0.7": 5,
+    "0.8": 5,
+    "0.9": 5
+   },
+   "babel_mu1_signed": 0.46985113620758057,
+   "n_dup_abs": {
+    "0.5": 1872,
+    "0.6": 862,
+    "0.7": 334,
+    "0.8": 61,
+    "0.9": 11
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 777,
+    "0.6": 376,
+    "0.7": 153,
+    "0.8": 27,
+    "0.9": 4
+   },
+   "n_dup_removed_abs": {
+    "0.5": 1095,
+    "0.6": 486,
+    "0.7": 181,
+    "0.8": 34,
+    "0.9": 7
+   },
+   "largest_dup_abs": {
+    "0.5": 10,
+    "0.6": 6,
+    "0.7": 5,
+    "0.8": 5,
+    "0.9": 5
+   },
+   "babel_mu1_abs": 0.47586241364479065,
+   "dup_excludes_dead": true
+  },
+  "seed1": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 67,
+   "l0_raw": 67.30862905273437,
+   "recon_r2": 0.9009085995812536,
+   "l1_per_token": 208.1974304222417,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 42,
+   "dense_rate": 0.0068359375,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 1759,
+    "0.6": 802,
+    "0.7": 299,
+    "0.8": 44,
+    "0.9": 8
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 733,
+    "0.6": 353,
+    "0.7": 137,
+    "0.8": 21,
+    "0.9": 3
+   },
+   "n_dup_removed_signed": {
+    "0.5": 1026,
+    "0.6": 449,
+    "0.7": 162,
+    "0.8": 23,
+    "0.9": 5
+   },
+   "largest_dup_signed": {
+    "0.5": 11,
+    "0.6": 6,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_signed": 0.46892330050468445,
+   "n_dup_abs": {
+    "0.5": 1823,
+    "0.6": 848,
+    "0.7": 327,
+    "0.8": 56,
+    "0.9": 8
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 760,
+    "0.6": 373,
+    "0.7": 151,
+    "0.8": 27,
+    "0.9": 3
+   },
+   "n_dup_removed_abs": {
+    "0.5": 1063,
+    "0.6": 475,
+    "0.7": 176,
+    "0.8": 29,
+    "0.9": 5
+   },
+   "largest_dup_abs": {
+    "0.5": 10,
+    "0.6": 6,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_abs": 0.474450021982193,
+   "dup_excludes_dead": true
+  },
+  "seed2": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 67,
+   "l0_raw": 66.92253725585938,
+   "recon_r2": 0.9245507933825287,
+   "l1_per_token": 206.50177240079768,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 35,
+   "dense_rate": 0.005696614583333333,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 1787,
+    "0.6": 776,
+    "0.7": 288,
+    "0.8": 50,
+    "0.9": 10
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 741,
+    "0.6": 338,
+    "0.7": 130,
+    "0.8": 23,
+    "0.9": 3
+   },
+   "n_dup_removed_signed": {
+    "0.5": 1046,
+    "0.6": 438,
+    "0.7": 158,
+    "0.8": 27,
+    "0.9": 7
+   },
+   "largest_dup_signed": {
+    "0.5": 11,
+    "0.6": 9,
+    "0.7": 6,
+    "0.8": 6,
+    "0.9": 6
+   },
+   "babel_mu1_signed": 0.46939826011657715,
+   "n_dup_abs": {
+    "0.5": 1852,
+    "0.6": 809,
+    "0.7": 322,
+    "0.8": 60,
+    "0.9": 12
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 776,
+    "0.6": 352,
+    "0.7": 147,
+    "0.8": 28,
+    "0.9": 4
+   },
+   "n_dup_removed_abs": {
+    "0.5": 1076,
+    "0.6": 457,
+    "0.7": 175,
+    "0.8": 32,
+    "0.9": 8
+   },
+   "largest_dup_abs": {
+    "0.5": 9,
+    "0.6": 9,
+    "0.7": 6,
+    "0.8": 6,
+    "0.9": 6
+   },
+   "babel_mu1_abs": 0.4753236770629883,
+   "dup_excludes_dead": true
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.8904410004615784,
+   "dcs_abs": 0.8906774520874023,
+   "frac_sign_flipped": 0.00439453125,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.8870011568069458,
+    "0.6": 0.889115035533905,
+    "0.7": 0.8894314765930176,
+    "0.8": 0.891057550907135,
+    "0.9": 0.8905840516090393
+   },
+   "unique_frac_signed": {
+    "0.5": 0.8304850260416666,
+    "0.6": 0.9259440104166666,
+    "0.7": 0.973388671875,
+    "0.8": 0.9956868489583334,
+    "0.9": 0.9990234375
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7366411788098048,
+    "0.6": 0.823270741724021,
+    "0.7": 0.8657625237246975,
+    "0.8": 0.887214285103255,
+    "0.9": 0.8897143406211399
+   },
+   "matched_frac_signed": {
+    "0.5": 0.7683919270833334,
+    "0.6": 0.8388671875,
+    "0.7": 0.8505859375,
+    "0.8": 0.8282877604166666,
+    "0.9": 0.73974609375
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8883788585662842,
+    "0.6": 0.8902012705802917,
+    "0.7": 0.8901354670524597,
+    "0.8": 0.891228973865509,
+    "0.9": 0.8908168077468872
+   },
+   "unique_frac_abs": {
+    "0.5": 0.8243815104166666,
+    "0.6": 0.9217936197916666,
+    "0.7": 0.970947265625,
+    "0.8": 0.994873046875,
+    "0.9": 0.9990234375
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7323631052471076,
+    "0.6": 0.820581851551348,
+    "0.7": 0.864274597770418,
+    "0.8": 0.8866596846928587,
+    "0.9": 0.8899468694580719
+   },
+   "matched_frac_abs": {
+    "0.5": 0.76318359375,
+    "0.6": 0.83642578125,
+    "0.7": 0.8494466145833334,
+    "0.8": 0.8284505208333334,
+    "0.9": 0.7399088541666666
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.8917345404624939,
+   "dcs_abs": 0.8918420672416687,
+   "frac_sign_flipped": 0.0026041666666666665,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.8859454989433289,
+    "0.6": 0.8901175856590271,
+    "0.7": 0.889983594417572,
+    "0.8": 0.8918033838272095,
+    "0.9": 0.8916692137718201
+   },
+   "unique_frac_signed": {
+    "0.5": 0.828857421875,
+    "0.6": 0.9268391927083334,
+    "0.7": 0.9737141927083334,
+    "0.8": 0.995361328125,
+    "0.9": 0.9988606770833334
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7343225021759281,
+    "0.6": 0.8249958645077035,
+    "0.7": 0.8665896571619669,
+    "0.8": 0.8876666005526204,
+    "0.9": 0.8906533146024838
+   },
+   "matched_frac_signed": {
+    "0.5": 0.76611328125,
+    "0.6": 0.83935546875,
+    "0.7": 0.8525390625,
+    "0.8": 0.8331705729166666,
+    "0.9": 0.7439778645833334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8873385787010193,
+    "0.6": 0.8913217782974243,
+    "0.7": 0.890244722366333,
+    "0.8": 0.8918389081954956,
+    "0.9": 0.891872763633728
+   },
+   "unique_frac_abs": {
+    "0.5": 0.8233235677083334,
+    "0.6": 0.9232584635416666,
+    "0.7": 0.9710286458333334,
+    "0.8": 0.99462890625,
+    "0.9": 0.998779296875
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7305667643813649,
+    "0.6": 0.822920375552106,
+    "0.7": 0.8644531272196522,
+    "0.8": 0.88704875780968,
+    "0.9": 0.890784051764058
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7605794270833334,
+    "0.6": 0.8367513020833334,
+    "0.7": 0.8507486979166666,
+    "0.8": 0.8312174479166666,
+    "0.9": 0.7439778645833334
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.8901703953742981,
+   "dcs_abs": 0.8903875946998596,
+   "frac_sign_flipped": 0.004069010416666667,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.8839718103408813,
+    "0.6": 0.8879483938217163,
+    "0.7": 0.8885824680328369,
+    "0.8": 0.8903875946998596,
+    "0.9": 0.8903201222419739
+   },
+   "unique_frac_signed": {
+    "0.5": 0.8313802083333334,
+    "0.6": 0.9278157552083334,
+    "0.7": 0.9739583333333334,
+    "0.8": 0.9959309895833334,
+    "0.9": 0.9990234375
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7349166678419958,
+    "0.6": 0.8238525095997223,
+    "0.7": 0.8654422995944818,
+    "0.8": 0.8867645983021551,
+    "0.9": 0.889450668997597
+   },
+   "matched_frac_signed": {
+    "0.5": 0.767578125,
+    "0.6": 0.8369140625,
+    "0.7": 0.8505859375,
+    "0.8": 0.82568359375,
+    "0.9": 0.74072265625
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8842839002609253,
+    "0.6": 0.8890949487686157,
+    "0.7": 0.88875412940979,
+    "0.8": 0.8904471397399902,
+    "0.9": 0.8905950784683228
+   },
+   "unique_frac_abs": {
+    "0.5": 0.825927734375,
+    "0.6": 0.9241536458333334,
+    "0.7": 0.971435546875,
+    "0.8": 0.9950358072916666,
+    "0.9": 0.9989420572916666
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7303545982867945,
+    "0.6": 0.821660338396517,
+    "0.7": 0.8633673537406139,
+    "0.8": 0.8860267885417367,
+    "0.9": 0.8896528798989796
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7626953125,
+    "0.6": 0.8351236979166666,
+    "0.7": 0.8486328125,
+    "0.8": 0.8248697916666666,
+    "0.9": 0.7405598958333334
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 39,
+  "n_dense_std": 3.7859388972001824,
+  "l0_mean": 67,
+  "l0_std": 0.19345660160934608,
+  "recon_r2_mean": 0.9086914100376987,
+  "recon_r2_std": 0.013735414232934668,
+  "l1_per_token_mean": 207.39708540646822,
+  "l1_per_token_std": 0.8518087935391796,
+  "babel_mu1_signed_mean": 0.4693908989429474,
+  "babel_mu1_signed_std": 0.0004639616504088339,
+  "n_dup_signed_per_threshold": {
+   "0.5": 1781.0,
+   "0.6": 797.3333333333334,
+   "0.7": 295.6666666666667,
+   "0.8": 49.0,
+   "0.9": 9.666666666666666
+  },
+  "n_dup_signed_band": 287.9166666666667,
+  "n_dup_signed_band_std": 6.756170019569766,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 738.0,
+   "0.6": 348.0,
+   "0.7": 134.0,
+   "0.8": 22.333333333333332,
+   "0.9": 3.3333333333333335
+  },
+  "n_dup_clusters_signed_band": 126.91666666666666,
+  "n_dup_clusters_signed_band_std": 2.9615592739861434,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 11.0,
+   "0.6": 7.0,
+   "0.7": 5.0,
+   "0.8": 5.0,
+   "0.9": 5.0
+  },
+  "largest_dup_signed_band": 5.5,
+  "largest_dup_signed_band_std": 1.14564392373896,
+  "dcs_signed_mean": 0.8907819787661234,
+  "dcs_signed_std": 0.000835964782368009,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.735293449609243,
+   "0.6": 0.824039705277149,
+   "0.7": 0.8659314934937155,
+   "0.8": 0.8872151613193435,
+   "0.9": 0.8899394414070736
+  },
+  "dedup_dcs_signed_band": 0.8667814503743204,
+  "dedup_dcs_signed_band_std": 0.0006044529314523477,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.885639488697052,
+   "0.6": 0.8890603383382162,
+   "0.7": 0.8893325130144755,
+   "0.8": 0.8910828431447347,
+   "0.9": 0.8908577958742777
+  },
+  "matched_cos_signed_band": 0.890083372592926,
+  "matched_cos_signed_band_std": 0.0007925251094893675,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.8302408854166666,
+   "0.6": 0.9268663194444445,
+   "0.7": 0.9736870659722223,
+   "0.8": 0.9956597222222223,
+   "0.9": 0.9989691840277778
+  },
+  "unique_frac_signed_band": 0.9737955729166667,
+  "unique_frac_signed_band_std": 0.00034706058978741823,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.7673611111111112,
+   "0.6": 0.83837890625,
+   "0.7": 0.8512369791666666,
+   "0.8": 0.8290473090277777,
+   "0.9": 0.7414822048611112
+  },
+  "matched_frac_signed_band": 0.8150363498263888,
+  "matched_frac_signed_band_std": 0.0019776953420472827,
+  "babel_mu1_abs_mean": 0.475212037563324,
+  "babel_mu1_abs_std": 0.0007127833370371501,
+  "n_dup_abs_per_threshold": {
+   "0.5": 1849.0,
+   "0.6": 839.6666666666666,
+   "0.7": 327.6666666666667,
+   "0.8": 59.0,
+   "0.9": 10.333333333333334
+  },
+  "n_dup_abs_band": 309.16666666666663,
+  "n_dup_abs_band_std": 8.140689978947321,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 771.0,
+   "0.6": 367.0,
+   "0.7": 150.33333333333334,
+   "0.8": 27.333333333333332,
+   "0.9": 3.6666666666666665
+  },
+  "n_dup_clusters_abs_band": 137.08333333333334,
+  "n_dup_clusters_abs_band_std": 3.826987501068345,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 9.666666666666666,
+   "0.6": 7.0,
+   "0.7": 5.0,
+   "0.8": 5.0,
+   "0.9": 5.0
+  },
+  "largest_dup_abs_band": 5.5,
+  "largest_dup_abs_band_std": 1.14564392373896,
+  "dcs_abs_mean": 0.8909690380096436,
+  "dcs_abs_std": 0.0007698307322107153,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.7310948226384223,
+   "0.6": 0.821720855166657,
+   "0.7": 0.8640316929102281,
+   "0.8": 0.8865784103480917,
+   "0.9": 0.8901279337070364
+  },
+  "dedup_dcs_abs_band": 0.8656147230330032,
+  "dedup_dcs_abs_band_std": 0.0006022866540436029,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.8866671125094095,
+   "0.6": 0.890205999215444,
+   "0.7": 0.8897114396095276,
+   "0.8": 0.8911716739336649,
+   "0.9": 0.8910948832829794
+  },
+  "matched_cos_abs_band": 0.890545999010404,
+  "matched_cos_abs_band_std": 0.0007995156813636146,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.8245442708333334,
+   "0.6": 0.9230685763888888,
+   "0.7": 0.9711371527777778,
+   "0.8": 0.9948459201388888,
+   "0.9": 0.9989149305555555
+  },
+  "unique_frac_abs_band": 0.9719916449652777,
+  "unique_frac_abs_band_std": 0.0003708905412835067,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.7621527777777778,
+   "0.6": 0.8361002604166666,
+   "0.7": 0.849609375,
+   "0.8": 0.8281792534722222,
+   "0.9": 0.7414822048611112
+  },
+  "matched_frac_abs_band": 0.8138427734375,
+  "matched_frac_abs_band_std": 0.0017065606262622833,
+  "alive_source": "stats",
+  "frac_denominator": "d_sae",
+  "d_sae": 6144
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x8/l0_64/seed0/config.json b/lm/gpt2-small_layer8/relu/x8/l0_64/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..03af2764551fb0c668e41544170a64c51da927c1
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x8/l0_64/seed0/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x8/l0_64/seed1/config.json b/lm/gpt2-small_layer8/relu/x8/l0_64/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..2df92766c911a9cf15a36365f5f0d91bde86a2e6
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x8/l0_64/seed1/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/relu/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..ba0dd53599d86dbe49bd6883676c3e0b8222e5cf
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 67.30862905273437,
+  "l0_per_token_0.001": 67.27729438476563,
+  "l0_per_token_0.01": 66.99604287109375,
+  "l0_per_token_0.05": 65.7628591796875,
+  "l0_per_token_0.1": 64.26012250976562,
+  "l0_per_token_0.5": 53.620801416015624,
+  "l0_per_token_1.0": 43.169892236328124,
+  "l1_per_token": 208.1974304222417,
+  "r2_score": 0.9009085995812536,
+  "l2_loss": 2.3239345298252547
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/relu/x8/l0_64/seed2/config.json b/lm/gpt2-small_layer8/relu/x8/l0_64/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..6ff526fe86b1fee16310adf1b05c81222284af8c
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x8/l0_64/seed2/config.json
@@ -0,0 +1,10 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "relu",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "class_type": "VanillaSAE",
+    "module_name": "fastsae.models.sae"
+}
diff --git a/lm/gpt2-small_layer8/relu/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/relu/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/relu/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/config.json b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..84aeb0f446bc7d69e204efd84c8877a8b6b471d5
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/model.safetensors b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..399286dc1a705dfe1de663a4857ea213e1b8eaaf
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e30a167ed9813f06461038de33746c33dee61463b16c9bc5732e9fb2ab1ccf6
+size 75550008
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..643abbdc3ec3b135bd74bd77a201fcb5b2af3cfa
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:830c87baaa651011bc2002d59946156732123b0c167e1f9dbe04b206a0fcfb45
+size 99648
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b58279c0d01d25e2bc1ca7bcb2f8a7c526f31135
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3fef96977b26f2cd68b0b01561cbf2ceab5aa49455b9d8d7602f73c52560f35a
+size 50471
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..f63f6bf30756aaa3c53a8803b789beacbd59fe63
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 127.99712387695313,
+  "l0_per_token_0.001": 127.9970890625,
+  "l0_per_token_0.01": 127.99690952148437,
+  "l0_per_token_0.05": 127.99579125976562,
+  "l0_per_token_0.1": 127.99433369140625,
+  "l0_per_token_0.5": 127.9272203125,
+  "l0_per_token_1.0": 127.74159506835937,
+  "l1_per_token": 589.6484688185383,
+  "r2_score": 0.9535042271919859,
+  "l2_loss": 1.0904390437801526
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c97e8dc5024e601f9948e3158bd84e82c6686301
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:64897fe3f5c7123f553404b8d43f0e76a860db5dbd944577294baebf928266e5
+size 50382
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..bf8b341eaf42f42a8897c7f9ad6abef89059d3d2
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e11253e1cbadb31f294b2b3fea5c67275d2739ee38e680fa159cda9b8762992d
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..117c0fbbf42d7e75957d1cb4da0b1b60d4805998
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b692591dfdcb01bdbe9273d3ee605649a3d1ff153b2e5e1954b3505552b9d32
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/config.json b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..d879da4919f95086b4f9449f0f09e9218fc9c99e
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/model.safetensors b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b5ffde731fd500aaff02cb71e3e8794e72521271
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff4671485b566949b658c2c2ad47b027c2c9cbce5e128abc480e32faae8107a0
+size 75550008
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9d849b8237529314bd80f4c0c8c9da3c8c10d2ed
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ccdac8ce3d32a625b23da2bec3f95a748ddcdc584de1552e3e3bb101788bd77e
+size 99648
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..59b78a6b16e4c7915fb58b27d11a01fd0065967a
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e978b12b507003f4ac60f778efe3a2a492c8ff2569e70849c19edd434a327355
+size 50471
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..463913462bf37eef28def63a922741b7af66acfe
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 127.99949423828124,
+  "l0_per_token_0.001": 127.99948979492187,
+  "l0_per_token_0.01": 127.99936943359376,
+  "l0_per_token_0.05": 127.99896567382812,
+  "l0_per_token_0.1": 127.99800229492187,
+  "l0_per_token_0.5": 127.93828989257813,
+  "l0_per_token_1.0": 127.7674515625,
+  "l1_per_token": 589.2194232140833,
+  "r2_score": 0.9535554051583998,
+  "l2_loss": 1.0892387958997847
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8f1da45434fb30eb4d457e22a7596a2bb2076dc1
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c454c42e430c46c4b14c17d6089dc3ba22135fd962d2b8aafe76468deeccb534
+size 50382
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..783dfbfb00b584e82aee9f6df414f8e090a403e2
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54f18259a2e8d8c3c1bb933c1f6bdd91af9d385df35e2627a04f5b7ec45704d1
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..35af8005a0d2dbbf7b09b5afb7c2767df694a77c
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3c5db0ab88bf8ce8e388b1cc03d34b074a0381ac993274d6fdb1d8e9553b442
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/config.json b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7f324f213a37a81da12e91e56716d2c9d4396801
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/model.safetensors b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..11413a7a67f8df2a6b0acdf1dd8ed07f597b67cc
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f84dd9cc4775a03e3b25ebd9895a8ec5f7a6a91a7b1c3b331583fe60bafb24d8
+size 75550008
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b076b3d46437a84ce5cdbbeafe4e7ea3d1ed0268
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:39fc59ad80dddb40f077ffe665f8faba406b61e553d0a1129e27a7f24028557b
+size 99648
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..981eede764a0ddc977ee2699648c733954c6cb12
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63aad0b09465321fa0352f3047f9c9650f60e9c53e6236e7ef58f02c20cc7984
+size 50471
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..7701a1c68d76004a3a8120b858fd5644ae26ab3c
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 127.99571987304688,
+  "l0_per_token_0.001": 127.99568291015625,
+  "l0_per_token_0.01": 127.99537001953125,
+  "l0_per_token_0.05": 127.99397299804687,
+  "l0_per_token_0.1": 127.9921455078125,
+  "l0_per_token_0.5": 127.9279611328125,
+  "l0_per_token_1.0": 127.7515001953125,
+  "l1_per_token": 588.8183802243086,
+  "r2_score": 0.9534961410314295,
+  "l2_loss": 1.0906286755513468
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2d19ee13978da6b93da941547a73431ecef90339
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:577d2521b6ebb2f838652ad1afd1e4a3601a0be0d4839e96a431e0501248bdce
+size 50382
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..617e3998dde3a24e23bf45c0b127b8b6c26d5ac2
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bcb8fcfbd617dd75b9ca8706e03dd0cb9babb381340020194754c981e22e962b
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..bc875c22f08cdc0e5e8b2cf29245df2909499c7e
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84ddc747a2e7727fd9ffffc3ebeb4a310a87830b5aae55b198668fb72abfd460
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/config.json b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..ab73283ef40e57bddc1476210efeeb8b48bcea00
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/model.safetensors b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d11523942ebd7dea66913ac81053f52ee6d2d410
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:961eda4e7839eea6309c68ed20b68215f8ecaf625cac9e506489a53c0c50911a
+size 75550008
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4c1a70877f677ffce1c09f32b1cd77bbce77501f
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae3e5d62d329618d36514c5eae841df5e57f89119a77a5adf1ab88c108a0274f
+size 99648
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..083e419304ed73c9676e3454f3c96decf61b9773
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3954c497031a063e2704814d08d3def3cebed24651c0d608fa7f73fc8fe8d688
+size 50471
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..748fa85a2c4238bff43f02f5c479162c4b82fbdf
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 31.999996875,
+  "l0_per_token_0.001": 31.999996826171873,
+  "l0_per_token_0.01": 31.99999482421875,
+  "l0_per_token_0.05": 31.999972265625,
+  "l0_per_token_0.1": 31.99985234375,
+  "l0_per_token_0.5": 31.99104384765625,
+  "l0_per_token_1.0": 31.987072509765625,
+  "l1_per_token": 303.2671258173008,
+  "r2_score": 0.9187235385163666,
+  "l2_loss": 1.9061308353206563
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3c9039429b4373c578d0640e085a7855572ef438
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72de035056d5aa59b7a1ac707cb04ced0fcaec5df4446495a5ed2523e122c9b1
+size 50382
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..92414c457bf9d19e8cc6cb36f0757e18bf1cd947
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:010bbc5fc64d494ea86dc91014de22a0643f19e59e7a6b28c326ffcceca3acae
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3fefc08084e9a908476ad1a2e19bc05193293539
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:952685a85d83d9a7b3e7bebfdf26de8baeb7c93fefd8bbc213a3e1fe2eef74b0
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/config.json b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0d3a43942ecb72bb8c80a4114780db11cce08c02
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/model.safetensors b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7f229b27083750e663fc94f0fa1d3debe9a21a39
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f39828dbd74b7cec2663c27efc0cca0d3bafb30563c15ec6667e3f2c68608a5d
+size 75550008
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..51268f68332edd14e7e5615d3b4883189f82d635
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7fd32dca1816151cf5c2c48af9b5dce9feba3d5d22ddb2875aeb31bfc4f24039
+size 99648
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..469eed2c095aafb81e35df84e13467aa30668879
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:afa859d6c5ef99bf71309cb616e2dca93462d91f1ee470c10b95aa8e613b7887
+size 50471
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..b49871da299013ebc8c3ec4b8d8322f019250ebb
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 31.999999365234373,
+  "l0_per_token_0.001": 31.999999365234373,
+  "l0_per_token_0.01": 31.9999986328125,
+  "l0_per_token_0.05": 31.9999873046875,
+  "l0_per_token_0.1": 31.999914599609376,
+  "l0_per_token_0.5": 31.99065322265625,
+  "l0_per_token_1.0": 31.98571650390625,
+  "l1_per_token": 302.4908204021793,
+  "r2_score": 0.9187593952479948,
+  "l2_loss": 1.9052899076945924
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f9aff7637b73344b4a45368706575697a302da54
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7e750088d5933d86078cc433ee14d6db203dc3cfe7c8057bd2071c4dee0c5597
+size 50382
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9fde2d0590dfa250b829e39c38a144b0c2e36daa
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77f5afd11d8f3286327fb51e43b4ab748d35ebf94448371c100644db8e1cbbae
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3e93214f67b69265f7453852c421386cb58ac7b2
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f644505edbe1b3ef3f24f2a35687a703e96de33937c66c628024ef087cfa905
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/config.json b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..76a481dabef8665b8a11108a8ca26d94b7e3a872
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 16,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/model.safetensors b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8a4ecf48f2708f604000312b8d26f1e31f301d05
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca277a9c713c2410fa6fdc4dce0d2415010d27972d0d097d753bbdbf1c4cee65
+size 75550008
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d1b36f7008962670d03f8e9c357ca6743ce978be
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e142494a165405ba862577ac1a8966695a2af5debc648d2bd2230572ddcfb56d
+size 99648
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..ada0a38faffafc32c2887b070cbe5ac8d3faed5f
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e28dcfaeb9df7620bc3fd8011f9db3baec0d9dccd4af1c7c6419d8a7b59a59a
+size 50471
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..83582a76c40a99dceae135c465ac4ebf905ffd1e
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 31.99999990234375,
+  "l0_per_token_0.001": 31.999999853515625,
+  "l0_per_token_0.01": 31.99999970703125,
+  "l0_per_token_0.05": 31.999994482421876,
+  "l0_per_token_0.1": 31.999949169921877,
+  "l0_per_token_0.5": 31.991566162109375,
+  "l0_per_token_1.0": 31.98709970703125,
+  "l1_per_token": 299.460248829974,
+  "r2_score": 0.9186805598122308,
+  "l2_loss": 1.9071387905357322
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..849bc805cdf72ce6d3fb339e4c1faf3efb0085eb
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:add9440014a3f1976ddce3a5c73543802cd93762937180057b4fa61472b4beb7
+size 50382
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..225b8e185c73732a09a11a8e1eac2474799402e1
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dfc8f3007d0cce8f0f21a5fb9df54b355829647a58ab68918fec9a869ad6a3e8
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4b8faca25d86d593e8a4fd38c50480ce5e7d2f1f
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56b080333538988ecfa3850839bf7f712c701669718a82fbd698a32f34caa98c
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/model.safetensors b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c0ce6ce9d52cabce071475f9d48f48f5b8d91b0c
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a92ab5447511ebd020018d7aaa448fdba7afe3b88e07973f7f39c1543c1f599
+size 75550008
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8de83be649f2f0cc84109b3da72b06a86da913fc
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53a972b5048d9cd93f2f67cbe0716fdc3f9f7eef130ca0bcd46ceccf0b967cde
+size 99648
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c8c3a28aac8f1f9ca77962efe43ea06fb15e0e8f
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:228cac094619bd6083f03a7ee2edf19afe4a02b03540b32d3ee8c438e0a9134e
+size 50471
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..f89b1700a5b395a3b47d674af698774d8c6366aa
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.0,
+  "l0_per_token_0.001": 64.0,
+  "l0_per_token_0.01": 64.0,
+  "l0_per_token_0.05": 63.9999998046875,
+  "l0_per_token_0.1": 63.99998994140625,
+  "l0_per_token_0.5": 63.97103334960938,
+  "l0_per_token_1.0": 63.9458068359375,
+  "l1_per_token": 402.35637088713605,
+  "r2_score": 0.9365965701152947,
+  "l2_loss": 1.4869647443087117
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..742b92c17dae3727a86a3ac4effe659050c2bda6
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9416690e6e506ab4d3fcba8092752fcb42f78694c515e05df27d2663cdb5f07d
+size 50382
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..33a73c4bf6b1c39ff01b5ba42b7406d7f7014e98
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2523cc98f3cfc279dab31e7b29917fa320671b4e5c357dfecb376afc9940ac18
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..767e4672fdf9fff634436ce5464cfc4fc3fa831e
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b594b44919c8bbce35dc00568e326ed3991868f40fb24202dc1c2594ae627e90
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/model.safetensors b/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..073afe4d7d3ca8849ac0c36316e55c5cf5ae9988
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:faecbbdd38695d89450b85a054dc66d503240b49d09778cbbcfd473dd65a1c53
+size 75550008
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1800d99f7d9cede3c491c424d46e9e89fb96646d
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:11e65d33c83de4d9e0b579d507a98114b179946290a9e26967d23c55039dc966
+size 99648
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3f167bf3268aa8af0fc1ff07e27d2331d912eda4
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93b8506cd4aba92d485e039bdbc9719859ba6d0fc847ebed51328f58579d9129
+size 50471
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..503dcc37659d0eb6a9c2bcc4b403c08aaf307795
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:682e9e971305c27379380b62f6cbd239ee790d0e95e388af0213ef9f75b37e7c
+size 50382
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..da8b1013121edb27f29d70d6598a90bdbfe0f90a
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb79155f4b413b8dd785acdddf28c07c0c903e625fb27222cce749ba03f4f6c6
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e06b57dfb704fdfee007c9f700fccd757f33c876
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3135966f861ad52cd0ede3fb3ccb8714fa6aaf4e1667cd39c7fd9ce7afc7d84
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/model.safetensors b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b5fa4067a78541fd0cd96ad0ddbf7875697c505a
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dea572a2d63d43683e2872361ea0a2ccd1f6ae71d5be69d4ad38297d688b0803
+size 75550008
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8de83be649f2f0cc84109b3da72b06a86da913fc
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53a972b5048d9cd93f2f67cbe0716fdc3f9f7eef130ca0bcd46ceccf0b967cde
+size 99648
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e72a95e45e9ac577c1282489c6db130f3a1b4256
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6618d3feee18744aa4bbc56a29a8f2a709795a9b10a11a4a6d734ed684829948
+size 50471
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..039c56a2868773f9f7cbafe9fb2990de8c7c8144
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.0,
+  "l0_per_token_0.001": 64.0,
+  "l0_per_token_0.01": 64.0,
+  "l0_per_token_0.05": 63.99999741210937,
+  "l0_per_token_0.1": 63.99989135742187,
+  "l0_per_token_0.5": 63.96697768554687,
+  "l0_per_token_1.0": 63.9466037109375,
+  "l1_per_token": 404.4139333175605,
+  "r2_score": 0.9365662586589976,
+  "l2_loss": 1.487675621731976
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..09709ae8a282a5225253ea5d3e6316e4feadad85
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b81f64f0ccccc649449e31f20791d8979104c3a1f3ab9843634af5d850c2ebd
+size 50382
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..85ec6cde0beee6e103ebadd88c544627fa0d15e5
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:588bd97c71efdd7ed04ed99941a8062babe0d791ea2899fe35661a7d4be0b8f1
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2dfaf318f69f286c2b65f55156edf498d86c040
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 12288,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c6cb3a59ce9b684e653a4d96280c88642a3f5144
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:afda15059e7e9190727d50ea9e497fcdd8ec657b7166df90bba20cf9ec0350b8
+size 50387
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/code_metrics_step_final.json b/lm/gpt2-small_layer8/topk/x4/l0_128/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..6ca0f784612ee49fb3792b4aa2a4dd13b80daf83
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 127.99905666666666,
+   "recon_r2": 0.9399651943639552,
+   "l1_per_token": 638.8245164675636
+  },
+  "seed1": {
+   "l0": 127.99891333333333,
+   "recon_r2": 0.9400142279127632,
+   "l1_per_token": 640.9344261670653
+  },
+  "seed2": {
+   "l0": 127.99985333333333,
+   "recon_r2": 0.9399728100214055,
+   "l1_per_token": 640.5251561315667
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7247841801581286,
+   "mcc_abs": 0.7248827404639163,
+   "act_iou": 0.4030665323653994,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9612611269656903,
+   "matched_mcc_signed": {
+    "0.6": 0.8662596625669441,
+    "0.7": 0.8924555821029084,
+    "0.8": 0.9191605705367597,
+    "0.9": 0.9508587452940783
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8652897586218903,
+    "0.7": 0.8920822461010253,
+    "0.8": 0.9177644427646146,
+    "0.9": 0.9504347414162428
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7181457049393587,
+   "mcc_abs": 0.7182372418552787,
+   "act_iou": 0.39674744833753284,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9607786547617042,
+   "matched_mcc_signed": {
+    "0.6": 0.8712941182661332,
+    "0.7": 0.8942896665616288,
+    "0.8": 0.9186995694363597,
+    "0.9": 0.9497923203400457
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.868298143276389,
+    "0.7": 0.8928945244188097,
+    "0.8": 0.9174723432504773,
+    "0.9": 0.9489730707292353
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7211751897108386,
+   "mcc_abs": 0.721236163894222,
+   "act_iou": 0.4001246847675204,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9610729922392133,
+   "matched_mcc_signed": {
+    "0.6": 0.8690042481840657,
+    "0.7": 0.8935653068329374,
+    "0.8": 0.9166314519556468,
+    "0.9": 0.9505538109946816
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8664137892791247,
+    "0.7": 0.8915228092622354,
+    "0.8": 0.9162156208954684,
+    "0.9": 0.950304881147547
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 127.99927444444445,
+  "l0_std": 0.000506429037844174,
+  "recon_r2_mean": 0.9399840774327081,
+  "recon_r2_std": 2.6387272410603572e-05,
+  "l1_per_token_mean": 640.0946995887319,
+  "l1_per_token_std": 1.1188830865934323,
+  "input_r2_mean": 0.9610375913222026,
+  "input_r2_std": 0.0002431764289537593,
+  "mcc_signed_mean": 0.721368358269442,
+  "mcc_signed_std": 0.0033234505978847156,
+  "mcc_abs_mean": 0.7214520487378057,
+  "mcc_abs_std": 0.0033280050538750656,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8688526763390477,
+   "0.7": 0.8934368518324916,
+   "0.8": 0.9181638639762554,
+   "0.9": 0.9504016255429352
+  },
+  "matched_mcc_signed_band": 0.9077137544226824,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.866667230392468,
+   "0.7": 0.8921665265940234,
+   "0.8": 0.91715080230352,
+   "0.9": 0.949904231097675
+  },
+  "matched_mcc_abs_band": 0.9064721975969215,
+  "act_iou_mean": 0.39997955515681755,
+  "act_iou_std": 0.0031620409059240013,
+  "act_iou_n_pairs": 3072.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/dict_metrics_step_final.json b/lm/gpt2-small_layer8/topk/x4/l0_128/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..bd8498c9c370065b2d7dbeca1a174217acdabac1
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 127.99899467773437,
+   "recon_r2": 0.9398913293518665,
+   "l1_per_token": 638.6426693865146,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 300,
+   "dense_rate": 0.09765625,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 6,
+    "0.6": 4,
+    "0.7": 2,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 3,
+    "0.6": 2,
+    "0.7": 1,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 2,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.20269207656383514,
+   "n_dup_abs": {
+    "0.5": 481,
+    "0.6": 467,
+    "0.7": 451,
+    "0.8": 436,
+    "0.9": 380
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 240,
+    "0.6": 233,
+    "0.7": 225,
+    "0.8": 218,
+    "0.9": 190
+   },
+   "largest_dup_abs": {
+    "0.5": 3,
+    "0.6": 3,
+    "0.7": 3,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "babel_mu1_abs": 0.3444276750087738
+  },
+  "seed1": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 127.998926953125,
+   "recon_r2": 0.9399367378759114,
+   "l1_per_token": 640.7977344105819,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 296,
+   "dense_rate": 0.09635416666666667,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 6,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 3,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 2,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.20116938650608063,
+   "n_dup_abs": {
+    "0.5": 486,
+    "0.6": 462,
+    "0.7": 448,
+    "0.8": 436,
+    "0.9": 396
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 243,
+    "0.6": 231,
+    "0.7": 224,
+    "0.8": 218,
+    "0.9": 198
+   },
+   "largest_dup_abs": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 2,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "babel_mu1_abs": 0.34215202927589417
+  },
+  "seed2": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 127.99983637695313,
+   "recon_r2": 0.9398953350736131,
+   "l1_per_token": 640.4189135679461,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 297,
+   "dense_rate": 0.0966796875,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 4,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 2,
+    "0.6": 0,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 2,
+    "0.6": 1,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.20095916092395782,
+   "n_dup_abs": {
+    "0.5": 488,
+    "0.6": 464,
+    "0.7": 456,
+    "0.8": 444,
+    "0.9": 392
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 244,
+    "0.6": 232,
+    "0.7": 228,
+    "0.8": 222,
+    "0.9": 196
+   },
+   "largest_dup_abs": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 2,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "babel_mu1_abs": 0.34535422921180725
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.7153791785240173,
+   "dcs_abs": 0.7166199684143066,
+   "frac_sign_flipped": 0.030598958333333332,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.7151912450790405,
+    "0.6": 0.7156291604042053,
+    "0.7": 0.7156118750572205,
+    "0.8": 0.7153791785240173,
+    "0.9": 0.7153791785240173
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9990234375,
+    "0.6": 0.9996744791666666,
+    "0.7": 0.9998372395833334,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.714492816128768,
+    "0.6": 0.7153962082035529,
+    "0.7": 0.7154954017702646,
+    "0.8": 0.7153791785240173,
+    "0.9": 0.7153791785240173
+   },
+   "matched_frac_signed": {
+    "0.5": 0.7858072916666666,
+    "0.6": 0.7259114583333334,
+    "0.7": 0.6396484375,
+    "0.8": 0.5257161458333334,
+    "0.9": 0.2906901041666667
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7343917489051819,
+    "0.6": 0.7342474460601807,
+    "0.7": 0.7339069843292236,
+    "0.8": 0.7332110404968262,
+    "0.9": 0.7314574718475342
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9212239583333334,
+    "0.6": 0.92431640625,
+    "0.7": 0.9267578125,
+    "0.8": 0.9290364583333334,
+    "0.9": 0.9368489583333334
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.6765392738937711,
+    "0.6": 0.6786769606405869,
+    "0.7": 0.6801540313754231,
+    "0.8": 0.6811797882740697,
+    "0.9": 0.685265170565496
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7542317708333334,
+    "0.6": 0.7005208333333334,
+    "0.7": 0.6207682291666666,
+    "0.8": 0.5123697916666666,
+    "0.9": 0.2828776041666667
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.7059656977653503,
+   "dcs_abs": 0.707994282245636,
+   "frac_sign_flipped": 0.0390625,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.7056087255477905,
+    "0.6": 0.7061253190040588,
+    "0.7": 0.706142783164978,
+    "0.8": 0.7059656977653503,
+    "0.9": 0.7059656977653503
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9991861979166666,
+    "0.6": 0.9996744791666666,
+    "0.7": 0.9998372395833334,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7050344996969216,
+    "0.6": 0.7058954605017789,
+    "0.7": 0.706027851071364,
+    "0.8": 0.7059656977653503,
+    "0.9": 0.7059656977653503
+   },
+   "matched_frac_signed": {
+    "0.5": 0.7711588541666666,
+    "0.6": 0.7018229166666666,
+    "0.7": 0.6253255208333334,
+    "0.8": 0.5172526041666666,
+    "0.9": 0.27734375
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7263902425765991,
+    "0.6": 0.7266482710838318,
+    "0.7": 0.7262475490570068,
+    "0.8": 0.7258001565933228,
+    "0.9": 0.7236779928207397
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9210611979166666,
+    "0.6": 0.9241536458333334,
+    "0.7": 0.9261067708333334,
+    "0.8": 0.9283854166666666,
+    "0.9": 0.9371744791666666
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.6690498669825804,
+    "0.6": 0.6715346489606115,
+    "0.7": 0.6725827724828075,
+    "0.8": 0.6738222807956238,
+    "0.9": 0.6782125460061554
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7431640625,
+    "0.6": 0.6842447916666666,
+    "0.7": 0.6110026041666666,
+    "0.8": 0.5035807291666666,
+    "0.9": 0.2682291666666667
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.7102031707763672,
+   "dcs_abs": 0.7118776440620422,
+   "frac_sign_flipped": 0.034830729166666664,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.7101739048957825,
+    "0.6": 0.7102031707763672,
+    "0.7": 0.7102031707763672,
+    "0.8": 0.7102031707763672,
+    "0.9": 0.7102031707763672
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9991861979166666,
+    "0.6": 1.0,
+    "0.7": 1.0,
+    "0.8": 1.0,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7095959638924493,
+    "0.6": 0.7102031707763672,
+    "0.7": 0.7102031707763672,
+    "0.8": 0.7102031707763672,
+    "0.9": 0.7102031707763672
+   },
+   "matched_frac_signed": {
+    "0.5": 0.7809244791666666,
+    "0.6": 0.71484375,
+    "0.7": 0.6360677083333334,
+    "0.8": 0.5289713541666666,
+    "0.9": 0.2809244791666667
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7304647564888,
+    "0.6": 0.7303079962730408,
+    "0.7": 0.7300501465797424,
+    "0.8": 0.7295516133308411,
+    "0.9": 0.7261304259300232
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9207356770833334,
+    "0.6": 0.9246419270833334,
+    "0.7": 0.9264322916666666,
+    "0.8": 0.9283854166666666,
+    "0.9": 0.9358723958333334
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.6725649621512275,
+    "0.6": 0.6752733930382723,
+    "0.7": 0.6763420303274567,
+    "0.8": 0.6773050785219917,
+    "0.9": 0.6795654214026097
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7503255208333334,
+    "0.6": 0.693359375,
+    "0.7": 0.6194661458333334,
+    "0.8": 0.515625,
+    "0.9": 0.2737630208333333
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 298,
+  "n_dense_std": 2.0816659994661326,
+  "l0_mean": 128,
+  "l0_std": 0.000506638578729595,
+  "recon_r2_mean": 0.9399078007671303,
+  "recon_r2_std": null,
+  "l1_per_token_mean": 639.9531057883477,
+  "l1_per_token_std": 1.1505689927361935,
+  "babel_mu1_signed_mean": 0.20160687466462454,
+  "babel_mu1_signed_std": 0.0009456722848759274,
+  "n_dup_signed_per_threshold": {
+   "0.5": 5.333333333333333,
+   "0.6": 1.3333333333333333,
+   "0.7": 0.6666666666666666,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_signed_band": 0.5,
+  "n_dup_signed_band_std": 0.8660254037844386,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 2.6666666666666665,
+   "0.6": 0.6666666666666666,
+   "0.7": 0.3333333333333333,
+   "0.8": 0.0,
+   "0.9": 0.0
+  },
+  "n_dup_clusters_signed_band": 0.25,
+  "n_dup_clusters_signed_band_std": 0.4330127018922193,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 2.0,
+   "0.6": 1.3333333333333333,
+   "0.7": 1.3333333333333333,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "largest_dup_signed_band": 1.1666666666666665,
+  "largest_dup_signed_band_std": 0.2886751345948129,
+  "dcs_signed_mean": 0.7105160156885783,
+  "dcs_signed_std": 0.004714531679050497,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.7097077599060463,
+   "0.6": 0.7104982798272329,
+   "0.7": 0.7105754745393319,
+   "0.8": 0.7105160156885783,
+   "0.9": 0.7105160156885783
+  },
+  "dedup_dcs_signed_band": 0.7105264464359303,
+  "dedup_dcs_signed_band_std": 0.004732695478611044,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.7103246251742045,
+   "0.6": 0.7106525500615438,
+   "0.7": 0.7106526096661886,
+   "0.8": 0.7105160156885783,
+   "0.9": 0.7105160156885783
+  },
+  "matched_cos_signed_band": 0.7105842977762222,
+  "matched_cos_signed_band_std": 0.004736501249516561,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.9991319444444443,
+   "0.6": 0.999782986111111,
+   "0.7": 0.9998914930555557,
+   "0.8": 1.0,
+   "0.9": 1.0
+  },
+  "unique_frac_signed_band": 0.9999186197916667,
+  "unique_frac_signed_band_std": 7.047732778193673e-05,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.779296875,
+   "0.6": 0.7141927083333334,
+   "0.7": 0.6336805555555557,
+   "0.8": 0.5239800347222222,
+   "0.9": 0.28298611111111116
+  },
+  "matched_frac_signed_band": 0.5387098524305556,
+  "matched_frac_signed_band_std": 0.0076377540284836955,
+  "babel_mu1_abs_mean": 0.3439779778321584,
+  "babel_mu1_abs_std": 0.0016477838967480952,
+  "n_dup_abs_per_threshold": {
+   "0.5": 485.0,
+   "0.6": 464.3333333333333,
+   "0.7": 451.6666666666667,
+   "0.8": 438.6666666666667,
+   "0.9": 389.3333333333333
+  },
+  "n_dup_abs_band": 436.0,
+  "n_dup_abs_band_std": 2.7838821814150108,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 242.33333333333334,
+   "0.6": 232.0,
+   "0.7": 225.66666666666666,
+   "0.8": 219.33333333333334,
+   "0.9": 194.66666666666666
+  },
+  "n_dup_clusters_abs_band": 217.91666666666666,
+  "n_dup_clusters_abs_band_std": 1.5069284433354269,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 2.3333333333333335,
+   "0.6": 2.3333333333333335,
+   "0.7": 2.3333333333333335,
+   "0.8": 2.0,
+   "0.9": 2.0
+  },
+  "largest_dup_abs_band": 2.166666666666667,
+  "largest_dup_abs_band_std": 0.2886751345948129,
+  "dcs_abs_mean": 0.7121639649073283,
+  "dcs_abs_std": 0.004319965299621404,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.6727180343425263,
+   "0.6": 0.6751616675464902,
+   "0.7": 0.6763596113952292,
+   "0.8": 0.677435715863895,
+   "0.9": 0.6810143793247537
+  },
+  "dedup_dcs_abs_band": 0.677492843532592,
+  "dedup_dcs_abs_band_std": 0.003654641198044022,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.7304155826568604,
+   "0.6": 0.7304012378056844,
+   "0.7": 0.7300682266553243,
+   "0.8": 0.7295209368069967,
+   "0.9": 0.7270886301994324
+  },
+  "matched_cos_abs_band": 0.7292697578668594,
+  "matched_cos_abs_band_std": 0.0038127614492049137,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.9210069444444445,
+   "0.6": 0.9243706597222223,
+   "0.7": 0.9264322916666666,
+   "0.8": 0.9286024305555555,
+   "0.9": 0.9366319444444445
+  },
+  "unique_frac_abs_band": 0.9290093315972222,
+  "unique_frac_abs_band_std": 0.0002088053971129512,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.749240451388889,
+   "0.6": 0.6927083333333334,
+   "0.7": 0.6170789930555555,
+   "0.8": 0.510525173611111,
+   "0.9": 0.27495659722222227
+  },
+  "matched_frac_abs_band": 0.5238172743055556,
+  "matched_frac_abs_band_std": 0.006365021430582667,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/config.json b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..009a151a3e5d68ed8a2b355d1c0bf3372507f1cb
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/model.safetensors b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..27184c07ef80a14e304a787f83acb526eec10c78
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e69328e6d95f80719e213fa16c4123081ddcbb4f5e1a7d83ae3325cd847daf38
+size 18890032
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..31945af51b00341106d7415b708bad53562f6a5b
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3dcb4fc423a9d0d8ec34c5e43c85f7db6ccc94a393da3ef135705976e3ad8e31
+size 25920
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..457f8e36b26f334c12aa0b00f8e95209d3aa8e00
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c7e783e3d355cb0aea9fe00d6a4c9e2604e0becfa9ef1f420b8101b2441cab8
+size 13607
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..4659f73de26a01c859411ac57c509be8918423b4
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 127.99899467773437,
+  "l0_per_token_0.001": 127.99898940429688,
+  "l0_per_token_0.01": 127.99885361328126,
+  "l0_per_token_0.05": 127.99806484375,
+  "l0_per_token_0.1": 127.996898046875,
+  "l0_per_token_0.5": 127.95907001953125,
+  "l0_per_token_1.0": 127.88101645507813,
+  "l1_per_token": 638.6426693865146,
+  "r2_score": 0.9398913293518665,
+  "l2_loss": 1.4096946213094281
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4cbe9cb4e202a752a082b7349eaff6e0f08bf22d
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ee129221175b79e6c5813faabc5ef2266d229504414de4eb339fffc46a22331
+size 13518
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8842d43d986967c06f01e21cad4c2cf96a9d3872
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d61097e4972f4757cc0c50f6775aa1e8390a5856f30e9012783faa61d59d55a2
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..87629f8c71a4edb1108a0c367505998cab654028
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4582116bd8a33ae036f151563e20174dd315d54a6e9672782c8e64c4acf1058
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/config.json b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..2b7d71bdb08174d183d81a685eecf8b4ae367b6c
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/model.safetensors b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..fd0a439f4d674a04b948cc6be1f7abf6931c4e57
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c41a257578fb5881e049f42038b3294b5379e197dfb8bd6d27782065bc38a447
+size 18890032
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..84e6e953a95c89bb5db5c5786397a3a9ac3f78b1
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:243bcc63ea80a8370fb2492faeb64b903693e9cd7605074acd456ebb14f791c0
+size 25920
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d2b9dde249bd30d4a10d5193d4eaa149fa6f2caf
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b021dec302c274f5f912fa2784112d4cdfe6e9456e718f13e941af32585bb928
+size 13607
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..059f7e10536d00c0a900949d25d9f3263940c232
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 127.998926953125,
+  "l0_per_token_0.001": 127.99891826171876,
+  "l0_per_token_0.01": 127.99869458007812,
+  "l0_per_token_0.05": 127.99803676757813,
+  "l0_per_token_0.1": 127.99686875,
+  "l0_per_token_0.5": 127.95877490234375,
+  "l0_per_token_1.0": 127.87174907226563,
+  "l1_per_token": 640.7977344105819,
+  "r2_score": 0.9399367378759114,
+  "l2_loss": 1.4086296809037018
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9f9e12a77753e484de73182932c557869f8eec9d
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3851a7015b68ee2c89a12cbb5c080518e31cee44c17b8b0f6eb5b85ed10e3f6
+size 13518
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e7eb98873348a4ded2cf2604e0003bf05a86a31a
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c2bd63bb55e6049128636a423489f0db995273d8de3cd7ffc281e5e8257df52
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..839388dd47718a36cfe0be387f05e4b1732bc62e
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:367c250ef6e766679c63bd7d9532f7718475bba363a7f40f07057b9226686a4f
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/config.json b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7be6b2725311038b3a08e0d310124cae24d8de0b
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/model.safetensors b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..603222657a578bf51be09a119697badfc35d60bc
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d743d9c48d0946a1b0eabed9066de15c51a931894440dee53346c370a75e85de
+size 18890032
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..62a2ffbf5f844452bfd42fe2aae004505cc8c6a2
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31fd90df284bed05643f5f593c3ef0a69f5d440f940af7a306e9796e04e8b48e
+size 25920
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7688c3535fb4e1dfffeff1866d9e21eb39404810
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a573985d4647e4578a3f344d56f4798f55f644fcb6ee0134b9e04020ea3edde
+size 13607
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..7e877ca6412c5f349e27e6efe9e40e624d1a04da
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 127.99983637695313,
+  "l0_per_token_0.001": 127.99979868164063,
+  "l0_per_token_0.01": 127.99977861328125,
+  "l0_per_token_0.05": 127.99955434570313,
+  "l0_per_token_0.1": 127.99876064453125,
+  "l0_per_token_0.5": 127.9609740234375,
+  "l0_per_token_1.0": 127.877821484375,
+  "l1_per_token": 640.4189135679461,
+  "r2_score": 0.9398953350736131,
+  "l2_loss": 1.4096006773852028
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..02fcfdaa9f9df556eee791d117098327643e303d
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0b729ff5225b8fb4ba2b8db5d6d174edb9dcecc9a9924978d848008574d771f
+size 13518
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..246187546028f71c0d894c9ee94f97c7c59ac4b7
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:508508c0814a9e413fd2244d28e6dba308424ae46e3ffbebfe24709d231d2d6e
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..718e0cefeeff3db9b4cdf7c67c7dbc7133893413
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0369f8c414ff23093d75a1d50ed7f51837b8410ff37b41d425d8a74ae5674a20
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/code_metrics_step_final.json b/lm/gpt2-small_layer8/topk/x4/l0_32/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..b1899874c54b423f103dffb90c067c26221e4fe1
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 32.0,
+   "recon_r2": 0.8943229886409616,
+   "l1_per_token": 315.78752639425693
+  },
+  "seed1": {
+   "l0": 32.0,
+   "recon_r2": 0.8942915269803985,
+   "l1_per_token": 316.8972846276822
+  },
+  "seed2": {
+   "l0": 32.0,
+   "recon_r2": 0.8944520321088446,
+   "l1_per_token": 318.8952004175599
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7917707895507274,
+   "mcc_abs": 0.7919744246306947,
+   "act_iou": 0.5065899121814615,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9636530951755685,
+   "matched_mcc_signed": {
+    "0.6": 0.8729450613136136,
+    "0.7": 0.8900733578025835,
+    "0.8": 0.910303829977194,
+    "0.9": 0.9314589956982844
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8709188668131908,
+    "0.7": 0.8888369234617941,
+    "0.8": 0.9092717175483737,
+    "0.9": 0.9314242901550511
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7862820113068644,
+   "mcc_abs": 0.7863483803248501,
+   "act_iou": 0.5023446761391819,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9631246854583128,
+   "matched_mcc_signed": {
+    "0.6": 0.8720276980019969,
+    "0.7": 0.8913224754103,
+    "0.8": 0.9103394166008768,
+    "0.9": 0.9341129033989036
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8711121274773037,
+    "0.7": 0.8895934860119046,
+    "0.8": 0.9098531756633003,
+    "0.9": 0.9341285068865729
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.787947914346902,
+   "mcc_abs": 0.7879786120422271,
+   "act_iou": 0.5035492478118445,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9633295604238475,
+   "matched_mcc_signed": {
+    "0.6": 0.8714741175225699,
+    "0.7": 0.8891200366901018,
+    "0.8": 0.9088717722182696,
+    "0.9": 0.9323123232783198
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.870780026546757,
+    "0.7": 0.8890905181022508,
+    "0.8": 0.9086026466184641,
+    "0.9": 0.932328470816204
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 32.0,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.8943555159100683,
+  "recon_r2_std": 8.505287467821259e-05,
+  "l1_per_token_mean": 317.1933371464997,
+  "l1_per_token_std": 1.5748475415740022,
+  "input_r2_mean": 0.9633691136859096,
+  "input_r2_std": 0.0002664161270032116,
+  "mcc_signed_mean": 0.7886669050681646,
+  "mcc_signed_std": 0.0028141397256572745,
+  "mcc_abs_mean": 0.7887671389992573,
+  "mcc_abs_std": 0.0028947235972202963,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8721489589460601,
+   "0.7": 0.8901719566343284,
+   "0.8": 0.9098383395987802,
+   "0.9": 0.9326280741251692
+  },
+  "matched_mcc_signed_band": 0.9011968323260845,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.8709370069457506,
+   "0.7": 0.8891736425253165,
+   "0.8": 0.9092425132767127,
+   "0.9": 0.9326270892859426
+  },
+  "matched_mcc_abs_band": 0.9004950630084306,
+  "act_iou_mean": 0.5041612787108293,
+  "act_iou_std": 0.002187794238417243,
+  "act_iou_n_pairs": 3072.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/dict_metrics_step_final.json b/lm/gpt2-small_layer8/topk/x4/l0_32/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..0a05168191db5aec47831327ed2724cd52d744e3
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/dict_metrics_step_final.json
@@ -0,0 +1,595 @@
+{
+ "schema": 3,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 32.0,
+   "recon_r2": 0.8941865117189599,
+   "l1_per_token": 315.68421864943195,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 18,
+   "dense_rate": 0.005859375,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 187,
+    "0.6": 60,
+    "0.7": 26,
+    "0.8": 20,
+    "0.9": 15
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 83,
+    "0.6": 22,
+    "0.7": 6,
+    "0.8": 3,
+    "0.9": 1
+   },
+   "n_dup_removed_signed": {
+    "0.5": 104,
+    "0.6": 38,
+    "0.7": 20,
+    "0.8": 17,
+    "0.9": 14
+   },
+   "largest_dup_signed": {
+    "0.5": 16,
+    "0.6": 16,
+    "0.7": 16,
+    "0.8": 16,
+    "0.9": 15
+   },
+   "babel_mu1_signed": 0.32970544695854187,
+   "n_dup_abs": {
+    "0.5": 231,
+    "0.6": 101,
+    "0.7": 62,
+    "0.8": 48,
+    "0.9": 27
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 104,
+    "0.6": 42,
+    "0.7": 23,
+    "0.8": 17,
+    "0.9": 7
+   },
+   "n_dup_removed_abs": {
+    "0.5": 127,
+    "0.6": 59,
+    "0.7": 39,
+    "0.8": 31,
+    "0.9": 20
+   },
+   "largest_dup_abs": {
+    "0.5": 16,
+    "0.6": 16,
+    "0.7": 16,
+    "0.8": 16,
+    "0.9": 15
+   },
+   "babel_mu1_abs": 0.34352269768714905,
+   "dup_excludes_dead": true
+  },
+  "seed1": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 32.0,
+   "recon_r2": 0.8941567258902522,
+   "l1_per_token": 316.90389909196955,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 20,
+   "dense_rate": 0.006510416666666667,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 182,
+    "0.6": 60,
+    "0.7": 22,
+    "0.8": 14,
+    "0.9": 13
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 85,
+    "0.6": 25,
+    "0.7": 6,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "n_dup_removed_signed": {
+    "0.5": 97,
+    "0.6": 35,
+    "0.7": 16,
+    "0.8": 12,
+    "0.9": 11
+   },
+   "largest_dup_signed": {
+    "0.5": 12,
+    "0.6": 12,
+    "0.7": 12,
+    "0.8": 12,
+    "0.9": 11
+   },
+   "babel_mu1_signed": 0.3280099332332611,
+   "n_dup_abs": {
+    "0.5": 239,
+    "0.6": 99,
+    "0.7": 55,
+    "0.8": 30,
+    "0.9": 23
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 113,
+    "0.6": 44,
+    "0.7": 22,
+    "0.8": 10,
+    "0.9": 7
+   },
+   "n_dup_removed_abs": {
+    "0.5": 126,
+    "0.6": 55,
+    "0.7": 33,
+    "0.8": 20,
+    "0.9": 16
+   },
+   "largest_dup_abs": {
+    "0.5": 12,
+    "0.6": 12,
+    "0.7": 12,
+    "0.8": 12,
+    "0.9": 11
+   },
+   "babel_mu1_abs": 0.3428892195224762,
+   "dup_excludes_dead": true
+  },
+  "seed2": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 32.0,
+   "recon_r2": 0.8943315094264184,
+   "l1_per_token": 318.6969839403548,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 18,
+   "dense_rate": 0.005859375,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 163,
+    "0.6": 49,
+    "0.7": 21,
+    "0.8": 14,
+    "0.9": 10
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 75,
+    "0.6": 19,
+    "0.7": 6,
+    "0.8": 3,
+    "0.9": 1
+   },
+   "n_dup_removed_signed": {
+    "0.5": 88,
+    "0.6": 30,
+    "0.7": 15,
+    "0.8": 11,
+    "0.9": 9
+   },
+   "largest_dup_signed": {
+    "0.5": 10,
+    "0.6": 10,
+    "0.7": 10,
+    "0.8": 10,
+    "0.9": 10
+   },
+   "babel_mu1_signed": 0.3250880837440491,
+   "n_dup_abs": {
+    "0.5": 204,
+    "0.6": 86,
+    "0.7": 56,
+    "0.8": 38,
+    "0.9": 20
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 94,
+    "0.6": 37,
+    "0.7": 23,
+    "0.8": 15,
+    "0.9": 6
+   },
+   "n_dup_removed_abs": {
+    "0.5": 110,
+    "0.6": 49,
+    "0.7": 33,
+    "0.8": 23,
+    "0.9": 14
+   },
+   "largest_dup_abs": {
+    "0.5": 10,
+    "0.6": 10,
+    "0.7": 10,
+    "0.8": 10,
+    "0.9": 10
+   },
+   "babel_mu1_abs": 0.338547945022583,
+   "dup_excludes_dead": true
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.8359165191650391,
+   "dcs_abs": 0.8375988006591797,
+   "frac_sign_flipped": 0.016927083333333332,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.8336952924728394,
+    "0.6": 0.836276113986969,
+    "0.7": 0.8366695046424866,
+    "0.8": 0.8366279006004333,
+    "0.9": 0.8363686800003052
+   },
+   "unique_frac_signed": {
+    "0.5": 0.96728515625,
+    "0.6": 0.9881184895833334,
+    "0.7": 0.994140625,
+    "0.8": 0.9952799479166666,
+    "0.9": 0.9959309895833334
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.8064210812444799,
+    "0.6": 0.8263398906274233,
+    "0.7": 0.831767144263722,
+    "0.8": 0.8326789733352294,
+    "0.9": 0.8329654871292103
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8681640625,
+    "0.6": 0.8580729166666666,
+    "0.7": 0.8160807291666666,
+    "0.8": 0.7405598958333334,
+    "0.9": 0.6051432291666666
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8346836566925049,
+    "0.6": 0.8378351926803589,
+    "0.7": 0.8385981917381287,
+    "0.8": 0.8390131592750549,
+    "0.9": 0.8380686640739441
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9588216145833334,
+    "0.6": 0.9814453125,
+    "0.7": 0.98828125,
+    "0.8": 0.99169921875,
+    "0.9": 0.994140625
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8003127313762283,
+    "0.6": 0.8222894225036725,
+    "0.7": 0.8287708691786975,
+    "0.8": 0.8320486945740413,
+    "0.9": 0.8331581054953858
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8623046875,
+    "0.6": 0.8548177083333334,
+    "0.7": 0.8138020833333334,
+    "0.8": 0.7389322916666666,
+    "0.9": 0.6038411458333334
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.8309705853462219,
+   "dcs_abs": 0.832480251789093,
+   "frac_sign_flipped": 0.016927083333333332,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.8316115736961365,
+    "0.6": 0.8327215909957886,
+    "0.7": 0.8319627046585083,
+    "0.8": 0.832119345664978,
+    "0.9": 0.831850528717041
+   },
+   "unique_frac_signed": {
+    "0.5": 0.96875,
+    "0.6": 0.9889322916666666,
+    "0.7": 0.9943033854166666,
+    "0.8": 0.9954427083333334,
+    "0.9": 0.9962565104166666
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.8056237120181322,
+    "0.6": 0.8235052713037778,
+    "0.7": 0.8272233337823611,
+    "0.8": 0.828327135105307,
+    "0.9": 0.8287365049278984
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8665364583333334,
+    "0.6": 0.8518880208333334,
+    "0.7": 0.806640625,
+    "0.8": 0.7353515625,
+    "0.9": 0.595703125
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8332535028457642,
+    "0.6": 0.8347722887992859,
+    "0.7": 0.834211528301239,
+    "0.8": 0.8344824314117432,
+    "0.9": 0.8334306478500366
+   },
+   "unique_frac_abs": {
+    "0.5": 0.96142578125,
+    "0.6": 0.982421875,
+    "0.7": 0.98828125,
+    "0.8": 0.9912109375,
+    "0.9": 0.9944661458333334
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.8011113999527879,
+    "0.6": 0.8200985571602359,
+    "0.7": 0.8244356119539589,
+    "0.8": 0.8271481131669134,
+    "0.9": 0.828818564186804
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8616536458333334,
+    "0.6": 0.8473307291666666,
+    "0.7": 0.8053385416666666,
+    "0.8": 0.7350260416666666,
+    "0.9": 0.5947265625
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.832785427570343,
+   "dcs_abs": 0.834144115447998,
+   "frac_sign_flipped": 0.018229166666666668,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.8293605446815491,
+    "0.6": 0.8329628109931946,
+    "0.7": 0.8328039646148682,
+    "0.8": 0.8326433300971985,
+    "0.9": 0.8327431082725525
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9698893229166666,
+    "0.6": 0.9894205729166666,
+    "0.7": 0.9949544270833334,
+    "0.8": 0.9962565104166666,
+    "0.9": 0.9967447916666666
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.8043879371349855,
+    "0.6": 0.8241505416711636,
+    "0.7": 0.8286019914861148,
+    "0.8": 0.8295263384643476,
+    "0.9": 0.8300323559669778
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8626302083333334,
+    "0.6": 0.8531901041666666,
+    "0.7": 0.8108723958333334,
+    "0.8": 0.7373046875,
+    "0.9": 0.6015625
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8313796520233154,
+    "0.6": 0.8345975875854492,
+    "0.7": 0.8343507051467896,
+    "0.8": 0.8343520760536194,
+    "0.9": 0.8339505195617676
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9615885416666666,
+    "0.6": 0.9830729166666666,
+    "0.7": 0.9892578125,
+    "0.8": 0.9930013020833334,
+    "0.9": 0.9951171875
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7994451471604407,
+    "0.6": 0.8204702846705914,
+    "0.7": 0.8253879534313455,
+    "0.8": 0.8285126979171764,
+    "0.9": 0.8298784955404699
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8564453125,
+    "0.6": 0.8489583333333334,
+    "0.7": 0.8082682291666666,
+    "0.8": 0.7373046875,
+    "0.9": 0.6005859375
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 19,
+  "n_dense_std": 1.1547005383792515,
+  "l0_mean": 32,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.8942249156785435,
+  "recon_r2_std": 9.350651962543287e-05,
+  "l1_per_token_mean": 317.09503389391875,
+  "l1_per_token_std": 1.5154497876651998,
+  "babel_mu1_signed_mean": 0.327601154645284,
+  "babel_mu1_signed_std": 0.002335666010826296,
+  "n_dup_signed_per_threshold": {
+   "0.5": 177.33333333333334,
+   "0.6": 56.333333333333336,
+   "0.7": 23.0,
+   "0.8": 16.0,
+   "0.9": 12.666666666666666
+  },
+  "n_dup_signed_band": 27.000000000000004,
+  "n_dup_signed_band_std": 3.3819373146171707,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 81.0,
+   "0.6": 22.0,
+   "0.7": 6.0,
+   "0.8": 2.6666666666666665,
+   "0.9": 1.3333333333333333
+  },
+  "n_dup_clusters_signed_band": 8.0,
+  "n_dup_clusters_signed_band_std": 0.75,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 12.666666666666666,
+   "0.6": 12.666666666666666,
+   "0.7": 12.666666666666666,
+   "0.8": 12.666666666666666,
+   "0.9": 12.0
+  },
+  "largest_dup_signed_band": 12.5,
+  "largest_dup_signed_band_std": 2.947456530637899,
+  "dcs_signed_mean": 0.8332241773605347,
+  "dcs_signed_std": 0.002501987483746608,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.8054775767991992,
+   "0.6": 0.8246652345341215,
+   "0.7": 0.8291974898440659,
+   "0.8": 0.830177482301628,
+   "0.9": 0.8305781160080289
+  },
+  "dedup_dcs_signed_band": 0.8286545806719611,
+  "dedup_dcs_signed_band_std": 0.002056490227205099,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.8315558036168417,
+   "0.6": 0.8339868386586508,
+   "0.7": 0.8338120579719543,
+   "0.8": 0.8337968587875366,
+   "0.9": 0.8336541056632996
+  },
+  "matched_cos_signed_band": 0.8338124652703603,
+  "matched_cos_signed_band_std": 0.0023359403415814267,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.9686414930555555,
+   "0.6": 0.9888237847222222,
+   "0.7": 0.9944661458333334,
+   "0.8": 0.9956597222222222,
+   "0.9": 0.9963107638888888
+  },
+  "unique_frac_signed_band": 0.9938151041666666,
+  "unique_frac_signed_band_std": 0.0004933412944735068,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.8657769097222223,
+   "0.6": 0.8543836805555555,
+   "0.7": 0.8111979166666666,
+   "0.8": 0.7377387152777778,
+   "0.9": 0.6008029513888888
+  },
+  "matched_frac_signed_band": 0.7510308159722222,
+  "matched_frac_signed_band_std": 0.0037929929060347544,
+  "babel_mu1_abs_mean": 0.3416532874107361,
+  "babel_mu1_abs_std": 0.002707893490709062,
+  "n_dup_abs_per_threshold": {
+   "0.5": 224.66666666666666,
+   "0.6": 95.33333333333333,
+   "0.7": 57.666666666666664,
+   "0.8": 38.666666666666664,
+   "0.9": 23.333333333333332
+  },
+  "n_dup_abs_band": 53.75,
+  "n_dup_abs_band_std": 5.055937104039171,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 103.66666666666667,
+   "0.6": 41.0,
+   "0.7": 22.666666666666668,
+   "0.8": 14.0,
+   "0.9": 6.666666666666667
+  },
+  "n_dup_clusters_abs_band": 21.083333333333336,
+  "n_dup_clusters_abs_band_std": 1.0408329997330663,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 12.666666666666666,
+   "0.6": 12.666666666666666,
+   "0.7": 12.666666666666666,
+   "0.8": 12.666666666666666,
+   "0.9": 12.0
+  },
+  "largest_dup_abs_band": 12.5,
+  "largest_dup_abs_band_std": 2.947456530637899,
+  "dcs_abs_mean": 0.8347410559654236,
+  "dcs_abs_std": 0.0026109651701764582,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.8002897594964856,
+   "0.6": 0.8209527547781666,
+   "0.7": 0.8261981448546672,
+   "0.8": 0.8292365018860437,
+   "0.9": 0.8306183884075532
+  },
+  "dedup_dcs_abs_band": 0.8267514474816077,
+  "dedup_dcs_abs_band_std": 0.002059152680302845,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.8331056038538615,
+   "0.6": 0.835735023021698,
+   "0.7": 0.8357201417287191,
+   "0.8": 0.8359492222468058,
+   "0.9": 0.8351499438285828
+  },
+  "matched_cos_abs_band": 0.8356385827064514,
+  "matched_cos_abs_band_std": 0.002373511969100615,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.9606119791666666,
+   "0.6": 0.9823133680555555,
+   "0.7": 0.9886067708333334,
+   "0.8": 0.9919704861111112,
+   "0.9": 0.9945746527777778
+  },
+  "unique_frac_abs_band": 0.9893663194444444,
+  "unique_frac_abs_band_std": 0.0006540019233529217,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.8601345486111112,
+   "0.6": 0.8503689236111112,
+   "0.7": 0.8091362847222222,
+   "0.8": 0.737087673611111,
+   "0.9": 0.5997178819444445
+  },
+  "matched_frac_abs_band": 0.7490776909722223,
+  "matched_frac_abs_band_std": 0.0036306276059527376,
+  "alive_source": "stats",
+  "frac_denominator": "d_sae",
+  "d_sae": 3072
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/config.json b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..6540cca8248886d0bfee03d02e40e7763e433c55
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/model.safetensors b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4586470d10ed7b4dae0b7d0d09ea648c2b02a10c
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f549241567182b43a5fa434a57cedcec1a14e641251da8cd027a1352c9e72e1d
+size 18890032
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..22476d226c364cbdda1c2808c14b7ca570e296dc
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af8cf66876f6b155c9ad52c0309e49a57494f55abcd7294eb01e0be14019e09c
+size 25920
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..69bf8a7ea3cd4e90587cb25ce39112d3f20c4b4e
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99a7b2f96069911c6ed08f93868a63fab51944d58000403303b2cb0e04f71308
+size 13607
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..a4cae01053bbde5362a6480d49526fdc7b8b848c
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 32.0,
+  "l0_per_token_0.001": 32.0,
+  "l0_per_token_0.01": 32.0,
+  "l0_per_token_0.05": 32.0,
+  "l0_per_token_0.1": 31.999999755859374,
+  "l0_per_token_0.5": 31.99116162109375,
+  "l0_per_token_1.0": 31.982162158203124,
+  "l1_per_token": 315.68421864943195,
+  "r2_score": 0.8941865117189599,
+  "l2_loss": 2.4815838518621653
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..da19b4db9d2ab1321ff596afa92a9d6d07b2577f
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b934bc2d8e052f711a3d1832ebf58ca3fd920401eb92f695b15a321aafebb3e9
+size 13518
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7470d95bfcfaed1f6152a4b73ef09238f84707c7
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:639ed97faf07a77ca5ff8b22d078a8a467449e366b6962cd2d267c28def04d6c
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e790609c84e061676d565a3c2239ebec8bc76af9
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76fe673e070d7df55ccfd461cfa4f534b88165e3b21cd5356dffcbda5c822cf2
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/config.json b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7199eb4b86125283cef31c7588b4bb855bfd40e9
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/model.safetensors b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..632ff8d4172dcbed9e6845bc9ce87008f60bdd8c
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9326577e6c5e3c8d8bcbac2b114f9837df606a8dcc14d39a8c5ed7f5a02e095a
+size 18890032
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..22476d226c364cbdda1c2808c14b7ca570e296dc
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af8cf66876f6b155c9ad52c0309e49a57494f55abcd7294eb01e0be14019e09c
+size 25920
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8ded180e520a2bc4c4177de82fe50f2b55710acb
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b25d5e3fe5b780a6e09fd0e654f5e4f5b132f7fd46b59457f922368256c2284e
+size 13607
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..6dd254904f3a553b97190f0f4b24ecbfe5a54c3e
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 32.0,
+  "l0_per_token_0.001": 32.0,
+  "l0_per_token_0.01": 32.0,
+  "l0_per_token_0.05": 32.0,
+  "l0_per_token_0.1": 32.0,
+  "l0_per_token_0.5": 31.99252666015625,
+  "l0_per_token_1.0": 31.98349755859375,
+  "l1_per_token": 316.90389909196955,
+  "r2_score": 0.8941567258902522,
+  "l2_loss": 2.48228240204453
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..cd059a58ef3b2a8349f6afff0da890450ed81069
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e5060cb96f1a4b90911da64dd99e0a1a392a4fbecba64df929e27e47f7f7f4e
+size 13518
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f2d6e7b91fe6d1489d6075697c5824a41e86ea65
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5273cc7e9599bf4b40535f630e5d0db95bf507dfda2328b78aae9c877ab1cbd
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..032f1aeddaea715d98edcecf3f1ab4d8dd878d54
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:165da684452f371834d1825e4230db47f84a0af4aa9d7ac64641db90fc413661
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/config.json b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..4494383ba4aba86812dc6cc4811de072f5218538
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/model.safetensors b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..353f98a2f83a2e3ba3afca188cf4d34e2bfe35bb
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37eb03ff795ce70c928945c9d11680e79dc2aaebd515dc6ac239404d2ea9364d
+size 18890032
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..22476d226c364cbdda1c2808c14b7ca570e296dc
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af8cf66876f6b155c9ad52c0309e49a57494f55abcd7294eb01e0be14019e09c
+size 25920
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a97be6b71330d3483b286ac6a31b3292f9cc8570
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:78a1d45dbf0ad434b2b67370ea526336bc207bc2349bfd3a4afb2bc1fb246b2f
+size 13607
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..56e2e60e296d6dcacc33f4a55402c525b3821e0b
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 32.0,
+  "l0_per_token_0.001": 32.0,
+  "l0_per_token_0.01": 32.0,
+  "l0_per_token_0.05": 32.0,
+  "l0_per_token_0.1": 31.99999931640625,
+  "l0_per_token_0.5": 31.989003564453125,
+  "l0_per_token_1.0": 31.981450927734375,
+  "l1_per_token": 318.6969839403548,
+  "r2_score": 0.8943315094264184,
+  "l2_loss": 2.478183283663912
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..aa50c0a95d102d12941c1991bd80467b0ef90f1c
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d396dd3844d15b4a8c6598319cc0f262f56e24f0a1741b3e8f54163def444206
+size 13518
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..48a42812654b599ac5253d4194480d7944e35834
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf6f2a5d63749cbcb4ec7d73421a3e2962a2426cca89b29772d36eb3d16afa11
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a759e7987c6b76b78965b7841553f1d52cbd9546
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e9249762a7aad78d19f0d96bbf4234877283020d4876932d284116729d897cce
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/code_metrics_step_final.json b/lm/gpt2-small_layer8/topk/x4/l0_64/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..72ae36ebcac18d45489e782a6079f26f1ced39d8
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 64.0,
+   "recon_r2": 0.9170001183922643,
+   "l1_per_token": 437.37387798521064
+  },
+  "seed1": {
+   "l0": 63.999991666666666,
+   "recon_r2": 0.9170672498269145,
+   "l1_per_token": 437.78611818490407
+  },
+  "seed2": {
+   "l0": 64.0,
+   "recon_r2": 0.917073391774114,
+   "l1_per_token": 439.4674978476801
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.778839252509887,
+   "mcc_abs": 0.7788863152607887,
+   "act_iou": 0.4683020411595065,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9649442446221492,
+   "matched_mcc_signed": {
+    "0.6": 0.8743534818044798,
+    "0.7": 0.8949230195456378,
+    "0.8": 0.9158185376163305,
+    "0.9": 0.939230951994286
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8723980159711286,
+    "0.7": 0.8939292699223388,
+    "0.8": 0.9151534317517604,
+    "0.9": 0.9390581985254495
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7790937454077156,
+   "mcc_abs": 0.7791289225879613,
+   "act_iou": 0.46912024123981966,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9646878856661039,
+   "matched_mcc_signed": {
+    "0.6": 0.875133803154284,
+    "0.7": 0.8957420556274147,
+    "0.8": 0.9158323655016293,
+    "0.9": 0.9385161954883471
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.87355062524931,
+    "0.7": 0.8944974257678223,
+    "0.8": 0.9154052597378326,
+    "0.9": 0.9381442084636107
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7758606095981998,
+   "mcc_abs": 0.775941693989337,
+   "act_iou": 0.4667327268238927,
+   "act_iou_n_pairs": 3072,
+   "input_r2": 0.9646054386753542,
+   "matched_mcc_signed": {
+    "0.6": 0.8773679946452828,
+    "0.7": 0.895517019791829,
+    "0.8": 0.9153656859236953,
+    "0.9": 0.9400943274122432
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8760190328452466,
+    "0.7": 0.8944739494040904,
+    "0.8": 0.9146501681210821,
+    "0.9": 0.9401021423765152
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 63.99999722222222,
+  "l0_std": 4.811252243406598e-06,
+  "recon_r2_mean": 0.9170469199977642,
+  "recon_r2_std": 4.064755327054002e-05,
+  "l1_per_token_mean": 438.20916467259826,
+  "l1_per_token_std": 1.109070457723238,
+  "input_r2_mean": 0.9647458563212025,
+  "input_r2_std": 0.00017668563901838995,
+  "mcc_signed_mean": 0.7779312025052674,
+  "mcc_signed_std": 0.0017976951629407117,
+  "mcc_abs_mean": 0.777985643946029,
+  "mcc_abs_std": 0.0017742641144698162,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8756184265346821,
+   "0.7": 0.8953940316549605,
+   "0.8": 0.9156721963472183,
+   "0.9": 0.9392804916316254
+  },
+  "matched_mcc_signed_band": 0.9064912865421215,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.8739892246885619,
+   "0.7": 0.8943002150314171,
+   "0.8": 0.915069619870225,
+   "0.9": 0.9391015164551918
+  },
+  "matched_mcc_abs_band": 0.9056151440113489,
+  "act_iou_mean": 0.468051669741073,
+  "act_iou_std": 0.0012132891893283568,
+  "act_iou_n_pairs": 3072.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/dict_metrics_step_final.json b/lm/gpt2-small_layer8/topk/x4/l0_64/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..5089e540edd3e48734a5610f20f009d2509bd8f0
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 64.0,
+   "recon_r2": 0.9169102820043679,
+   "l1_per_token": 437.2697283288448,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 64,
+   "dense_rate": 0.020833333333333332,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 44,
+    "0.6": 6,
+    "0.7": 2,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 22,
+    "0.6": 3,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "largest_dup_signed": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 2,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "babel_mu1_signed": 0.25215375423431396,
+   "n_dup_abs": {
+    "0.5": 184,
+    "0.6": 137,
+    "0.7": 111,
+    "0.8": 97,
+    "0.9": 53
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 91,
+    "0.6": 68,
+    "0.7": 55,
+    "0.8": 48,
+    "0.9": 26
+   },
+   "largest_dup_abs": {
+    "0.5": 3,
+    "0.6": 3,
+    "0.7": 3,
+    "0.8": 3,
+    "0.9": 3
+   },
+   "babel_mu1_abs": 0.29581913352012634
+  },
+  "seed1": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 63.99999213867188,
+   "recon_r2": 0.9169844455894797,
+   "l1_per_token": 437.6918106018625,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 67,
+   "dense_rate": 0.021809895833333332,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 48,
+    "0.6": 12,
+    "0.7": 2,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 24,
+    "0.6": 6,
+    "0.7": 1,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 2,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.2503739297389984,
+   "n_dup_abs": {
+    "0.5": 189,
+    "0.6": 147,
+    "0.7": 118,
+    "0.8": 96,
+    "0.9": 56
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 94,
+    "0.6": 73,
+    "0.7": 59,
+    "0.8": 48,
+    "0.9": 28
+   },
+   "largest_dup_abs": {
+    "0.5": 3,
+    "0.6": 3,
+    "0.7": 2,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "babel_mu1_abs": 0.29480278491973877
+  },
+  "seed2": {
+   "d_sae": 3072,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 64.0,
+   "recon_r2": 0.9169728717403118,
+   "l1_per_token": 439.34843191899427,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 66,
+   "dense_rate": 0.021484375,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 38,
+    "0.6": 12,
+    "0.7": 2,
+    "0.8": 2,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 19,
+    "0.6": 6,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 2,
+    "0.8": 2,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.2500469386577606,
+   "n_dup_abs": {
+    "0.5": 213,
+    "0.6": 163,
+    "0.7": 139,
+    "0.8": 117,
+    "0.9": 64
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 106,
+    "0.6": 81,
+    "0.7": 69,
+    "0.8": 58,
+    "0.9": 32
+   },
+   "largest_dup_abs": {
+    "0.5": 3,
+    "0.6": 3,
+    "0.7": 3,
+    "0.8": 3,
+    "0.9": 2
+   },
+   "babel_mu1_abs": 0.29904496669769287
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.8001644611358643,
+   "dcs_abs": 0.8014826774597168,
+   "frac_sign_flipped": 0.015950520833333332,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.7991164922714233,
+    "0.6": 0.8005691170692444,
+    "0.7": 0.8004101514816284,
+    "0.8": 0.8004142642021179,
+    "0.9": 0.8004142642021179
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9925130208333334,
+    "0.6": 0.99853515625,
+    "0.7": 0.9996744791666666,
+    "0.8": 0.9998372395833334,
+    "0.9": 0.9998372395833334
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7931335237420475,
+    "0.6": 0.7993964084016625,
+    "0.7": 0.8001496013021097,
+    "0.8": 0.8002839884429704,
+    "0.9": 0.8002839884429704
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8626302083333334,
+    "0.6": 0.8291015625,
+    "0.7": 0.7692057291666666,
+    "0.8": 0.6790364583333334,
+    "0.9": 0.5166015625
+   },
+   "matched_cos_abs": {
+    "0.5": 0.802560031414032,
+    "0.6": 0.8040739893913269,
+    "0.7": 0.8038158416748047,
+    "0.8": 0.8024335503578186,
+    "0.9": 0.802299976348877
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9694010416666666,
+    "0.6": 0.9767252604166666,
+    "0.7": 0.9812825520833334,
+    "0.8": 0.9842122395833334,
+    "0.9": 0.9910481770833334
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7780025304527953,
+    "0.6": 0.7853593766825118,
+    "0.7": 0.788770460523665,
+    "0.8": 0.7897649217144742,
+    "0.9": 0.795117929034556
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8486328125,
+    "0.6": 0.81640625,
+    "0.7": 0.7591145833333334,
+    "0.8": 0.6722005208333334,
+    "0.9": 0.5130208333333334
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.8008715510368347,
+   "dcs_abs": 0.8024069666862488,
+   "frac_sign_flipped": 0.018229166666666668,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.8005561828613281,
+    "0.6": 0.8012158870697021,
+    "0.7": 0.8011359572410583,
+    "0.8": 0.8011359572410583,
+    "0.9": 0.8011359572410583
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9933268229166666,
+    "0.6": 0.99853515625,
+    "0.7": 0.9996744791666666,
+    "0.8": 0.9996744791666666,
+    "0.9": 0.9998372395833334
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.795213929687937,
+    "0.6": 0.8000422309851274,
+    "0.7": 0.8008751707966439,
+    "0.8": 0.8008751707966439,
+    "0.9": 0.8010055640188511
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8645833333333334,
+    "0.6": 0.8251953125,
+    "0.7": 0.7666015625,
+    "0.8": 0.6832682291666666,
+    "0.9": 0.5185546875
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8066564202308655,
+    "0.6": 0.806505024433136,
+    "0.7": 0.8066691160202026,
+    "0.8": 0.8052306175231934,
+    "0.9": 0.8031167984008789
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9674479166666666,
+    "0.6": 0.9754231770833334,
+    "0.7": 0.9794921875,
+    "0.8": 0.982421875,
+    "0.9": 0.9903971354166666
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7803980732181419,
+    "0.6": 0.7866836932662409,
+    "0.7": 0.7901260970393196,
+    "0.8": 0.7910761730745435,
+    "0.9": 0.795404576541235
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8489583333333334,
+    "0.6": 0.8121744791666666,
+    "0.7": 0.7591145833333334,
+    "0.8": 0.6761067708333334,
+    "0.9": 0.5146484375
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.7974991798400879,
+   "dcs_abs": 0.7989159226417542,
+   "frac_sign_flipped": 0.020182291666666668,
+   "alive_source": "stats",
+   "n_alive_a": 3072,
+   "n_alive_b": 3072,
+   "matched_cos_signed": {
+    "0.5": 0.7971132397651672,
+    "0.6": 0.7975146770477295,
+    "0.7": 0.79775470495224,
+    "0.8": 0.7977558374404907,
+    "0.9": 0.7974991798400879
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9930013020833334,
+    "0.6": 0.998046875,
+    "0.7": 0.9996744791666666,
+    "0.8": 0.9998372395833334,
+    "0.9": 1.0
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7915344849946754,
+    "0.6": 0.7959570311941206,
+    "0.7": 0.7974950191758883,
+    "0.8": 0.7976259943679906,
+    "0.9": 0.7974991798400879
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8597005208333334,
+    "0.6": 0.8180338541666666,
+    "0.7": 0.7639973958333334,
+    "0.8": 0.6826171875,
+    "0.9": 0.5081380208333334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.802304208278656,
+    "0.6": 0.8020427823066711,
+    "0.7": 0.8027011752128601,
+    "0.8": 0.8018078804016113,
+    "0.9": 0.799677312374115
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9671223958333334,
+    "0.6": 0.974609375,
+    "0.7": 0.97900390625,
+    "0.8": 0.9825846354166666,
+    "0.9": 0.990234375
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7759263680976195,
+    "0.6": 0.7816784147871658,
+    "0.7": 0.7858475860848557,
+    "0.8": 0.7878441038386275,
+    "0.9": 0.7918679636204615
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8427734375,
+    "0.6": 0.8030598958333334,
+    "0.7": 0.7545572916666666,
+    "0.8": 0.6744791666666666,
+    "0.9": 0.5042317708333334
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 66,
+  "n_dense_std": 1.5275252316519468,
+  "l0_mean": 64,
+  "l0_std": 4.5387399073123225e-06,
+  "recon_r2_mean": 0.9169558664447198,
+  "recon_r2_std": null,
+  "l1_per_token_mean": 438.1033236165672,
+  "l1_per_token_std": 1.0987534903299723,
+  "babel_mu1_signed_mean": 0.250858207543691,
+  "babel_mu1_signed_std": 0.0011338261385637365,
+  "n_dup_signed_per_threshold": {
+   "0.5": 43.333333333333336,
+   "0.6": 10.0,
+   "0.7": 2.0,
+   "0.8": 1.3333333333333333,
+   "0.9": 0.6666666666666666
+  },
+  "n_dup_signed_band": 3.5,
+  "n_dup_signed_band_std": 0.5,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 21.666666666666668,
+   "0.6": 5.0,
+   "0.7": 1.0,
+   "0.8": 0.6666666666666666,
+   "0.9": 0.3333333333333333
+  },
+  "n_dup_clusters_signed_band": 1.75,
+  "n_dup_clusters_signed_band_std": 0.25,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 2.0,
+   "0.6": 2.0,
+   "0.7": 2.0,
+   "0.8": 1.6666666666666667,
+   "0.9": 1.3333333333333333
+  },
+  "largest_dup_signed_band": 1.75,
+  "largest_dup_signed_band_std": 0.25,
+  "dcs_signed_mean": 0.799511730670929,
+  "dcs_signed_std": 0.0017784163368156788,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.7932939794748867,
+   "0.6": 0.7984652235269701,
+   "0.7": 0.7995065970915473,
+   "0.8": 0.799595051202535,
+   "0.9": 0.7995962441006365
+  },
+  "dedup_dcs_signed_band": 0.7992907789804222,
+  "dedup_dcs_signed_band_std": 0.0018889367534882992,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.7989286382993063,
+   "0.6": 0.7997665603955587,
+   "0.7": 0.7997669378916422,
+   "0.8": 0.7997686862945557,
+   "0.9": 0.799683133761088
+  },
+  "matched_cos_signed_band": 0.7997463295857111,
+  "matched_cos_signed_band_std": 0.0018653547551670977,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.9929470486111112,
+   "0.6": 0.9983723958333334,
+   "0.7": 0.9996744791666666,
+   "0.8": 0.9997829861111112,
+   "0.9": 0.9998914930555557
+  },
+  "unique_frac_signed_band": 0.9994303385416667,
+  "unique_frac_signed_band_std": 4.069010416668517e-05,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.8623046875,
+   "0.6": 0.8241102430555555,
+   "0.7": 0.7666015625,
+   "0.8": 0.681640625,
+   "0.9": 0.5144314236111112
+  },
+  "matched_frac_signed_band": 0.6966959635416666,
+  "matched_frac_signed_band_std": 0.00303079825024998,
+  "babel_mu1_abs_mean": 0.296555628379186,
+  "babel_mu1_abs_std": 0.002214914234765633,
+  "n_dup_abs_per_threshold": {
+   "0.5": 195.33333333333334,
+   "0.6": 149.0,
+   "0.7": 122.66666666666667,
+   "0.8": 103.33333333333333,
+   "0.9": 57.666666666666664
+  },
+  "n_dup_abs_band": 108.16666666666667,
+  "n_dup_abs_band_std": 11.153288005486694,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 97.0,
+   "0.6": 74.0,
+   "0.7": 61.0,
+   "0.8": 51.333333333333336,
+   "0.9": 28.666666666666668
+  },
+  "n_dup_clusters_abs_band": 53.75,
+  "n_dup_clusters_abs_band_std": 5.584576975922169,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 3.0,
+   "0.6": 3.0,
+   "0.7": 2.6666666666666665,
+   "0.8": 2.6666666666666665,
+   "0.9": 2.3333333333333335
+  },
+  "largest_dup_abs_band": 2.6666666666666665,
+  "largest_dup_abs_band_std": 0.3818813079129867,
+  "dcs_abs_mean": 0.8009351889292399,
+  "dcs_abs_std": 0.001808771654577537,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.7781089905895189,
+   "0.6": 0.7845738282453061,
+   "0.7": 0.7882480478826134,
+   "0.8": 0.7895617328758817,
+   "0.9": 0.7941301563987508
+  },
+  "dedup_dcs_abs_band": 0.789128441350638,
+  "dedup_dcs_abs_band_std": 0.002078219211877963,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.8038402199745178,
+   "0.6": 0.8042072653770447,
+   "0.7": 0.8043953776359558,
+   "0.8": 0.8031573494275411,
+   "0.9": 0.8016980290412903
+  },
+  "matched_cos_abs_band": 0.803364505370458,
+  "matched_cos_abs_band_std": 0.0019200735431181587,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.9679904513888888,
+   "0.6": 0.9755859375,
+   "0.7": 0.9799262152777778,
+   "0.8": 0.9830729166666666,
+   "0.9": 0.9905598958333334
+  },
+  "unique_frac_abs_band": 0.9822862413194444,
+  "unique_frac_abs_band_std": 0.0009074288565363911,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.8467881944444445,
+   "0.6": 0.810546875,
+   "0.7": 0.7575954861111112,
+   "0.8": 0.6742621527777778,
+   "0.9": 0.5106336805555557
+  },
+  "matched_frac_abs_band": 0.6882595486111112,
+  "matched_frac_abs_band_std": 0.0036214954687286798,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/config.json b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..1f74b95b2c2d080144475eee812aca8a0d7290d1
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/model.safetensors b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5d439ef8c317024fde1f8554e36baac46e34339e
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f58ae90bc9b8d414696d58d3f10e10cd72dff06add1b0d191695c1d52189905
+size 18890032
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..05342aec6090b6546064ba3d08efbd5678ed2326
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d32f23fd81286996eebbb066afd1b434695b0b36f36a328d7242c2e286a4e12
+size 25920
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c4ea1eccff501186dffcc46af1255a52788d80b1
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:915aebb32e0f83a73a60b18b6b48206a5d69551cf18979844dedc643335fd914
+size 13607
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..ad711901d005b99582c584093192374953c2fbf1
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.0,
+  "l0_per_token_0.001": 64.0,
+  "l0_per_token_0.01": 64.0,
+  "l0_per_token_0.05": 64.0,
+  "l0_per_token_0.1": 64.0,
+  "l0_per_token_0.5": 63.988106396484376,
+  "l0_per_token_1.0": 63.953277587890625,
+  "l1_per_token": 437.2697283288448,
+  "r2_score": 0.9169102820043679,
+  "l2_loss": 1.9486561136949205
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..39c63d86e1e6387645ae714c9b3b38a94f40134d
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5132439db5db99d7c719450df99ff8f0518ff1de94ae554e85515835d80c5fee
+size 13518
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e8cd375256245e5bafd69147b510a76f79fc7ea6
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:389abb428b4d4c39bd1b12cc1798bee4825ba482551a1fd10fd1b4486c7df91c
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7dd70b1afa8436fa2e7d231b451802afefb75afc
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:508eff3ebc76f6d72ea98a1effc1cd50555a85d9a806d6dc3fa8e14fde66956c
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/config.json b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..e47f2db1dc70bd7814656dc23d655f2f47731ade
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/model.safetensors b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..59a4585c29e7a0fd83f6f2a8b6590148dbe660ee
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:97514abcf0b7b7ec6add0ffebcbc7f3277bfcb969b7bdc47b43dbe8869155bbf
+size 18890032
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..18414496e9701af50803cc82cf3a9678e14f9653
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e4f3a8f63295764851ea817aea79bab5699164617ed1668966894059b4bac80b
+size 25920
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2143a156fcfe42a9a2f88bc990b97824e07faf6e
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:803891bcf2c543f1d73d7f0ae8c8f95f6d7d3e1a31b11f2d4713e910f17947db
+size 13607
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..192ed69a51094b186e1ddf403922161b6fbd6f54
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 63.99999213867188,
+  "l0_per_token_0.001": 63.9999919921875,
+  "l0_per_token_0.01": 63.9999908203125,
+  "l0_per_token_0.05": 63.999984130859374,
+  "l0_per_token_0.1": 63.999966357421876,
+  "l0_per_token_0.5": 63.984091064453125,
+  "l0_per_token_1.0": 63.945371875,
+  "l1_per_token": 437.6918106018625,
+  "r2_score": 0.9169844455894797,
+  "l2_loss": 1.9469167971220898
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..20812a595ee1d251b00b00ff764d8ab8249575c2
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86b201477bff8a1109b5d044759ebe3fe9376c5eca781d4e72dfc97b544203e0
+size 13518
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fbad6e70ec4287cf3def8533369fc190eb0b4338
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6bd1c8a8bbb0a828d19b5125fa26fe9b58842dbffcbff6b53fe9cb080887de97
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c35c053f49590d3e78f1d0ded0c3a838b412c454
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4fd002df14aa5b6c2ab5086dffefcca40d7229ecfee8f9fc5e64487354c70058
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/config.json b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..e9cd36975be00817755013b4d36e0b4519ef39de
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 4,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/model.safetensors b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..993482fe0950b26220e69cf0110c05ffaf0f0533
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2aadb9dda22316adad64962625d1644cfcd7561faa9382cb02e5ad18868ee529
+size 18890032
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..05342aec6090b6546064ba3d08efbd5678ed2326
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d32f23fd81286996eebbb066afd1b434695b0b36f36a328d7242c2e286a4e12
+size 25920
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1b009bc439c17e9438322d53840ee1810210d3ef
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:318edd77fd5b25a4b7dd87d8196f48d2a9b0d05c9f6c0d90085b03950f3328e4
+size 13607
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..ad29d6d4d0b19ff2991cee4798124ac696af525c
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 64.0,
+  "l0_per_token_0.001": 64.0,
+  "l0_per_token_0.01": 64.0,
+  "l0_per_token_0.05": 64.0,
+  "l0_per_token_0.1": 64.0,
+  "l0_per_token_0.5": 63.98780625,
+  "l0_per_token_1.0": 63.95329291992188,
+  "l1_per_token": 439.34843191899427,
+  "r2_score": 0.9169728717403118,
+  "l2_loss": 1.9471882320539216
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..46acb50615a05deabc13e94173e9cea07594ee0b
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16870516f85e12442369bb98c4540055ad45b658c5a47401fae30e58c7b0a586
+size 13518
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f01989c1ab67f3f86acaf6745fe8145f59a0a3ca
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:60982085957e4b71648e99ced2f1bff80d9285163bd1aa1d65d0e770fea783f7
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8050d46eebc38f6dfad35f53ad2b37df49594c4b
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 3072,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2bf52b4a2d0a2884b03c07481aa3d5e63a86e215
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3d5feadc6e1190603085f2eb21f4ec9e32338f82ba578c953b8dcc0741c3837
+size 13523
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/code_metrics_step_final.json b/lm/gpt2-small_layer8/topk/x8/l0_128/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..0b3ac3d40c58c7dce5a6f95a139df5cd00049ffe
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 127.99970166666667,
+   "recon_r2": 0.947772136062153,
+   "l1_per_token": 612.8000049998055
+  },
+  "seed1": {
+   "l0": 127.984445,
+   "recon_r2": 0.9477406721700997,
+   "l1_per_token": 613.9913856234108
+  },
+  "seed2": {
+   "l0": 127.99820833333334,
+   "recon_r2": 0.94775507654279,
+   "l1_per_token": 611.7738154068963
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7197505963762758,
+   "mcc_abs": 0.7197898394031813,
+   "act_iou": 0.3816880025431102,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9675973769940596,
+   "matched_mcc_signed": {
+    "0.6": 0.8607113841109026,
+    "0.7": 0.8866170166805184,
+    "0.8": 0.9137377612800062,
+    "0.9": 0.9450061731844813
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8612491950213407,
+    "0.7": 0.8872511574406977,
+    "0.8": 0.9146286656502456,
+    "0.9": 0.9449821630141415
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7222552817436686,
+   "mcc_abs": 0.7222927265231499,
+   "act_iou": 0.38390789335663844,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.967873232113979,
+   "matched_mcc_signed": {
+    "0.6": 0.8595772223071652,
+    "0.7": 0.8875933458062601,
+    "0.8": 0.9132034368993933,
+    "0.9": 0.944076057091245
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8592052515377617,
+    "0.7": 0.8880371517491169,
+    "0.8": 0.914165981187403,
+    "0.9": 0.9443717860421766
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7223994191881647,
+   "mcc_abs": 0.7224401197141063,
+   "act_iou": 0.3831886750519173,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9677526258056897,
+   "matched_mcc_signed": {
+    "0.6": 0.861872934959526,
+    "0.7": 0.8887721170838965,
+    "0.8": 0.9157306821529779,
+    "0.9": 0.9445422292437436
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8619444858049542,
+    "0.7": 0.8891683195566282,
+    "0.8": 0.9157179392102717,
+    "0.9": 0.9448128455252092
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 127.99411833333333,
+  "l0_std": 0.008410561482120143,
+  "recon_r2_mean": 0.9477559615916809,
+  "recon_r2_std": 1.575060663721799e-05,
+  "l1_per_token_mean": 612.8550686767043,
+  "l1_per_token_std": 1.1098100840586123,
+  "input_r2_mean": 0.967741078304576,
+  "input_r2_std": 0.00013828962500101098,
+  "mcc_signed_mean": 0.7214684324360364,
+  "mcc_signed_std": 0.0014894342707507363,
+  "mcc_abs_mean": 0.7215075618801459,
+  "mcc_abs_std": 0.0014894156804219957,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8607205137925312,
+   "0.7": 0.8876608265235584,
+   "0.8": 0.9142239601107924,
+   "0.9": 0.94454148650649
+  },
+  "matched_mcc_signed_band": 0.901786696733343,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.8607996441213522,
+   "0.7": 0.8881522095821476,
+   "0.8": 0.9148375286826401,
+   "0.9": 0.9447222648605091
+  },
+  "matched_mcc_abs_band": 0.9021279118116623,
+  "act_iou_mean": 0.382928190317222,
+  "act_iou_std": 0.0011326376422994867,
+  "act_iou_n_pairs": 6144.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/dict_metrics_step_final.json b/lm/gpt2-small_layer8/topk/x8/l0_128/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..df96208f03ac3a5c0f95b7142a444595696418c5
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/dict_metrics_step_final.json
@@ -0,0 +1,545 @@
+{
+ "schema": 2,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 127.99970029296875,
+   "recon_r2": 0.9476983010517099,
+   "l1_per_token": 612.6701362468667,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 274,
+   "dense_rate": 0.044596354166666664,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 90,
+    "0.6": 18,
+    "0.7": 8,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 45,
+    "0.6": 9,
+    "0.7": 4,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "largest_dup_signed": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 2,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "babel_mu1_signed": 0.25818774104118347,
+   "n_dup_abs": {
+    "0.5": 729,
+    "0.6": 617,
+    "0.7": 578,
+    "0.8": 545,
+    "0.9": 469
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 363,
+    "0.6": 307,
+    "0.7": 288,
+    "0.8": 272,
+    "0.9": 234
+   },
+   "largest_dup_abs": {
+    "0.5": 3,
+    "0.6": 3,
+    "0.7": 3,
+    "0.8": 3,
+    "0.9": 3
+   },
+   "babel_mu1_abs": 0.3451387882232666
+  },
+  "seed1": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 127.98429594726562,
+   "recon_r2": 0.9476596205713882,
+   "l1_per_token": 613.7922145471359,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 282,
+   "dense_rate": 0.0458984375,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 64,
+    "0.6": 6,
+    "0.7": 2,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 32,
+    "0.6": 3,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "largest_dup_signed": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 2,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "babel_mu1_signed": 0.2580951452255249,
+   "n_dup_abs": {
+    "0.5": 717,
+    "0.6": 613,
+    "0.7": 583,
+    "0.8": 551,
+    "0.9": 455
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 357,
+    "0.6": 306,
+    "0.7": 291,
+    "0.8": 275,
+    "0.9": 227
+   },
+   "largest_dup_abs": {
+    "0.5": 3,
+    "0.6": 3,
+    "0.7": 3,
+    "0.8": 3,
+    "0.9": 3
+   },
+   "babel_mu1_abs": 0.34565138816833496
+  },
+  "seed2": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 128,
+   "l0_raw": 127.99805942382812,
+   "recon_r2": 0.9476707713898209,
+   "l1_per_token": 611.6318825384176,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 269,
+   "dense_rate": 0.043782552083333336,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 76,
+    "0.6": 16,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 38,
+    "0.6": 8,
+    "0.7": 0,
+    "0.8": 0,
+    "0.9": 0
+   },
+   "largest_dup_signed": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 1,
+    "0.8": 1,
+    "0.9": 1
+   },
+   "babel_mu1_signed": 0.25812530517578125,
+   "n_dup_abs": {
+    "0.5": 696,
+    "0.6": 606,
+    "0.7": 572,
+    "0.8": 538,
+    "0.9": 448
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 348,
+    "0.6": 303,
+    "0.7": 286,
+    "0.8": 269,
+    "0.9": 224
+   },
+   "largest_dup_abs": {
+    "0.5": 2,
+    "0.6": 2,
+    "0.7": 2,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "babel_mu1_abs": 0.3437892496585846
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.7193899154663086,
+   "dcs_abs": 0.7209102511405945,
+   "frac_sign_flipped": 0.028483072916666668,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.7183835506439209,
+    "0.6": 0.7194072604179382,
+    "0.7": 0.7194593548774719,
+    "0.8": 0.7193480730056763,
+    "0.9": 0.7193480730056763
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9937337239583334,
+    "0.6": 0.9990234375,
+    "0.7": 0.9995930989583334,
+    "0.8": 0.9998372395833334,
+    "0.9": 0.9998372395833334
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7138819610117935,
+    "0.6": 0.7187047142651863,
+    "0.7": 0.7191666061165355,
+    "0.8": 0.7192309916135855,
+    "0.9": 0.7192309916135855
+   },
+   "matched_frac_signed": {
+    "0.5": 0.7841796875,
+    "0.6": 0.7234700520833334,
+    "0.7": 0.64013671875,
+    "0.8": 0.5236002604166666,
+    "0.9": 0.3064778645833333
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7281427979469299,
+    "0.6": 0.7288630604743958,
+    "0.7": 0.7285162806510925,
+    "0.8": 0.72804856300354,
+    "0.9": 0.7262296676635742
+   },
+   "unique_frac_abs": {
+    "0.5": 0.94091796875,
+    "0.6": 0.9497884114583334,
+    "0.7": 0.95263671875,
+    "0.8": 0.955322265625,
+    "0.9": 0.9623209635416666
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.685122642404167,
+    "0.6": 0.6922656883786356,
+    "0.7": 0.6940113591554109,
+    "0.8": 0.6955210026935674,
+    "0.9": 0.698866033538555
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7581380208333334,
+    "0.6": 0.7023111979166666,
+    "0.7": 0.6241861979166666,
+    "0.8": 0.5115559895833334,
+    "0.9": 0.2994791666666667
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.722062885761261,
+   "dcs_abs": 0.7232913970947266,
+   "frac_sign_flipped": 0.023763020833333332,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.7211748361587524,
+    "0.6": 0.7221399545669556,
+    "0.7": 0.7223819494247437,
+    "0.8": 0.7221794128417969,
+    "0.9": 0.7221794128417969
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9932454427083334,
+    "0.6": 0.9986165364583334,
+    "0.7": 0.9996744791666666,
+    "0.8": 0.9999186197916666,
+    "0.9": 0.9999186197916666
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7163036194106098,
+    "0.6": 0.7211409002678314,
+    "0.7": 0.7221467990505819,
+    "0.8": 0.7221206417307258,
+    "0.9": 0.7221206417307258
+   },
+   "matched_frac_signed": {
+    "0.5": 0.78515625,
+    "0.6": 0.7272135416666666,
+    "0.7": 0.6398111979166666,
+    "0.8": 0.5255533854166666,
+    "0.9": 0.3126627604166667
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7315207123756409,
+    "0.6": 0.7316517233848572,
+    "0.7": 0.7312687635421753,
+    "0.8": 0.7306620478630066,
+    "0.9": 0.7295863628387451
+   },
+   "unique_frac_abs": {
+    "0.5": 0.94189453125,
+    "0.6": 0.9501139322916666,
+    "0.7": 0.953125,
+    "0.8": 0.9558919270833334,
+    "0.9": 0.962646484375
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.6890153584827203,
+    "0.6": 0.6951524959731614,
+    "0.7": 0.6969905402511358,
+    "0.8": 0.6984339529784241,
+    "0.9": 0.7023337472346611
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7604166666666666,
+    "0.6": 0.7080078125,
+    "0.7": 0.6236979166666666,
+    "0.8": 0.5128580729166666,
+    "0.9": 0.3068033854166667
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.720814049243927,
+   "dcs_abs": 0.7222006320953369,
+   "frac_sign_flipped": 0.023111979166666668,
+   "alive_source": "stats",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.7199376821517944,
+    "0.6": 0.7211160659790039,
+    "0.7": 0.720928966999054,
+    "0.8": 0.720928966999054,
+    "0.9": 0.720928966999054
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9943033854166666,
+    "0.6": 0.9991048177083334,
+    "0.7": 0.9999186197916666,
+    "0.8": 0.9999186197916666,
+    "0.9": 0.9999186197916666
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7158364746525573,
+    "0.6": 0.7204705356465032,
+    "0.7": 0.720870297649526,
+    "0.8": 0.720870297649526,
+    "0.9": 0.720870297649526
+   },
+   "matched_frac_signed": {
+    "0.5": 0.78662109375,
+    "0.6": 0.7213541666666666,
+    "0.7": 0.6344401041666666,
+    "0.8": 0.5218098958333334,
+    "0.9": 0.314453125
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7307726740837097,
+    "0.6": 0.7307832837104797,
+    "0.7": 0.7308114171028137,
+    "0.8": 0.7302628755569458,
+    "0.9": 0.7284812927246094
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9423828125,
+    "0.6": 0.9503580729166666,
+    "0.7": 0.9529622395833334,
+    "0.8": 0.9556477864583334,
+    "0.9": 0.9632161458333334
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.6886676079011522,
+    "0.6": 0.6945057932268052,
+    "0.7": 0.696435684755367,
+    "0.8": 0.6978741005586926,
+    "0.9": 0.7016849430898826
+   },
+   "matched_frac_abs": {
+    "0.5": 0.7594401041666666,
+    "0.6": 0.7018229166666666,
+    "0.7": 0.6194661458333334,
+    "0.8": 0.5110677083333334,
+    "0.9": 0.30712890625
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 275,
+  "n_dense_std": 6.557438524302,
+  "l0_mean": 128,
+  "l0_std": 0.008459901547544018,
+  "recon_r2_mean": 0.9476762310043063,
+  "recon_r2_std": null,
+  "l1_per_token_mean": 612.6980777774735,
+  "l1_per_token_std": 1.0804369940572296,
+  "babel_mu1_signed_mean": 0.2581360638141632,
+  "babel_mu1_signed_std": 4.7226131476590836e-05,
+  "n_dup_signed_per_threshold": {
+   "0.5": 76.66666666666667,
+   "0.6": 13.333333333333334,
+   "0.7": 3.3333333333333335,
+   "0.8": 1.3333333333333333,
+   "0.9": 1.3333333333333333
+  },
+  "n_dup_signed_band": 4.833333333333333,
+  "n_dup_signed_band_std": 2.362907813126304,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 38.333333333333336,
+   "0.6": 6.666666666666667,
+   "0.7": 1.6666666666666667,
+   "0.8": 0.6666666666666666,
+   "0.9": 0.6666666666666666
+  },
+  "n_dup_clusters_signed_band": 2.4166666666666665,
+  "n_dup_clusters_signed_band_std": 1.181453906563152,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 2.0,
+   "0.6": 2.0,
+   "0.7": 1.6666666666666667,
+   "0.8": 1.6666666666666667,
+   "0.9": 1.6666666666666667
+  },
+  "largest_dup_signed_band": 1.7500000000000002,
+  "largest_dup_signed_band_std": 0.4330127018922193,
+  "dcs_signed_mean": 0.7207556168238322,
+  "dcs_signed_std": 0.0013374428250258297,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.7153406850249868,
+   "0.6": 0.7201053833931738,
+   "0.7": 0.7207279009388811,
+   "0.8": 0.7207406436646124,
+   "0.9": 0.7207406436646124
+  },
+  "dedup_dcs_signed_band": 0.7205786429153199,
+  "dedup_dcs_signed_band_std": 0.0014092741968945073,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.7198320229848226,
+   "0.6": 0.7208877603212992,
+   "0.7": 0.7209234237670898,
+   "0.8": 0.720818817615509,
+   "0.9": 0.720818817615509
+  },
+  "matched_cos_signed_band": 0.7208622048298517,
+  "matched_cos_signed_band_std": 0.0014181587880145743,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.9937608506944445,
+   "0.6": 0.9989149305555557,
+   "0.7": 0.9997287326388888,
+   "0.8": 0.9998914930555555,
+   "0.9": 0.9998914930555555
+  },
+  "unique_frac_signed_band": 0.9996066623263888,
+  "unique_frac_signed_band_std": 9.614696505230722e-05,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.7853190104166666,
+   "0.6": 0.7240125868055555,
+   "0.7": 0.6381293402777777,
+   "0.8": 0.5236545138888888,
+   "0.9": 0.3111979166666667
+  },
+  "matched_frac_signed_band": 0.5492485894097221,
+  "matched_frac_signed_band_std": 0.001796979916468915,
+  "babel_mu1_abs_mean": 0.3448598086833954,
+  "babel_mu1_abs_std": 0.000961905476189784,
+  "n_dup_abs_per_threshold": {
+   "0.5": 714.0,
+   "0.6": 612.0,
+   "0.7": 577.6666666666666,
+   "0.8": 544.6666666666666,
+   "0.9": 457.3333333333333
+  },
+  "n_dup_abs_band": 547.9166666666666,
+  "n_dup_abs_band_std": 6.053580207888,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 356.0,
+   "0.6": 305.3333333333333,
+   "0.7": 288.3333333333333,
+   "0.8": 272.0,
+   "0.9": 228.33333333333334
+  },
+  "n_dup_clusters_abs_band": 273.5,
+  "n_dup_clusters_abs_band_std": 2.6100766272276377,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 2.6666666666666665,
+   "0.6": 2.6666666666666665,
+   "0.7": 2.6666666666666665,
+   "0.8": 2.6666666666666665,
+   "0.9": 2.6666666666666665
+  },
+  "largest_dup_abs_band": 2.6666666666666665,
+  "largest_dup_abs_band_std": 0.5773502691896258,
+  "dcs_abs_mean": 0.7221340934435526,
+  "dcs_abs_std": 0.001191966676487159,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.6876018695960132,
+   "0.6": 0.6939746591928674,
+   "0.7": 0.6958125280539713,
+   "0.8": 0.6972763520768948,
+   "0.9": 0.7009615746210329
+  },
+  "dedup_dcs_abs_band": 0.6970062784861916,
+  "dedup_dcs_abs_band_std": 0.001621936686133217,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.7301453948020935,
+   "0.6": 0.7304326891899109,
+   "0.7": 0.7301988204320272,
+   "0.8": 0.7296578288078308,
+   "0.9": 0.7280991077423096
+  },
+  "matched_cos_abs_band": 0.7295971115430196,
+  "matched_cos_abs_band_std": 0.001499599275847057,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.9417317708333334,
+   "0.6": 0.9500868055555555,
+   "0.7": 0.9529079861111112,
+   "0.8": 0.9556206597222223,
+   "0.9": 0.9627278645833334
+  },
+  "unique_frac_abs_band": 0.9553358289930556,
+  "unique_frac_abs_band_std": 0.0002806830795162885,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.7593315972222222,
+   "0.6": 0.7040473090277777,
+   "0.7": 0.6224500868055555,
+   "0.8": 0.5118272569444445,
+   "0.9": 0.3044704861111111
+  },
+  "matched_frac_abs_band": 0.5356987847222222,
+  "matched_frac_abs_band_std": 0.0018718922137746484,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/config.json b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..20071d712b97551e507642401ad3a55036610413
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/model.safetensors b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..966a43d1c32dfabb46708ed0d490bb358affaf09
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:30eb0e23b86c37ef8515df48fbcf02b00930093c6b8e1b1c8250acea9b7c1b67
+size 37776696
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c2454087ca716dbdddcba9a2df6e8213a3e8204e
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99af137e331265b473a1ea6d4ece532bc0f82f8945a08e6c3db2a0c3820eff76
+size 50496
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d980492904ab8db4228f2982bb05bbaacc858f52
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:510ae19324bbfd2244923e1b3299572dd1f82581efc92aa2a2908330f9d60f42
+size 25895
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..0001804f2650999e2eeb61b995c45bf3ed4c6717
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 127.99970029296875,
+  "l0_per_token_0.001": 127.99969741210937,
+  "l0_per_token_0.01": 127.9996708984375,
+  "l0_per_token_0.05": 127.9994197265625,
+  "l0_per_token_0.1": 127.9987361328125,
+  "l0_per_token_0.5": 127.9546068359375,
+  "l0_per_token_1.0": 127.83810830078124,
+  "l1_per_token": 612.6701362468667,
+  "r2_score": 0.9476983010517099,
+  "l2_loss": 1.2266021340639794
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..55648fccdb559f43693251596accf369bf9ed35f
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e4b12f060b889bec1ceef4844be0423fbe1e11b08576e010b7bbcccffd8806fe
+size 25806
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..86a5a3daf83e1c1454c47c16ac1081eec93ebe49
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f119a4fbc3f48ed87a5603e60012302e28c1e7aaa36f9fe99780feb7340b6463
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3cb14a73fc13320f881ac300c1ad1d5b3240d17f
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e612cb8bcc8f93d2bac52567dd9c9df1960e8d5696a4650b8bba2167a552c9c
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/config.json b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..556ec95876818cad5126cb5a7fb3e52b5c9b02f6
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/model.safetensors b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..64ad66ad7d44907be00143058f88a8198eb482d2
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d78c4d4d948e8ec2f124954481f742cde9d016f9a042745d53441d9a0c76318
+size 37776696
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a883299da1c04cb4c9ca3b7063232c19008f226f
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9f5545dc15eb6379812e62781a3b3492ed06dda635f4baaae6bc8926d02182e
+size 50496
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c7842d892688cef8ff5a94adb43d0c9803115702
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c827fbe5a0dfc4c133efb6bbf135e05d7d2d5092deda9bb72fd29d044e9bbf46
+size 25895
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..2d9b74566248fd8a80b7e1c93accb713c6a6855d
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 127.98429594726562,
+  "l0_per_token_0.001": 127.9842587890625,
+  "l0_per_token_0.01": 127.98393344726563,
+  "l0_per_token_0.05": 127.98229125976563,
+  "l0_per_token_0.1": 127.98031923828125,
+  "l0_per_token_0.5": 127.93932919921875,
+  "l0_per_token_1.0": 127.83192153320313,
+  "l1_per_token": 613.7922145471359,
+  "r2_score": 0.9476596205713882,
+  "l2_loss": 1.227509285469448
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..88f4a2ec09acc6868e351e9988863b01809c1cc0
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:180ea975a5d9e775ff8b5e9f52835a38869db997c0d82b20b18a380026e7df99
+size 25806
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..16ba67e7842aa2a61d9b08fa5c8ba22fb4ffe34c
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2721533016ed742da07a7ebe8f2c8e91fdfd5dae94f404e379579b1ff0e88533
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d43c705f9e5194ad285acf2e6680285c0ac23d99
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0025d87f5247a46aadd11825738bce1a262624d01a005a6fc1845e0be025a055
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/config.json b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..c633eec338943c277944540dbafcbdd6750fb4af
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 128,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/model.safetensors b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..73334d08576b796cb6b5b0f3467c9acd7a8d898c
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f588ddd57d00a0e2b093af52b0d54e77b2c277c0f9e8f798fa33b52f25840a41
+size 37776696
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..502959bbdd66a7eed9d8b0107364e03b23ec53fe
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4bc87e4067a82d1487c9486946286ffadfdedb7f62470522c91ac0de74bba2b5
+size 50496
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b4a64fa8fe3da3ce2ce6eaed53c564849999ca59
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8f51c4e820f774590c5cba9648f5aa2af68409c8eedc0ae32b8fb96c061fdfa
+size 25895
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..64cbddab6ee7c7e832bc17df13380923b6cb481a
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 127.99805942382812,
+  "l0_per_token_0.001": 127.99800805664063,
+  "l0_per_token_0.01": 127.99763745117187,
+  "l0_per_token_0.05": 127.99632573242188,
+  "l0_per_token_0.1": 127.99492973632813,
+  "l0_per_token_0.5": 127.9496794921875,
+  "l0_per_token_1.0": 127.83888159179688,
+  "l1_per_token": 611.6318825384176,
+  "r2_score": 0.9476707713898209,
+  "l2_loss": 1.227247771637944
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..6c764a00366d429886cf9433f2a8b0efc21720fb
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9bb2ae51b1fdd7077f2c9cdb00f246a2bbad739553c7ac272c55d65f1799ba0f
+size 25806
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a26d08e26c229204a64237d68dbedd31714f79b8
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79c6a13fb788eff4be3f1b778ce1a5650ae99706dac93a356962654cf001f70d
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..dc40f88cffb3877991d80950c012c2a13758f692
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d9e6065c1085ef0accd1d0aab22069274faf74dc6218cda0f69156aff1855bb
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/code_metrics_step_final.json b/lm/gpt2-small_layer8/topk/x8/l0_32/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..c476d7c53fc95c968e40414bc3f2215cf76b6ed2
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 32.0,
+   "recon_r2": 0.9080914980888211,
+   "l1_per_token": 305.8815994780672
+  },
+  "seed1": {
+   "l0": 32.0,
+   "recon_r2": 0.9081274089303089,
+   "l1_per_token": 311.7732023085558
+  },
+  "seed2": {
+   "l0": 32.0,
+   "recon_r2": 0.908004007309433,
+   "l1_per_token": 306.38841370745865
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7659129303233656,
+   "mcc_abs": 0.7659362918531286,
+   "act_iou": 0.46789434659765455,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9657026544753862,
+   "matched_mcc_signed": {
+    "0.6": 0.8548901317965886,
+    "0.7": 0.8796267680666926,
+    "0.8": 0.9015021593760298,
+    "0.9": 0.9272282066449454
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8546360068477926,
+    "0.7": 0.8794098558680772,
+    "0.8": 0.9008836092853577,
+    "0.9": 0.9269834807115763
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.76869142592109,
+   "mcc_abs": 0.7687148480438436,
+   "act_iou": 0.47107953231746585,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9656036297800568,
+   "matched_mcc_signed": {
+    "0.6": 0.8576381329380721,
+    "0.7": 0.8798542070833352,
+    "0.8": 0.9003892098839262,
+    "0.9": 0.9296024942712704
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.856832138584284,
+    "0.7": 0.879326431108144,
+    "0.8": 0.9001180264521071,
+    "0.9": 0.9294984960156434
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7644747178394456,
+   "mcc_abs": 0.7645026828548748,
+   "act_iou": 0.46444407337560084,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9649890720500524,
+   "matched_mcc_signed": {
+    "0.6": 0.851385125061681,
+    "0.7": 0.8768435127871552,
+    "0.8": 0.9005939641632161,
+    "0.9": 0.9304257793086937
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8516894156487563,
+    "0.7": 0.8765639320138194,
+    "0.8": 0.9000005186486586,
+    "0.9": 0.9302246447370714
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 32.0,
+  "l0_std": 0.0,
+  "recon_r2_mean": 0.9080743047761876,
+  "recon_r2_std": 6.34720214597053e-05,
+  "l1_per_token_mean": 308.01440516469387,
+  "l1_per_token_std": 3.265062348915699,
+  "input_r2_mean": 0.9654317854351651,
+  "input_r2_std": 0.00038658482715926197,
+  "mcc_signed_mean": 0.7663596913613003,
+  "mcc_signed_std": 0.0021435608995812455,
+  "mcc_abs_mean": 0.766384607583949,
+  "mcc_abs_std": 0.0021415704846364423,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8546377965987806,
+   "0.7": 0.8787748293123944,
+   "0.8": 0.9008284444743907,
+   "0.9": 0.9290854934083032
+  },
+  "matched_mcc_signed_band": 0.8908316409484672,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.854385853693611,
+   "0.7": 0.8784334063300135,
+   "0.8": 0.9003340514620412,
+   "0.9": 0.9289022071547638
+  },
+  "matched_mcc_abs_band": 0.8905138796601074,
+  "act_iou_mean": 0.4678059840969071,
+  "act_iou_std": 0.0033186118771178787,
+  "act_iou_n_pairs": 6144.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/dict_metrics_step_final.json b/lm/gpt2-small_layer8/topk/x8/l0_32/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..7997e50711975e30cfe00ff3c62be1ed0c54bc5b
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/dict_metrics_step_final.json
@@ -0,0 +1,595 @@
+{
+ "schema": 3,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 32.0,
+   "recon_r2": 0.9079532803360462,
+   "l1_per_token": 305.9132832406872,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 18,
+   "dense_rate": 0.0029296875,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 676,
+    "0.6": 250,
+    "0.7": 90,
+    "0.8": 26,
+    "0.9": 24
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 312,
+    "0.6": 109,
+    "0.7": 33,
+    "0.8": 2,
+    "0.9": 2
+   },
+   "n_dup_removed_signed": {
+    "0.5": 364,
+    "0.6": 141,
+    "0.7": 57,
+    "0.8": 24,
+    "0.9": 22
+   },
+   "largest_dup_signed": {
+    "0.5": 24,
+    "0.6": 24,
+    "0.7": 24,
+    "0.8": 24,
+    "0.9": 22
+   },
+   "babel_mu1_signed": 0.3792347013950348,
+   "n_dup_abs": {
+    "0.5": 789,
+    "0.6": 337,
+    "0.7": 158,
+    "0.8": 72,
+    "0.9": 36
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 365,
+    "0.6": 151,
+    "0.7": 66,
+    "0.8": 25,
+    "0.9": 8
+   },
+   "n_dup_removed_abs": {
+    "0.5": 424,
+    "0.6": 186,
+    "0.7": 92,
+    "0.8": 47,
+    "0.9": 28
+   },
+   "largest_dup_abs": {
+    "0.5": 24,
+    "0.6": 24,
+    "0.7": 24,
+    "0.8": 24,
+    "0.9": 22
+   },
+   "babel_mu1_abs": 0.390911340713501,
+   "dup_excludes_dead": true
+  },
+  "seed1": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 32.0,
+   "recon_r2": 0.9079910218319425,
+   "l1_per_token": 311.5163819885298,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 18,
+   "dense_rate": 0.0029296875,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 631,
+    "0.6": 209,
+    "0.7": 75,
+    "0.8": 36,
+    "0.9": 25
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 293,
+    "0.6": 91,
+    "0.7": 26,
+    "0.8": 8,
+    "0.9": 5
+   },
+   "n_dup_removed_signed": {
+    "0.5": 338,
+    "0.6": 118,
+    "0.7": 49,
+    "0.8": 28,
+    "0.9": 20
+   },
+   "largest_dup_signed": {
+    "0.5": 21,
+    "0.6": 21,
+    "0.7": 21,
+    "0.8": 21,
+    "0.9": 13
+   },
+   "babel_mu1_signed": 0.3778785765171051,
+   "n_dup_abs": {
+    "0.5": 740,
+    "0.6": 300,
+    "0.7": 146,
+    "0.8": 84,
+    "0.9": 41
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 342,
+    "0.6": 134,
+    "0.7": 60,
+    "0.8": 31,
+    "0.9": 12
+   },
+   "n_dup_removed_abs": {
+    "0.5": 398,
+    "0.6": 166,
+    "0.7": 86,
+    "0.8": 53,
+    "0.9": 29
+   },
+   "largest_dup_abs": {
+    "0.5": 21,
+    "0.6": 21,
+    "0.7": 21,
+    "0.8": 21,
+    "0.9": 13
+   },
+   "babel_mu1_abs": 0.3900517523288727,
+   "dup_excludes_dead": true
+  },
+  "seed2": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 32,
+   "l0_raw": 31.999998486328124,
+   "recon_r2": 0.9078798382662103,
+   "l1_per_token": 306.22344391044953,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 20,
+   "dense_rate": 0.0032552083333333335,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 653,
+    "0.6": 220,
+    "0.7": 76,
+    "0.8": 33,
+    "0.9": 25
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 305,
+    "0.6": 97,
+    "0.7": 27,
+    "0.8": 7,
+    "0.9": 3
+   },
+   "n_dup_removed_signed": {
+    "0.5": 348,
+    "0.6": 123,
+    "0.7": 49,
+    "0.8": 26,
+    "0.9": 22
+   },
+   "largest_dup_signed": {
+    "0.5": 21,
+    "0.6": 21,
+    "0.7": 21,
+    "0.8": 21,
+    "0.9": 21
+   },
+   "babel_mu1_signed": 0.3786081373691559,
+   "n_dup_abs": {
+    "0.5": 762,
+    "0.6": 301,
+    "0.7": 144,
+    "0.8": 78,
+    "0.9": 40
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 355,
+    "0.6": 134,
+    "0.7": 59,
+    "0.8": 28,
+    "0.9": 10
+   },
+   "n_dup_removed_abs": {
+    "0.5": 407,
+    "0.6": 167,
+    "0.7": 85,
+    "0.8": 50,
+    "0.9": 30
+   },
+   "largest_dup_abs": {
+    "0.5": 21,
+    "0.6": 21,
+    "0.7": 21,
+    "0.8": 21,
+    "0.9": 21
+   },
+   "babel_mu1_abs": 0.39009419083595276,
+   "dup_excludes_dead": true
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.822551965713501,
+   "dcs_abs": 0.8231813311576843,
+   "frac_sign_flipped": 0.01025390625,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.8178921937942505,
+    "0.6": 0.8224523663520813,
+    "0.7": 0.822685182094574,
+    "0.8": 0.8227839469909668,
+    "0.9": 0.8226262927055359
+   },
+   "unique_frac_signed": {
+    "0.5": 0.94287109375,
+    "0.6": 0.9789225260416666,
+    "0.7": 0.9913736979166666,
+    "0.8": 0.9957682291666666,
+    "0.9": 0.99658203125
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7711669073323719,
+    "0.6": 0.8051171480183257,
+    "0.7": 0.8155884511943441,
+    "0.8": 0.8193021138819555,
+    "0.9": 0.81981458174414
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8304036458333334,
+    "0.6": 0.8263346354166666,
+    "0.7": 0.7841796875,
+    "0.8": 0.71240234375,
+    "0.9": 0.5589192708333334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8192251920700073,
+    "0.6": 0.82337886095047,
+    "0.7": 0.8237749934196472,
+    "0.8": 0.8237028121948242,
+    "0.9": 0.8229721188545227
+   },
+   "unique_frac_abs": {
+    "0.5": 0.93310546875,
+    "0.6": 0.9713541666666666,
+    "0.7": 0.9855143229166666,
+    "0.8": 0.9918619791666666,
+    "0.9": 0.995361328125
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.764423506858293,
+    "0.6": 0.7997924873294929,
+    "0.7": 0.8118420548756451,
+    "0.8": 0.8169995015487075,
+    "0.9": 0.8191546212328831
+   },
+   "matched_frac_abs": {
+    "0.5": 0.822265625,
+    "0.6": 0.8216145833333334,
+    "0.7": 0.7820638020833334,
+    "0.8": 0.7107747395833334,
+    "0.9": 0.5579427083333334
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.8262729048728943,
+   "dcs_abs": 0.827171802520752,
+   "frac_sign_flipped": 0.010091145833333334,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.8199571967124939,
+    "0.6": 0.8254126310348511,
+    "0.7": 0.8265976905822754,
+    "0.8": 0.8262924551963806,
+    "0.9": 0.8259562849998474
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9420572916666666,
+    "0.6": 0.978515625,
+    "0.7": 0.9913736979166666,
+    "0.8": 0.9959309895833334,
+    "0.9": 0.9964192708333334
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7724466560175642,
+    "0.6": 0.8076791565399617,
+    "0.7": 0.819467209201927,
+    "0.8": 0.8229302625889735,
+    "0.9": 0.8229987592397568
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8274739583333334,
+    "0.6": 0.82958984375,
+    "0.7": 0.7931315104166666,
+    "0.8": 0.7233072916666666,
+    "0.9": 0.5641276041666666
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8209528923034668,
+    "0.6": 0.8271118998527527,
+    "0.7": 0.8278838992118835,
+    "0.8": 0.8272501230239868,
+    "0.9": 0.8268340229988098
+   },
+   "unique_frac_abs": {
+    "0.5": 0.932373046875,
+    "0.6": 0.9712727864583334,
+    "0.7": 0.985595703125,
+    "0.8": 0.9921061197916666,
+    "0.9": 0.9952799479166666
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7654343495378271,
+    "0.6": 0.803351279682829,
+    "0.7": 0.815958813749603,
+    "0.8": 0.8207199096505065,
+    "0.9": 0.8229313233459834
+   },
+   "matched_frac_abs": {
+    "0.5": 0.81787109375,
+    "0.6": 0.8260091145833334,
+    "0.7": 0.7919921875,
+    "0.8": 0.7220052083333334,
+    "0.9": 0.56396484375
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.8203430771827698,
+   "dcs_abs": 0.8211563229560852,
+   "frac_sign_flipped": 0.0107421875,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.8139887452125549,
+    "0.6": 0.8187050223350525,
+    "0.7": 0.8198452591896057,
+    "0.8": 0.8202481269836426,
+    "0.9": 0.820119321346283
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9441731770833334,
+    "0.6": 0.9803873697916666,
+    "0.7": 0.9920247395833334,
+    "0.8": 0.99560546875,
+    "0.9": 0.99658203125
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7685463396774139,
+    "0.6": 0.8026480634822898,
+    "0.7": 0.813306779746199,
+    "0.8": 0.816643520956859,
+    "0.9": 0.8173161791346502
+   },
+   "matched_frac_signed": {
+    "0.5": 0.83056640625,
+    "0.6": 0.8292643229166666,
+    "0.7": 0.7864583333333334,
+    "0.8": 0.7019856770833334,
+    "0.9": 0.5418294270833334
+   },
+   "matched_cos_abs": {
+    "0.5": 0.8153641223907471,
+    "0.6": 0.8199777007102966,
+    "0.7": 0.8212684392929077,
+    "0.8": 0.8211085200309753,
+    "0.9": 0.8208771347999573
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9344889322916666,
+    "0.6": 0.972900390625,
+    "0.7": 0.986083984375,
+    "0.8": 0.9916178385416666,
+    "0.9": 0.9951985677083334
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.761948748161861,
+    "0.6": 0.7977566253248369,
+    "0.7": 0.8098396548593882,
+    "0.8": 0.8142258558412625,
+    "0.9": 0.816935748817438
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8221028645833334,
+    "0.6": 0.8230794270833334,
+    "0.7": 0.783203125,
+    "0.8": 0.70068359375,
+    "0.9": 0.54150390625
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 19,
+  "n_dense_std": 1.1547005383792515,
+  "l0_mean": 32,
+  "l0_std": 8.739188649817303e-07,
+  "recon_r2_mean": 0.907941380144733,
+  "recon_r2_std": 5.65389886467391e-05,
+  "l1_per_token_mean": 307.88436971322216,
+  "l1_per_token_std": 3.1492355876078713,
+  "babel_mu1_signed_mean": 0.37857380509376526,
+  "babel_mu1_signed_std": 0.0006787140045594136,
+  "n_dup_signed_per_threshold": {
+   "0.5": 653.3333333333334,
+   "0.6": 226.33333333333334,
+   "0.7": 80.33333333333333,
+   "0.8": 31.666666666666668,
+   "0.9": 24.666666666666668
+  },
+  "n_dup_signed_band": 90.75000000000001,
+  "n_dup_signed_band_std": 5.952940449895329,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 303.3333333333333,
+   "0.6": 99.0,
+   "0.7": 28.666666666666668,
+   "0.8": 5.666666666666667,
+   "0.9": 3.3333333333333335
+  },
+  "n_dup_clusters_signed_band": 34.16666666666667,
+  "n_dup_clusters_signed_band_std": 2.0816659994661326,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 22.0,
+   "0.6": 22.0,
+   "0.7": 22.0,
+   "0.8": 22.0,
+   "0.9": 18.666666666666668
+  },
+  "largest_dup_signed_band": 21.166666666666668,
+  "largest_dup_signed_band_std": 2.254624876411447,
+  "dcs_signed_mean": 0.8230559825897217,
+  "dcs_signed_std": 0.002996871513308221,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.7707199676757833,
+   "0.6": 0.8051481226801923,
+   "0.7": 0.8161208133808233,
+   "0.8": 0.819625299142596,
+   "0.9": 0.820043173372849
+  },
+  "dedup_dcs_signed_band": 0.8152343521441151,
+  "dedup_dcs_signed_band_std": 0.0029051547461622198,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.8172793785730997,
+   "0.6": 0.822190006573995,
+   "0.7": 0.8230427106221517,
+   "0.8": 0.82310817639033,
+   "0.9": 0.822900633017222
+  },
+  "matched_cos_signed_band": 0.8228103816509247,
+  "matched_cos_signed_band_std": 0.003171225425110527,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.9430338541666666,
+   "0.6": 0.979275173611111,
+   "0.7": 0.9915907118055555,
+   "0.8": 0.9957682291666666,
+   "0.9": 0.9965277777777778
+  },
+  "unique_frac_signed_band": 0.9907904730902777,
+  "unique_frac_signed_band_std": 0.00031540299355015697,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.8294813368055557,
+   "0.6": 0.828396267361111,
+   "0.7": 0.7879231770833334,
+   "0.8": 0.7125651041666666,
+   "0.9": 0.5549587673611112
+  },
+  "matched_frac_signed_band": 0.7209608289930556,
+  "matched_frac_signed_band_std": 0.006342219880356085,
+  "babel_mu1_abs_mean": 0.39035242795944214,
+  "babel_mu1_abs_std": 0.0004844975301361429,
+  "n_dup_abs_per_threshold": {
+   "0.5": 763.6666666666666,
+   "0.6": 312.6666666666667,
+   "0.7": 149.33333333333334,
+   "0.8": 78.0,
+   "0.9": 39.0
+  },
+  "n_dup_abs_band": 144.75,
+  "n_dup_abs_band_std": 5.291502622129181,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 354.0,
+   "0.6": 139.66666666666666,
+   "0.7": 61.666666666666664,
+   "0.8": 28.0,
+   "0.9": 10.0
+  },
+  "n_dup_clusters_abs_band": 59.83333333333333,
+  "n_dup_clusters_abs_band_std": 2.4281337140555777,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 22.0,
+   "0.6": 22.0,
+   "0.7": 22.0,
+   "0.8": 22.0,
+   "0.9": 18.666666666666668
+  },
+  "largest_dup_abs_band": 21.166666666666668,
+  "largest_dup_abs_band_std": 2.254624876411447,
+  "dcs_abs_mean": 0.8238364855448405,
+  "dcs_abs_std": 0.003060787325425975,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.7639355348526603,
+   "0.6": 0.8003001307790529,
+   "0.7": 0.8125468411615454,
+   "0.8": 0.817315089013492,
+   "0.9": 0.8196738977987682
+  },
+  "dedup_dcs_abs_band": 0.8124589896882146,
+  "dedup_dcs_abs_band_std": 0.003057727965492809,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.8185140689214071,
+   "0.6": 0.8234894871711731,
+   "0.7": 0.8243091106414795,
+   "0.8": 0.8240204850832621,
+   "0.9": 0.8235610922177633
+  },
+  "matched_cos_abs_band": 0.8238450437784195,
+  "matched_cos_abs_band_std": 0.0032484306342325225,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.9333224826388888,
+   "0.6": 0.9718424479166666,
+   "0.7": 0.9857313368055555,
+   "0.8": 0.9918619791666666,
+   "0.9": 0.9952799479166666
+  },
+  "unique_frac_abs_band": 0.9861789279513888,
+  "unique_frac_abs_band_std": 0.0002358037468901147,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.8207465277777778,
+   "0.6": 0.8235677083333334,
+   "0.7": 0.7857530381944445,
+   "0.8": 0.711154513888889,
+   "0.9": 0.5544704861111112
+  },
+  "matched_frac_abs_band": 0.7187364366319445,
+  "matched_frac_abs_band_std": 0.006959593990540867,
+  "alive_source": "stats",
+  "frac_denominator": "d_sae",
+  "d_sae": 6144
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/config.json b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..3ce12067a8863b91a9d79c7c4c7ea13fbc128cee
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/model.safetensors b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..6a41d8f88e6ef90cade4fc82faece21c62c56530
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1cbe998a3fb76a8a7c452025dada1ea81cd3d5a389a5839b3af29a9a8716c557
+size 37776696
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0d846a9c18af168b9ab640277af5d8fd537dd811
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bd44067a3b156c4c657c2ac2d0d08d2df65ac472c009e7ae3a381d80cb26c91
+size 50496
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..776f2107905c1d5f9fbad31025bf653768f4f3e2
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:388e47a7a6beb5f7035db467a908aef1d6c52f294c9c3e0d3b7fc088cfe887c2
+size 25895
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..163d04a860649931099d89fd86c27f963673bdec
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 32.0,
+  "l0_per_token_0.001": 32.0,
+  "l0_per_token_0.01": 32.0,
+  "l0_per_token_0.05": 31.999999755859374,
+  "l0_per_token_0.1": 31.999990380859376,
+  "l0_per_token_0.5": 31.988533056640627,
+  "l0_per_token_1.0": 31.982305419921875,
+  "l1_per_token": 305.9132832406872,
+  "r2_score": 0.9079532803360462,
+  "l2_loss": 2.158719602684836
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8119e12a10a85ed1db38aed5fd2e676e5065628c
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c5e48c5aeced80a2f92c493e70fef88fdf438e32dfe8fd4763d65cc0c39c4d32
+size 25806
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3641ba496d3313151d2277d61a94f1249c4f2f39
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bed0272f0d77451aeee91b8e8f9e380ba5fa63a81de2de072c16285d98fb9063
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..03dba2c36c1383c278446daddbdcc5c47d1bce13
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3557af1d845537868df8dafd9592a11ef07f582da6e11e5d2b6c88e0ad6685f
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/config.json b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7964558bf0dbf0fe06b1ae7d224de8b79e6126db
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/model.safetensors b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..bbfec1c7f26d2aaf0c1cfbf1d166cf225dc42641
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f7069a300a8d9130798ef4da851e6af1cd82bbab1469d94a40f5011e9a8b521
+size 37776696
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0d846a9c18af168b9ab640277af5d8fd537dd811
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bd44067a3b156c4c657c2ac2d0d08d2df65ac472c009e7ae3a381d80cb26c91
+size 50496
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e823ea3cedf0bc8c99cb354365cd47dc9d4babb9
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b079a2a6e722cc5850c1af172bb96dda33b83af50aa6af075bd9866eea8cb4b
+size 25895
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..6fcc1dc3be1039c6fd0b10da75676548fdf58a86
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 32.0,
+  "l0_per_token_0.001": 32.0,
+  "l0_per_token_0.01": 32.0,
+  "l0_per_token_0.05": 32.0,
+  "l0_per_token_0.1": 31.9999970703125,
+  "l0_per_token_0.5": 31.990776416015624,
+  "l0_per_token_1.0": 31.9842193359375,
+  "l1_per_token": 311.5163819885298,
+  "r2_score": 0.9079910218319425,
+  "l2_loss": 2.157834472749479
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f41f268f4b15482a63a28ec18a5114bcc8c34f58
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75a0d06d237272fee08df200c71d5e3a54ed731274adb91f3d94e91bb08a4a1a
+size 25806
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..95e2819b7f7ed6ade88f2cd93b2293a51325d915
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a422fce5c0bf9d0cd7ad0c1d98ca501b570217b18693e70b6817c5edd7a3405
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..aaad18038c94cba65e937ea313f6e93766edf84a
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:068d7248bb61953a780f6a71c65f51fc842c5fbff5ce7ff47d75019cb541063a
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/config.json b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..09e46fd324567f67c3c939dc55a6795e157259cf
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 32,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/model.safetensors b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..966da32d547e3b0c2b25eba4ac9412aa973eaf41
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:27b0b2429b01437f5c2c9e5a1041e6dd14acdbbb9f81f511080f202c9e7136c3
+size 37776696
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..84e55ae2fb0f9b48de06d58f20be3015f4e6d9c0
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4fa406fddb0215b32335780dc37aefd08adac672ed6f924774f401cf39538e37
+size 50496
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..edbb6acb9abc33979b1b8647b94d7b3a13af3e07
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12c929792c98c35218ff4fb6d7cb1984eb297fd1741b587236a410a69ecf2fcd
+size 25895
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..53f22fcbdd0f2782b8b839386876227b18114956
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 31.999998486328124,
+  "l0_per_token_0.001": 31.99999833984375,
+  "l0_per_token_0.01": 31.999996826171873,
+  "l0_per_token_0.05": 31.999971484375,
+  "l0_per_token_0.1": 31.999798095703124,
+  "l0_per_token_0.5": 31.988127392578125,
+  "l0_per_token_1.0": 31.983656298828127,
+  "l1_per_token": 306.22344391044953,
+  "r2_score": 0.9078798382662103,
+  "l2_loss": 2.1604419979683986
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9b54b01f1830a9e49bf625b41f4cbb4d9c673065
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d76be1fcb76f530a32af3cbacc5e90d41dcd23247e8b27ab985b1b6e1299afc1
+size 25806
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2526fe4ae0b6c21d84cd9af363130878ccea71ac
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:165ac1a2bb864e28e7a45f8a7c25b383f41586c01fad10c151dbd2edaadd169e
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8c548cc20dad7df61c40387ca523af89e30a89ef
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:425f8066a6bc2f78e4456c973093c017848ebbb3a00716d56d4f47dd14f799b9
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/code_metrics_step_final.json b/lm/gpt2-small_layer8/topk/x8/l0_64/code_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..7cad873625abf59b2e3b4bf3ddb4c1b40bf47adf
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/code_metrics_step_final.json
@@ -0,0 +1,136 @@
+{
+ "schema": 1,
+ "step": "final",
+ "n_tokens": 600000,
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "l0": 63.999995,
+   "recon_r2": 0.9281248629629932,
+   "l1_per_token": 419.5693799648451
+  },
+  "seed1": {
+   "l0": 63.999985,
+   "recon_r2": 0.9281894901783466,
+   "l1_per_token": 420.1214003578948
+  },
+  "seed2": {
+   "l0": 63.999995,
+   "recon_r2": 0.9281173386914648,
+   "l1_per_token": 416.5700457365233
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7448058340032525,
+   "mcc_abs": 0.7448540881816004,
+   "act_iou": 0.4210180881359154,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9679482290423354,
+   "matched_mcc_signed": {
+    "0.6": 0.8598297625152763,
+    "0.7": 0.8838237827914244,
+    "0.8": 0.9084769192611802,
+    "0.9": 0.9353589183670821
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8584737223480051,
+    "0.7": 0.8831324120689428,
+    "0.8": 0.9078389816038842,
+    "0.9": 0.9352761201232028
+   }
+  },
+  "seed0\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7457140571636623,
+   "mcc_abs": 0.745743508125054,
+   "act_iou": 0.41991465838405845,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9674758407827344,
+   "matched_mcc_signed": {
+    "0.6": 0.8585736770577648,
+    "0.7": 0.8824358977369756,
+    "0.8": 0.9073538421677813,
+    "0.9": 0.936153924412134
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8580881652634673,
+    "0.7": 0.8825997825901667,
+    "0.8": 0.9072116673199714,
+    "0.9": 0.9355979969843716
+   }
+  },
+  "seed1\u2194seed2": {
+   "alive_source": "stats",
+   "mcc_signed": 0.7430915967234429,
+   "mcc_abs": 0.7431443280069036,
+   "act_iou": 0.416635686403366,
+   "act_iou_n_pairs": 6144,
+   "input_r2": 0.9669713477423478,
+   "matched_mcc_signed": {
+    "0.6": 0.8571810921922965,
+    "0.7": 0.8809095106267517,
+    "0.8": 0.9052964889301378,
+    "0.9": 0.9354387034434406
+   },
+   "matched_mcc_abs": {
+    "0.6": 0.8563361294549862,
+    "0.7": 0.8806854733533145,
+    "0.8": 0.904914938947301,
+    "0.9": 0.9352389822443224
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "l0_mean": 63.999991666666666,
+  "l0_std": 5.773502689626526e-06,
+  "recon_r2_mean": 0.9281438972776016,
+  "recon_r2_std": 3.966343548243802e-05,
+  "l1_per_token_mean": 418.7536086864211,
+  "l1_per_token_std": 1.9110578211321514,
+  "input_r2_mean": 0.9674651391891391,
+  "input_r2_std": 0.0004885285678907985,
+  "mcc_signed_mean": 0.7445371626301193,
+  "mcc_signed_std": 0.0013317142787343395,
+  "mcc_abs_mean": 0.7445806414378526,
+  "mcc_abs_std": 0.0013209898420869027,
+  "matched_mcc_signed_per_threshold": {
+   "0.6": 0.8585281772551125,
+   "0.7": 0.8823897303850506,
+   "0.8": 0.9070424167863664,
+   "0.9": 0.9356505154075522
+  },
+  "matched_mcc_signed_band": 0.8959027099585204,
+  "matched_mcc_abs_per_threshold": {
+   "0.6": 0.8576326723554861,
+   "0.7": 0.882139222670808,
+   "0.8": 0.9066551959570521,
+   "0.9": 0.935371033117299
+  },
+  "matched_mcc_abs_band": 0.8954495310251613,
+  "act_iou_mean": 0.4191894776411133,
+  "act_iou_std": 0.0022794246135410677,
+  "act_iou_n_pairs": 6144.0,
+  "alive_source": "stats"
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/dict_metrics_step_final.json b/lm/gpt2-small_layer8/topk/x8/l0_64/dict_metrics_step_final.json
new file mode 100644
index 0000000000000000000000000000000000000000..b22dd6c655ad8a6a71d0b41550a94b73e3f8bdd2
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/dict_metrics_step_final.json
@@ -0,0 +1,595 @@
+{
+ "schema": 3,
+ "step": "final",
+ "linkage": "complete",
+ "thresholds": [
+  0.5,
+  0.6,
+  0.7,
+  0.8,
+  0.9
+ ],
+ "seeds": [
+  "seed0",
+  "seed1",
+  "seed2"
+ ],
+ "per_seed": {
+  "seed0": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 63.99999272460938,
+   "recon_r2": 0.9280184127316149,
+   "l1_per_token": 419.4997001473017,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 59,
+   "dense_rate": 0.009602864583333334,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 190,
+    "0.6": 60,
+    "0.7": 20,
+    "0.8": 7,
+    "0.9": 5
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 91,
+    "0.6": 27,
+    "0.7": 8,
+    "0.8": 2,
+    "0.9": 1
+   },
+   "n_dup_removed_signed": {
+    "0.5": 99,
+    "0.6": 33,
+    "0.7": 12,
+    "0.8": 5,
+    "0.9": 4
+   },
+   "largest_dup_signed": {
+    "0.5": 5,
+    "0.6": 5,
+    "0.7": 5,
+    "0.8": 5,
+    "0.9": 5
+   },
+   "babel_mu1_signed": 0.3016592562198639,
+   "n_dup_abs": {
+    "0.5": 450,
+    "0.6": 279,
+    "0.7": 212,
+    "0.8": 161,
+    "0.9": 113
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 217,
+    "0.6": 135,
+    "0.7": 103,
+    "0.8": 79,
+    "0.9": 55
+   },
+   "n_dup_removed_abs": {
+    "0.5": 233,
+    "0.6": 144,
+    "0.7": 109,
+    "0.8": 82,
+    "0.9": 58
+   },
+   "largest_dup_abs": {
+    "0.5": 5,
+    "0.6": 5,
+    "0.7": 5,
+    "0.8": 5,
+    "0.9": 5
+   },
+   "babel_mu1_abs": 0.3336229622364044,
+   "dup_excludes_dead": true
+  },
+  "seed1": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 63.999965771484376,
+   "recon_r2": 0.9280729060131212,
+   "l1_per_token": 420.0123807715552,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 58,
+   "dense_rate": 0.009440104166666666,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 186,
+    "0.6": 49,
+    "0.7": 26,
+    "0.8": 10,
+    "0.9": 6
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 89,
+    "0.6": 22,
+    "0.7": 11,
+    "0.8": 3,
+    "0.9": 1
+   },
+   "n_dup_removed_signed": {
+    "0.5": 97,
+    "0.6": 27,
+    "0.7": 15,
+    "0.8": 7,
+    "0.9": 5
+   },
+   "largest_dup_signed": {
+    "0.5": 6,
+    "0.6": 6,
+    "0.7": 6,
+    "0.8": 6,
+    "0.9": 6
+   },
+   "babel_mu1_signed": 0.3019880950450897,
+   "n_dup_abs": {
+    "0.5": 431,
+    "0.6": 262,
+    "0.7": 202,
+    "0.8": 157,
+    "0.9": 98
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 208,
+    "0.6": 126,
+    "0.7": 98,
+    "0.8": 76,
+    "0.9": 47
+   },
+   "n_dup_removed_abs": {
+    "0.5": 223,
+    "0.6": 136,
+    "0.7": 104,
+    "0.8": 81,
+    "0.9": 51
+   },
+   "largest_dup_abs": {
+    "0.5": 6,
+    "0.6": 6,
+    "0.7": 6,
+    "0.8": 6,
+    "0.9": 6
+   },
+   "babel_mu1_abs": 0.33322784304618835,
+   "dup_excludes_dead": true
+  },
+  "seed2": {
+   "d_sae": 6144,
+   "d_in": 768,
+   "l0": 64,
+   "l0_raw": 63.99998994140625,
+   "recon_r2": 0.9279888365150281,
+   "l1_per_token": 416.4976577540383,
+   "stats_source": "stats",
+   "n_dead": 0,
+   "dead_rate": 0.0,
+   "alive_source": "stats",
+   "n_dense": 65,
+   "dense_rate": 0.010579427083333334,
+   "dense_threshold": 0.1,
+   "n_dup_signed": {
+    "0.5": 193,
+    "0.6": 71,
+    "0.7": 26,
+    "0.8": 10,
+    "0.9": 8
+   },
+   "n_dup_clusters_signed": {
+    "0.5": 93,
+    "0.6": 34,
+    "0.7": 12,
+    "0.8": 4,
+    "0.9": 3
+   },
+   "n_dup_removed_signed": {
+    "0.5": 100,
+    "0.6": 37,
+    "0.7": 14,
+    "0.8": 6,
+    "0.9": 5
+   },
+   "largest_dup_signed": {
+    "0.5": 4,
+    "0.6": 4,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_signed": 0.30270087718963623,
+   "n_dup_abs": {
+    "0.5": 428,
+    "0.6": 287,
+    "0.7": 213,
+    "0.8": 158,
+    "0.9": 104
+   },
+   "n_dup_clusters_abs": {
+    "0.5": 205,
+    "0.6": 139,
+    "0.7": 104,
+    "0.8": 77,
+    "0.9": 50
+   },
+   "n_dup_removed_abs": {
+    "0.5": 223,
+    "0.6": 148,
+    "0.7": 109,
+    "0.8": 81,
+    "0.9": 54
+   },
+   "largest_dup_abs": {
+    "0.5": 4,
+    "0.6": 4,
+    "0.7": 4,
+    "0.8": 4,
+    "0.9": 4
+   },
+   "babel_mu1_abs": 0.33315521478652954,
+   "dup_excludes_dead": true
+  }
+ },
+ "per_pair": {
+  "seed0\u2194seed1": {
+   "dcs_signed": 0.7765676379203796,
+   "dcs_abs": 0.777982234954834,
+   "frac_sign_flipped": 0.016764322916666668,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.7738843560218811,
+    "0.6": 0.7761965990066528,
+    "0.7": 0.776680588722229,
+    "0.8": 0.7766603231430054,
+    "0.9": 0.7765572667121887
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9840494791666666,
+    "0.6": 0.9951171875,
+    "0.7": 0.997802734375,
+    "0.8": 0.9990234375,
+    "0.9": 0.999267578125
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7615404974785633,
+    "0.6": 0.7724065765505657,
+    "0.7": 0.7749740151630249,
+    "0.8": 0.775901865796186,
+    "0.9": 0.7759884991828585
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8294270833333334,
+    "0.6": 0.78662109375,
+    "0.7": 0.7195638020833334,
+    "0.8": 0.6222330729166666,
+    "0.9": 0.4523111979166667
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7778541445732117,
+    "0.6": 0.7792920470237732,
+    "0.7": 0.7796674370765686,
+    "0.8": 0.7789324522018433,
+    "0.9": 0.7786384224891663
+   },
+   "unique_frac_abs": {
+    "0.5": 0.962890625,
+    "0.6": 0.9772135416666666,
+    "0.7": 0.982666015625,
+    "0.8": 0.9867350260416666,
+    "0.9": 0.9911295572916666
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7489884634269401,
+    "0.6": 0.761534741264768,
+    "0.7": 0.7661526939045871,
+    "0.8": 0.7685999335080851,
+    "0.9": 0.771731554971969
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8162434895833334,
+    "0.6": 0.7763671875,
+    "0.7": 0.712890625,
+    "0.8": 0.6184895833333334,
+    "0.9": 0.4495442708333333
+   }
+  },
+  "seed0\u2194seed2": {
+   "dcs_signed": 0.7766640186309814,
+   "dcs_abs": 0.7781643867492676,
+   "frac_sign_flipped": 0.016927083333333332,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.7743046879768372,
+    "0.6": 0.7760435938835144,
+    "0.7": 0.7766110897064209,
+    "0.8": 0.7767396569252014,
+    "0.9": 0.7767677307128906
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9838053385416666,
+    "0.6": 0.9943033854166666,
+    "0.7": 0.9978841145833334,
+    "0.8": 0.9991048177083334,
+    "0.9": 0.999267578125
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7617650856894518,
+    "0.6": 0.7716227726292951,
+    "0.7": 0.7749678696272895,
+    "0.8": 0.7760443333390867,
+    "0.9": 0.7761988090351224
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8302408854166666,
+    "0.6": 0.7869466145833334,
+    "0.7": 0.72314453125,
+    "0.8": 0.6266276041666666,
+    "0.9": 0.44921875
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7777813673019409,
+    "0.6": 0.7790558338165283,
+    "0.7": 0.7793118357658386,
+    "0.8": 0.7791107892990112,
+    "0.9": 0.7786908149719238
+   },
+   "unique_frac_abs": {
+    "0.5": 0.962890625,
+    "0.6": 0.9762369791666666,
+    "0.7": 0.9822591145833334,
+    "0.8": 0.9867350260416666,
+    "0.9": 0.9908854166666666
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7489183868747205,
+    "0.6": 0.7605431138072163,
+    "0.7": 0.7654861537836648,
+    "0.8": 0.7687759049683033,
+    "0.9": 0.7715933726479609
+   },
+   "matched_frac_abs": {
+    "0.5": 0.81591796875,
+    "0.6": 0.77685546875,
+    "0.7": 0.7161458333333334,
+    "0.8": 0.6219075520833334,
+    "0.9": 0.4466145833333333
+   }
+  },
+  "seed1\u2194seed2": {
+   "dcs_signed": 0.7748768329620361,
+   "dcs_abs": 0.7761378288269043,
+   "frac_sign_flipped": 0.01904296875,
+   "alive_source": "stats",
+   "frac_denominator": "d_sae",
+   "n_alive_a": 6144,
+   "n_alive_b": 6144,
+   "matched_cos_signed": {
+    "0.5": 0.7724420428276062,
+    "0.6": 0.7748798727989197,
+    "0.7": 0.7751132249832153,
+    "0.8": 0.7751100659370422,
+    "0.9": 0.7750077843666077
+   },
+   "unique_frac_signed": {
+    "0.5": 0.9839680989583334,
+    "0.6": 0.9947916666666666,
+    "0.7": 0.9976399739583334,
+    "0.8": 0.9989420572916666,
+    "0.9": 0.9991861979166666
+   },
+   "dedup_dcs_signed": {
+    "0.5": 0.7600583284365712,
+    "0.6": 0.7708440401280919,
+    "0.7": 0.7732839375870147,
+    "0.8": 0.7742900438946283,
+    "0.9": 0.7743770814170905
+   },
+   "matched_frac_signed": {
+    "0.5": 0.8282877604166666,
+    "0.6": 0.7861328125,
+    "0.7": 0.7197265625,
+    "0.8": 0.6201171875,
+    "0.9": 0.4386393229166667
+   },
+   "matched_cos_abs": {
+    "0.5": 0.7756493091583252,
+    "0.6": 0.7775540947914124,
+    "0.7": 0.7778685092926025,
+    "0.8": 0.777348518371582,
+    "0.9": 0.7771089673042297
+   },
+   "unique_frac_abs": {
+    "0.5": 0.9637044270833334,
+    "0.6": 0.9768880208333334,
+    "0.7": 0.982666015625,
+    "0.8": 0.98681640625,
+    "0.9": 0.991455078125
+   },
+   "dedup_dcs_abs": {
+    "0.5": 0.7474966731000071,
+    "0.6": 0.759583280751637,
+    "0.7": 0.76438494870672,
+    "0.8": 0.7671002713032067,
+    "0.9": 0.7704686318902532
+   },
+   "matched_frac_abs": {
+    "0.5": 0.8134765625,
+    "0.6": 0.7762044270833334,
+    "0.7": 0.7122395833333334,
+    "0.8": 0.61669921875,
+    "0.9": 0.4371744791666667
+   }
+  }
+ },
+ "summary": {
+  "n_seeds": 3,
+  "n_pairs": 3,
+  "report_band": [
+   0.6,
+   0.7,
+   0.8,
+   0.9
+  ],
+  "n_dead_mean": 0,
+  "n_dead_std": 0.0,
+  "n_dense_mean": 61,
+  "n_dense_std": 3.7859388972001824,
+  "l0_mean": 64,
+  "l0_std": 1.482341779238718e-05,
+  "recon_r2_mean": 0.9280267184199213,
+  "recon_r2_std": 4.264573215565237e-05,
+  "l1_per_token_mean": 418.6699128909651,
+  "l1_per_token_std": 1.8986125567239265,
+  "babel_mu1_signed_mean": 0.30211607615152997,
+  "babel_mu1_signed_std": 0.0005324734114222187,
+  "n_dup_signed_per_threshold": {
+   "0.5": 189.66666666666666,
+   "0.6": 60.0,
+   "0.7": 24.0,
+   "0.8": 9.0,
+   "0.9": 6.333333333333333
+  },
+  "n_dup_signed_band": 24.833333333333332,
+  "n_dup_signed_band_std": 3.3942353090693835,
+  "n_dup_clusters_signed_per_threshold": {
+   "0.5": 91.0,
+   "0.6": 27.666666666666668,
+   "0.7": 10.333333333333334,
+   "0.8": 3.0,
+   "0.9": 1.6666666666666667
+  },
+  "n_dup_clusters_signed_band": 10.666666666666666,
+  "n_dup_clusters_signed_band_std": 2.2407216099581255,
+  "largest_dup_signed_per_threshold": {
+   "0.5": 5.0,
+   "0.6": 5.0,
+   "0.7": 5.0,
+   "0.8": 5.0,
+   "0.9": 5.0
+  },
+  "largest_dup_signed_band": 5.0,
+  "largest_dup_signed_band_std": 1.0,
+  "dcs_signed_mean": 0.7760361631711324,
+  "dcs_signed_std": 0.0010051652653352249,
+  "dedup_dcs_signed_per_threshold": {
+   "0.5": 0.7611213038681953,
+   "0.6": 0.7716244631026509,
+   "0.7": 0.7744086074591098,
+   "0.8": 0.775412081009967,
+   "0.9": 0.7755214632116904
+  },
+  "dedup_dcs_signed_band": 0.7742416536958544,
+  "dedup_dcs_signed_band_std": 0.0009048104982694627,
+  "matched_cos_signed_per_threshold": {
+   "0.5": 0.7735436956087748,
+   "0.6": 0.7757066885630289,
+   "0.7": 0.7761349678039551,
+   "0.8": 0.776170015335083,
+   "0.9": 0.7761109272638956
+  },
+  "matched_cos_signed_band": 0.7760306497414906,
+  "matched_cos_signed_band_std": 0.0008685886251900496,
+  "unique_frac_signed_per_threshold": {
+   "0.5": 0.9839409722222222,
+   "0.6": 0.9947374131944443,
+   "0.7": 0.997775607638889,
+   "0.8": 0.9990234375,
+   "0.9": 0.9992404513888888
+  },
+  "unique_frac_signed_band": 0.9976942274305555,
+  "unique_frac_signed_band_std": 9.396977037592633e-05,
+  "matched_frac_signed_per_threshold": {
+   "0.5": 0.8293185763888888,
+   "0.6": 0.7865668402777778,
+   "0.7": 0.7208116319444445,
+   "0.8": 0.6229926215277778,
+   "0.9": 0.44672309027777785
+  },
+  "matched_frac_signed_band": 0.6442735460069444,
+  "matched_frac_signed_band_std": 0.0027789682910077237,
+  "babel_mu1_abs_mean": 0.33333534002304077,
+  "babel_mu1_abs_std": 0.0002517213126463861,
+  "n_dup_abs_per_threshold": {
+   "0.5": 436.3333333333333,
+   "0.6": 276.0,
+   "0.7": 209.0,
+   "0.8": 158.66666666666666,
+   "0.9": 105.0
+  },
+  "n_dup_abs_band": 187.16666666666666,
+  "n_dup_abs_band_std": 6.433959382319205,
+  "n_dup_clusters_abs_per_threshold": {
+   "0.5": 210.0,
+   "0.6": 133.33333333333334,
+   "0.7": 101.66666666666667,
+   "0.8": 77.33333333333333,
+   "0.9": 50.666666666666664
+  },
+  "n_dup_clusters_abs_band": 90.75,
+  "n_dup_clusters_abs_band_std": 3.473110997362451,
+  "largest_dup_abs_per_threshold": {
+   "0.5": 5.0,
+   "0.6": 5.0,
+   "0.7": 5.0,
+   "0.8": 5.0,
+   "0.9": 5.0
+  },
+  "largest_dup_abs_band": 5.0,
+  "largest_dup_abs_band_std": 1.0,
+  "dcs_abs_mean": 0.777428150177002,
+  "dcs_abs_std": 0.0011211564159127054,
+  "dedup_dcs_abs_per_threshold": {
+   "0.5": 0.7484678411338893,
+   "0.6": 0.7605537119412071,
+   "0.7": 0.7653412654649907,
+   "0.8": 0.768158703259865,
+   "0.9": 0.7712645198367277
+  },
+  "dedup_dcs_abs_band": 0.7663295501256977,
+  "dedup_dcs_abs_band_std": 0.0008433105205189363,
+  "matched_cos_abs_per_threshold": {
+   "0.5": 0.7770949403444926,
+   "0.6": 0.7786339918772379,
+   "0.7": 0.7789492607116699,
+   "0.8": 0.7784639199574789,
+   "0.9": 0.7781460682551066
+  },
+  "matched_cos_abs_band": 0.7785483102003733,
+  "matched_cos_abs_band_std": 0.0009349147526076311,
+  "unique_frac_abs_per_threshold": {
+   "0.5": 0.9631618923611112,
+   "0.6": 0.9767795138888888,
+   "0.7": 0.9825303819444445,
+   "0.8": 0.9867621527777777,
+   "0.9": 0.9911566840277777
+  },
+  "unique_frac_abs_band": 0.9843071831597222,
+  "unique_frac_abs_band_std": 0.00024101231070853223,
+  "matched_frac_abs_per_threshold": {
+   "0.5": 0.8152126736111112,
+   "0.6": 0.7764756944444445,
+   "0.7": 0.7137586805555557,
+   "0.8": 0.6190321180555556,
+   "0.9": 0.4444444444444444
+  },
+  "matched_frac_abs_band": 0.638427734375,
+  "matched_frac_abs_band_std": 0.0025227864583333686,
+  "alive_source": "stats",
+  "frac_denominator": "d_sae",
+  "d_sae": 6144
+ }
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/config.json b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..1ad25b8fbad399dcd95ee6f7012618be6449b4e3
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 0,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/model.safetensors b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0ec68b9f04c7b105581275cc7c47027fd9c46d00
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e767a41a82d6cca906b427ad1ababb796897d8f6d1994d2b88bad2f04eb14a
+size 37776696
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..99d64d2b50066f9fd8d17342c9e8148b22c0dabb
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0fe91a4dcc26b644a7a00d646afb91414c13ad0478de92225f5e541fa6679bbf
+size 50496
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..688bfce14eacbfb84a227da6fcdf9db334fe9856
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b587fc901f6f10e0744fd9fe34a51c1e87427103a9d1ed0df8c4ac8cc78b1d1
+size 25895
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..0cf5beee0720a5610ad88b3b08337d07915e9ec7
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 63.99999272460938,
+  "l0_per_token_0.001": 63.99999262695312,
+  "l0_per_token_0.01": 63.99999018554688,
+  "l0_per_token_0.05": 63.99998154296875,
+  "l0_per_token_0.1": 63.9999611328125,
+  "l0_per_token_0.5": 63.969056689453126,
+  "l0_per_token_1.0": 63.946680078125,
+  "l1_per_token": 419.4997001473017,
+  "r2_score": 0.9280184127316149,
+  "l2_loss": 1.6881434257435655
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..63ee7ff43de14f30bd407c92e2eb6b066efc5d48
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e6d9a6f1beaf8bcef071af9246c6d6c392ccaddf95fb59cd7f0b1b9a6fe3c044
+size 25806
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0f62d15c0bd297f253f90739501798857ead9d7a
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a808ce191992dfd5f75d8a936d78777345069cab1c101070f71f7a0bb5b4c8b
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..20c4314adcc30c56b17f2d3e3d925685483e8ab9
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f01aeb9e1fb64bb32acdea57662e41fd9bec9983bbecd4f2f67f947e4cd9e60
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/config.json b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b86d7e5789f4f17317558623f7569acf12145128
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 1,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/model.safetensors b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9491d4e02915888a249b7dd1cfd05e48151b6d3d
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:44921ba6a04179e9a11f863b5f249e8d5a33228c3c841861bb076d736d6a3589
+size 37776696
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b0490a8eb2d76976334276707f93088a78c52b3a
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4ab33d095224bcc80b214e9e8f83c3a81bbf52a4dc43368c4a21a46cc9d8c3fd
+size 50496
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d0ec62127342d71c53ddb23202e4d2753d3f0f20
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf3161c0dd10553c59984b81f3decf2ec891cc1b0116c348374ecbe66cfe146c
+size 25895
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..3b1d33c8bd7eaf514bcf40e24f08694ed051ef52
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 63.999965771484376,
+  "l0_per_token_0.001": 63.999965283203124,
+  "l0_per_token_0.01": 63.99995947265625,
+  "l0_per_token_0.05": 63.99993168945313,
+  "l0_per_token_0.1": 63.999835302734375,
+  "l0_per_token_0.5": 63.96527041015625,
+  "l0_per_token_1.0": 63.941302294921876,
+  "l1_per_token": 420.0123807715552,
+  "r2_score": 0.9280729060131212,
+  "l2_loss": 1.6868654256547508
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a361ef60d13ad711c0d01fefa40611417d9e0619
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:945c14519ea2f605e772a2ad6f40351ec390003281b743f3b9f1bf509edbbc60
+size 25806
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a847c8ae588d1d398a2f0530c0cd16c20b875964
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd96567accb13cae0636b29a68ad5d36a064dec307716fbe8a54f410298d3013
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2d64595816afd19745850f24cc4d815c9895ef20
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff9d114f9c1aefd25887cfa43a5a6ea66261cb2f8ab3cf39e46990e28abc8c62
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/config.json b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..ff904d2425408ced117183df3820ef7f96f15324
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/config.json
@@ -0,0 +1,12 @@
+{
+    "d_in": 768,
+    "d_sae_factor": 8,
+    "architecture": "topk",
+    "activation_name": "relu",
+    "device": "cpu",
+    "seed": 2,
+    "top_k": 64,
+    "top_k_inference": 0,
+    "class_type": "TopKSAE",
+    "module_name": "consdl.models.topk"
+}
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/model.safetensors b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..2b903a7c7deae3075b96d1b4e2a2645069fc3c7c
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ffdf173124d80538e877ae5a29ad58fdd21640eb793642df6b13df66740a8327
+size 37776696
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5a68d16c80af1e58575fcc02d34add58b5dc5a15
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6fc9189132b3e39feb7b12d3fe8a60b7cdb68e06d03647ffd73d40384e6f4695
+size 50496
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1ca754ba588f12ee95cae5e7261b362507a86fdf
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09735b73dfe0c5c7e29634780823a7682439dfef78275ef2e2025edba72cbcb3
+size 25895
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
new file mode 100644
index 0000000000000000000000000000000000000000..9b34eb234f3c88eef4e099760649cce96c08f161
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/l0_per_token.json
@@ -0,0 +1,14 @@
+{
+  "n_samples": 20000,
+  "n_positions": 20480000,
+  "l0_per_token_0": 63.99998994140625,
+  "l0_per_token_0.001": 63.9999888671875,
+  "l0_per_token_0.01": 63.99998813476562,
+  "l0_per_token_0.05": 63.9999765625,
+  "l0_per_token_0.1": 63.99993173828125,
+  "l0_per_token_0.5": 63.9729158203125,
+  "l0_per_token_1.0": 63.94719580078125,
+  "l1_per_token": 416.4976577540383,
+  "r2_score": 0.9279888365150281,
+  "l2_loss": 1.6888370470417493
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2b56d327aceb4910fb7b3f4727cdcdfececeb3df
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f967353fc018cdfe090618942166ba7c7311d42abc6fba9666fcaf1068a4f4e7
+size 25806
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e60d846578d38b494346c24d032ce06f962c7754
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3bff6195e1039ec9eb22e3b84166bd33291fe86165bd2ba0efedc9ceef557faf
+size 25811
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c933ddf2467a7fe8b200254aaf4a7548573f1371
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/meta.json
@@ -0,0 +1,26 @@
+{
+  "dataset": "fineweb",
+  "split": "train",
+  "hf_dataset": "HuggingFaceFW/fineweb",
+  "hf_dataset_config": null,
+  "hf_dataset_text_field": "text",
+  "context_length": 1024,
+  "num_samples_requested": 20000,
+  "num_samples_actual": 20000,
+  "batch_seqs": 32,
+  "stats_top_k": 100,
+  "stream_seed": 0,
+  "sparsity_thresholds": [
+    0.001,
+    0.01,
+    0.05,
+    0.1,
+    0.5,
+    1.0
+  ],
+  "model_hf_name": "openai-community/gpt2",
+  "hook_target_name": "transformer.h.8",
+  "tokenizer_hf_name": "openai-community/gpt2",
+  "d_sae": 6144,
+  "d_in": 768
+}
\ No newline at end of file
diff --git a/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..941c5510dc773e308c63436938b50b702be4a85e
--- /dev/null
+++ b/lm/gpt2-small_layer8/topk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b069559ec83c957f51d0f7fddb641e2ad2319202ce4d354270456e070f2fa397
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..78d54a4c0f3fe8aac6bca689f3233356c90b2a34
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:336fb502f376a92dad8fb579931ac1377994a9bcaa42ea91c2e8c0efd6ebc536
+size 75550084
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..cc8414d7ef899627657e2bad2a03ffddff0394b7
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86582db09fe2f106d87aa63d19bb3b564e2487bba198fb15210a8995f5657e39
+size 99648
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9853aa8ef33c2818c1775b36aed346dd6e9af435
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f69bf58da1deb1b44e921a7f3f5a1aa29d19327dcc6eda94e4437db225918c6b
+size 50471
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..897c03a474bd497b24bf85a9756de5b78bf1d7d0
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:456e16b5da4304b5beba9bef08b14f142a89df7529c2a85097f4a99b2cdbdfed
+size 50382
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2753ce66340a76ca437e9066e75877053b45cae9
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3269b8a9222d9bed7a20d7d94e1a79876cc8756b5986bb2280e9470f704c3fcf
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a02929f03cf48eda0397ef406c6fc7771a3f4159
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b314c4756a9a9a9886b20dd9c35936d7f79725e00d125046c6dc7b15293b934
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a9e9b5664ab8141a8a8f4140ec843a08467e5e93
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:250c81f398e5388b9a6c05eb14ef4c154ed5c38a615abf5af5620e1bc19e4a12
+size 75550084
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a4ae2c0747553907ea8c3212b2d7a8f5639713ab
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2baffccc12ac826aad8b3a57dcfd815c69f93ea07c6fc8c8899fa2ccc76a64e3
+size 99648
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e84c3742da9531136048fb0c5854ac7ec2bdc55d
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d347e3a61660e2ac94726003a355a5d4aafdc2859dd9bcb8fbfe3cf58219a73d
+size 50471
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4f8edff78325cc97b722a41eefa831d29493726a
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7cd89d310143dd16ec525c532f7284255a7bdd6cec8e03397812b06a2d8a38d
+size 50382
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b6f76f677dae0803cf9995dcef72db160f34f19a
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6073e99b9a586bf10957ec7aa53eeb9c89caafee03abf70b830d623c6607a85
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..aace0d8283f00689c0db4fdb7c0ea82b4a5fbdf1
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd51d56585bd3c9a3ca708ae371b6cb8058b6e1b35fc7ac871bc5130c94661af
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e18848caa6b5c193edf1338d79463b563be910b8
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef4e11d2eb8516fa07c3f488540f31abbb4e9d9c0fbb4829f8d7e0a683213fb3
+size 75550084
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5b94189b48212684cbe68e33ed0458d299d6562e
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d88b33f1da23334cc25ec5863573cfecfc754285bde08c9e974b16af4e341357
+size 99648
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5d65a16f70888a0bfb6aba4e0786e93d83726b7f
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:caf7e3c7d5dea5b07434d268089f59e7d581320a26a5383e829ce6db677dbf3d
+size 50471
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..ced998d45ecdc5dee1e8e5b0109c50e126740fd6
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f962c569ff9000e897363dbe430567f972f67c032d07e184d5b0b57152babd13
+size 50382
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0b73a1002c6955e58798a8bf36998f2b81c82d5b
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ecdcb514a6c77382bdd0f2a236f8ed2d0466398a92497efbee2699ff88de3019
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..36d85b9b7ac4d048b6f3eea8e350242c7798d31d
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d1449b20918a00108b3e30550f82f0d0bf2fe177ebbf5fd57272ac17e13d845
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4bb81b3344adf1693e42a4efee79f65f6524e079
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9f505e224d5ebf9dd5bd655452e7a9599210f6be16a37af4edf4123fc7abfb8
+size 75550084
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..db71b1e59bdf0fca5a027bfb3f4726627ab46464
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f58d72c20e977ad6ab6c18ac70882274a4ca5ac2fcc375261941080b0e8c0edf
+size 99648
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4a757869a38f7326abc299c345a7c7cabdf2633c
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5fedd325f45d255222d3aa519f537244ac2a5515778bf0fc5ab614ecd90957d
+size 50471
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..174155b7d3acf71fb5d3d96d7d69f028fcf8edc9
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:617195a1969f437202e6ba785060ef8aa7def08c8351ab0ed76554e9703246c6
+size 50382
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3dff0b44199ceb228b73494b11f0a86e6597f262
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc56bc034b6fb7773cd2463ab7a368dbebf508e27b1bf40d3cac7c8fe1612ca2
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..ae42774b3fe6d39d70d3fed42303fc2dca160298
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09dd4d231dedc9796776e5b85efda357143a74ad14ae305ae2f6a3660f5cb5c7
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ea7adeecb983887409dda84f1bbc66286d82bda3
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5473e389c0e4ae086426871de10551e0d1283ec2781f368220ba800f225e3cda
+size 75550084
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..58478b20656f72a747424a3269baf8e70f62cdd8
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:940247b138cb6361f8e123ecab3b098a449ce8b5acbd2f5853339493c74722b6
+size 99648
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e344e1a57b13361e56ef8918253d062607110c62
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d361ebbfe369096bf8ac1a495fbe2b77ba9e4c00d881a85d92ba1c0e1b0ff6d
+size 50471
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..869b4cd25dd5a78b950746fbbf444c2a4471092b
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10f978071429784d6b0e58cf83fe804ef6a68088dbe062f262b97c5be04f38cc
+size 50382
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fd11b7b5a80016dccb6a5b30f4907743eb8e1db3
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:990914834fe81ae15532b8d1f4b91bc5717051c3bac373f56b2992e3beeefa08
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..64a2305d1375c546f1b15f175ee201c9352989fe
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49aa74b3337a6c71fc0d0cf996132373adefd3521ee864db122d40ff8fa62023
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..11fe2cf4fdbdc4eee1b1705351ba042d86bdc9fe
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e087f4fd3a5bf86a0bb663e6897695f016654571370b1372abb7640e5f47e02
+size 75550084
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..af47a4dacb2b9019c0f759c1abf32ef70490f83c
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7cbbb9b6fc853c80d922e07293236e5d9dfe37a03c457763607d9a0486687972
+size 99648
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..57e41e6f06644a1391d2b5e91f3eddb64f47dd7a
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:331e2e997007f6b3d0ac96993c792e077210e747d21cc4226a554d8910f3aad5
+size 50471
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..325abb7badd26638df9219686f93f42c1c017311
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a831ad86c4b80e5fbdabe29523de3f6ef2d0c615c89b4a82d1055156420444c
+size 50382
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..07f065666d03d99b55742278c4d00f7ec66cd784
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:115746a87ba7810fc2d18b715dee7398953bb6d22be93cdd21ca6dc69c3a5344
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..79f3a56c3f0cc3be75247e49129be2ed26ac8b08
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7657c7ce0cde96415e43c136dd7b9072aab7dd32578fa216189cf2fc27712afe
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f7f55e75f4d351c72dddf9eefed80b3d8476d7c0
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9eaadbe45dc38f16f1f31903020706d0c7b4e7a8952c643e74920051980d67df
+size 75550084
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a1af5c17dd52aaec8560881ef2d9c55000ede5d7
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:60239df057bcd50ba31fc8ea844e61e44a9525d984cee4fc3413974ee18a7f45
+size 99648
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..552e52e4edc4eab29920879d55e223d026ae7650
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2545f49613b29ebd1af24d4819471aaf00c243164647c3bc03e4776b0dc80fe8
+size 50471
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..33007e351e34b3bfcc92158bc6eadf689635ad56
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8881bb815bb522a1aa4fc93811fa68facd636e4f50b28c4ca6973ca6b4ffac96
+size 50382
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7e06d1857e3602beb66d4c5d4fb3b93be43868a3
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bfd938b5a7e570ec0da17c72293a17ea65482f9892e0d826b4a6d04d33debeac
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0db287ea3d999a371fa3f21dfa2d23df7892df67
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e4303da50e60a9ce0c7d6ef61bf89046e776a218147553a6587387e4dd7de3b4
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3db3a2f17b4d5971790d0c4fc27ce45adb840cb6
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8db7b45565920e2892664fc7f0a742e7cf0524a64787b80db420cd4c80500b23
+size 75550084
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a4941deaaf940b62f6a26f78b4903bda4c899eea
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4cd801ae057ed17df0012037b179083c4eb4f545d36c8d7f4df835fdaee21146
+size 99648
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2397735afc3d08276e6deaafd84a3286c8427149
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:020b9893e607ab8328e355c800981993467249080aa7e6d29af9b56d9d1b5f16
+size 50471
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1ac6bfbbe8d979a987252dd1c1eafdbb2419e4c2
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08d78163972ca6d8ee8738d1e6e03f951c134aaa96fe8e2d2bf3a304cd7ccb83
+size 50382
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a1bc793152e4e065fcbf62bf1da77198223d4d18
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a422544c43b2fd00352800cb9ef5ed59c96d9da724c3cd7486eeef10f0909c66
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c43f81d56cfd95e2c6462182732c411be2c2ef5c
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0a6ed432094ddeb3fd18a5b26d04bdecc53a264f5d9f2b4b6784b923d55194a
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..fd93e6adb0b2b4436088c8a225b6cae27d398785
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3a6a6c24cedfc9f8b409c341ba9a323bd68b93f7e2346ad39020a316a96b104
+size 75550084
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..dee862803427e45190d761a871efbf72172355ea
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00ea13d7a16342767891dcd96cf656b92b107d40a6a6533175d426d12930a0cc
+size 99648
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..643eb3c2ace66bb0dfcbca6e33f40fef67119b13
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c1c0118f7b7b240ba44ad2f9b1aa85e0975a910f7a8fa8125ae1e456b1899f18
+size 50471
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4e5edeb02da1ab46c50b551bbae59b55b66d42cd
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:07561aae66180ccfd966b5f693136089d48587f42d13c98f9f3ac2e48a80b469
+size 50382
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..476cbc71d369e9704d417d3dd1b1a79c64a448e8
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5b8bf0600cbd814b070330a308deb2568f014207fd54c0a281c85416de4402c
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c3a7fadedcbcfd3d249fb84f451147873ea094d9
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:261948314450f3e587a62b3ea8f4ebb78f5d8645d617b228c3d5f98af33ebcf0
+size 50387
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..bf37403d11e11db9922366e90fb88ac6ad56a60f
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f5782291f054ae66b037e4ea03e855c16acf2d4ef11d9f4d8186498a771ee50
+size 18890116
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..08c192ee076902fe5faa3137c8b2b97110e8c0c0
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:221d58f6624ab70f5f4f18aebe742dc0cc4e9b668dadbf95e1ff30d49430881b
+size 25920
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..11e51831797368b944369199b606a6ac4ac61293
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2d339cb67858382b5daab575f0141498828cfadabcc6fc113a857200bcb480b
+size 13607
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f208d6687d8ccb53cbff70981eefa4f3bcb59bc7
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0625c1fa7dc14609da66987fb947517c17bded07f9d30b261430993f102da22
+size 13518
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..dfe63826b88192ee40a5b6ba881bff4b8ea966ed
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b80141a9f6e96d7be0dd60f20a10b21a72400849a7e38b274f761adc463bbb6
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..6545a981ee08f918ebf3f6556f69df24c2f67a5e
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:debee4aa2c566f9e92237cf2f98595171c24c2cff906887465a1b499fd225a5e
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a2e53515717574e9337fcc163e04745c4f327aa2
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0422205959d51e8c913ba9e9a8a1a37ba291bcb91f8725a69d69aeebd92b9a5
+size 18890116
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..279f7eea72fec2a392a5b93f46c9b3086407d778
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:feb7ef89af2c15f77db2af5af3a247d9c9cad3fbfefd62c6af80654546cc2b02
+size 25920
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..611e144e3cc30c2a0ccecf10dbee34d2b3d1c7f1
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb8b32213cbcfcde159085cb715bb80294a4147adc8cc002aba020dbfa581225
+size 13607
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8b04c321053b26d37bd36988ef7ffe19db415530
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2bd8a50a7f7e76468fd364127b9f6f025432d10d1302575b3ff664716cf0a4f1
+size 13518
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..86cad538b5271ada3d9ea13b53b4d34e0042efd2
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6fa18285b3562bc69d3053ac1fca01b56af9032ab978c81ce39b163353f497b
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2d4fa88bf9e17bec809b2242eff1eb09e52cef9b
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1458c339bb55fd9969735c2d99a219d64dfb7d4dc435cfbc0d10fe48a6b22087
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7cb2f443405a96d88722fac5524583d839d5ec24
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8cfc5667b9253308d6999c920f791fd5fad40e441738c3a4aec080dcc7fd0599
+size 18890116
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..52179749082edb8ba9308e0c01b762b803597b9d
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:beb528822c5e34f4aafdae1e0661175dc8960ddd0b126657d3c5ca4f9bd1d189
+size 25920
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..6c2a381975ca5550a2f9dacde3ea57856dd704b7
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68f1230a5536bcd24b094c41d50771bbce2285f9bc19263d8d974f754a7121fc
+size 13607
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9fb27edbe2ccd863cc445ec05f5ebc1f77bbbf5e
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3bdbd2f35c7d47fd9ba574a52c344a1cf64324703a37d4c5e5eb293c72a7938
+size 13518
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..081f750317f696a69b6ec0035a79cbf3f5b0abc4
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c839407036d3e64e57bf8b737cdf5e8c118523ea591494da88989ce75f560c9
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a769906fc2a8118a90de3b2aef43d39cf2a3e169
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1de326149cfb32992c656b323fd068076d0fb117575525b423809716a0f1b4e5
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c5053b42af45a3a4bd02457cf57f8bfd2535d122
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa68328dcfe8857044f39a14ac9da8ecb7e16582b4f03b1438fef5ff138559bb
+size 18890116
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c0edc02625b8f1c63ffe83fa91d46124c6f235ed
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c92bacf67fb232785c015984367b70c9e67ed70925ed4925702d124f409c07a8
+size 25920
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5427a7173294a4cfe535eeaaa5fd5046f626ca31
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49a298034f3c4bc35853a2bd40551be4cacc315dceb1c718bc8cde07f717fb67
+size 13607
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2f5042d63d89449183d67eacf321a42277c62929
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6633e56a6b75ea57e445cb3a01a8fbb3cc5fe76047719d75d6174497d80a0869
+size 13518
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..64590718b0cfa6e26aa52fec9c2c732debd2535d
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7ed00f61882850053e80c74d2f8d58d0f208519094580d36991fee2f6d6f9be
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9bbeb030a786ccde71050076c3cd6a03937b5538
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:129a90d422c50f7a27cea78b1fad5d3dadad5ec235ec46b8c12b2f90a4c902e8
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5c9a53a6194e0960812ffd262e820635cfe3aafc
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c9ed3fe891e516d1af28d88ae1ff310fb294a682ef7dac6a3e219c8a5b50976c
+size 18890116
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..cfa37fdffe6832117a005d23cbfb6db9f2c00965
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ebbbcbcd730bf35561e11909d5f2d661ee3ff9df09a2d79fb301719953db414
+size 25920
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..081c8604353b70711e74357be3f5935ae01be414
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dadb8ddbbfc2ddc825c57299f235fbb9e7a1e6a1fdeb7c6d40589d27ec58cba1
+size 13607
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..6a80e5bade89da2fa523ac004cd81f2032b7c842
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6fc50342edfe9a6207884d7c8b3e74ce6a8dda5eca20b1bdc926a5baf433a94
+size 13518
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3b9372cb78dd2797cba98e5b049fb86b5fb8d4cf
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:27fae21caa433f3566d79de0debf4ab79d4ffe3ab817ba2cc552e1a1f91f7bce
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e2f8015a5b0e2e553946b063abd34c23daff118f
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:839194b967526eced32280f8d2f5b03ff7236217fb8b28d5f075684b79209822
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..78e4ac3dc4aecca8657f9d25a0aae6d60b0b6081
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f2212876db898eca162021c9fb687d1cc3e2a29857995cf7b09b19879c146f5
+size 18890116
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7dbf7c9c3daa74701098038b51dbbf000e2e69a3
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c83469a583e6f8010cd6f82220b80dfc4fb760baf6aca48636b68d48d86ae88
+size 25920
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e9ec6d5e4e0cf81a95e9039ad8eed25430786d5b
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3c811150ee35c03410928abb64f4b47a5a805c4cd553b14cd9c28aaab3150f6
+size 13607
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..06c1e311dd2ca984c732bb40453a4878e2038e17
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3eb5a8a14b943647ae25fbcf376bea09952f87e6123ab54f9c8aae5ae50ce52d
+size 13518
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..98dce202eb5f3c2252d58f7d3463ae7f26169373
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2902d6aa231a9331eeb989a2cb1a7fa08d57ff8510fb8daa16784faba9d7953
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5edc71316bf0e04c406763e923e3bdf030c2b043
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eccaba6ea62c328bf697bee4a7ae17e60680a6a2b93adbf359b381ed53f84540
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d8892ee28ed48308296c54d703a4918576ff12d5
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee39fa5e9ca68b3d17965bb3e78d0198a49879bc8ffd0e082f0b3914ff224343
+size 18890116
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..adb82cd82edc19912c782fabc9b2c8239ad03676
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a030be391fb4d791d4fcb1bc101458f004e0fcca8a52d68cfb8703a8988fbc95
+size 25920
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..bc116b46284af27d5eb0cb1b683e2e337c578a60
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:126e57381c5c7fb59720300e343f76015a13c72d3e6a360ebf51ab0330c680d8
+size 13607
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9a40d1eb8e30c32b6c792ffc36e064e053c58d25
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bac23636f96c4d584df00e0723bd27d85c35eacdcaa280fdc32b60335c7b55ae
+size 13518
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..403dbbbce403b8803a34255c050b5344c8148a47
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c96b6c1c9262a7bfab171058bc61f464a96507d6e57cf7ad6325960ec443ffa7
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..ac5a7829914adc98c663753708a1c99b52a2b1c0
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3755779205adcfacf7865d0be550cbbb3e0655f42d58488a225132af9aeacaa9
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8481419aceeaab74dea51d9e0c968b8316ddc6b8
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a5486d729d27ac97509b11452abf150bde153232fcc2be48bff9f0b6aeb12e4
+size 18890116
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..63bb10c84d081b36a4fda90c6024fb387e1bf4fb
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7ebbe1878c8c2dce73c8df1c64871035e7f2c74553194f504d66b3d111e8c71
+size 25920
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fa73bf0f61d64c92a68c76af707cc4308c51de12
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1e5a1fff6b6553f3c60ec5b57b2978b766b3d111925e679c255ebe5fd6ca599
+size 13607
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..bbb560961710031fda3818bad0a37b43bd3a0c5a
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3baf53c95a4d4e269001a1c394cc7580baf7922ed8a86f24166694dd7432301e
+size 13518
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..dbec4e2bd21b124fb8d31b172d75e572d6411776
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5bd0b6376dfc13096d957b717306b6f32aa5536417443213099e020a160a7dcd
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4c25d2e92b2e56cc9891537ee0b4f207f618dfdf
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8408a606b6e93929957be7df4fbd34a99c663e5164dac037ab6a5b41fbe1918d
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..34f7c87681a25b19ac12c80fba336660217152a6
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6db34595f211db4865692e883d4a3d8a25dbec40fbe6a45af6558a0524a2f5f
+size 18890116
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a16cca9f9ab159363ad1a7101ac3873ca4684ad5
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a89a8f915efa89d50edc0cd532acf968b6d259c0030bef5187db2f26cfcb5ea3
+size 25920
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1ddc0613bf525829d938e3a3fc8ee6fd8dad92e2
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56bf9da6e7080809ad6632189676c76801f271be1b94b254787653b83267e3fa
+size 13607
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..503a4050525ffeb77361ffdfa9912b90b81e71cd
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:241ff3e93315a49bfc75bfcc2987d851dec4527ad529fe5d23e67fcdd38c048c
+size 13518
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..70b4bfb992ca49bba5f7a98d6e3d83515ff4d36e
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a1b9bcabec2f4c80867ef48eebdec4fcfe635f76118b18594da7c27fca3d9da
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..ee830140c2133c86b9689efc3bc08117619a7e66
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8441c5a509338ca1c30579492057aba8d0fbac0bdcde3d703d905e9423d57e04
+size 13523
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b2bd9e759a212325af5fc1c6f00576f5276e63e2
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d67a084c71b305ceb8c804d1b6bd4d2eafafcd03539b38fde059f71445ec7d9
+size 37776772
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..05f0c89eb43f6ec6a4893147ff3fd49267040ff9
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:287d2daa4e9f3bd9196e48b238d400995de5f6052ccde099b4bbbba5bcfc4fbf
+size 50496
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e40a7886d96dcaa43aa73269db4512ee1d463d87
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd052216fbdf27609598e44100395e8659da50716f1f53e7157f0ec9ef4c6796
+size 25895
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3973ab978dd1eaed91a54c3beec5feecd099e3fa
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f75298e485aab5393723d65c8fb750874838752e6e82147704e76c450ddd86b9
+size 25806
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7d0f48867559524f828486cea676dac1200985e7
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dfb2306ba39e29f4cb93afbe6423abb9d80cfb5eb0ad61330ea91f061d613298
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..84d364db631768b95e29a24b22e844b815f8c4e8
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eef0810a0a8ec222188bbe2a9191f710d8e12f92d71747c5bfda6fd4a876abfd
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a80acce2f586530d1e18114b115cc678b8b69615
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c92522ef608ec8c7e0b5e43e6a6601ca5bd9522606dc2c4da49c39cd8c65850
+size 37776772
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4d8934d6ad4f8beaa0a0166908e2a37d0d309bd8
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:163536de07e7922e262b4905f57d04d951382e9bf4310fcf76003b2db35ca162
+size 50496
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7c77d395aced6e4fa0faafafd73abe61018fed85
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4ced29bbe67c00d85ca3a2dd7d47822046cb974d1884f2fdfe5c4a142cb87c0
+size 25895
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f146a641d85cabbad8b328e55b4037f861ffa52e
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b26662cfabeef3df7a54e7714e64f8c2d073a61839e50b28e057497895e1c25f
+size 25806
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..41e04622ce99f90bd1f1b9f11fe1b5eb7cb00d17
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ccaab7867c30b95b6fafa086bddbf98b4839733e5fe14005ad636d4c64a326c6
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..06fd4c975f0f512d03e7ae55bcba50e9431c9795
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c674dcc8d476a999dee3ca905ac145171351d17fc67dd154cd08c4cbc7919a09
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..63c7bd592b71c173a0639e64adbb3e3e37a4279c
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:954bc8c23f1fb56f6417fbc9132958759267421bb8c9123fb36624d4d99a0c6a
+size 37776772
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..67dde5c600f282c6a094ee2f50235498a6511131
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a707d0eaa299b383665aa7023a429a987c6b0d0d2e7ad4bef20a06db833b7303
+size 50496
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fbccf55ef3bd37c1edc8ac19814730418ec702ef
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4718dda8127d1f3c8be70bfed3912b7a4e8a91ea5d3353f9faeb2f8a7d0669cb
+size 25895
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2557530dbb03925dd4e3b3cc95cbc652775c0617
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46f78e969479dee583f23ea9090fad838691a603655eb8c9f7628de32ce0c22d
+size 25806
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3d21210116778d97380cedc267cdb0b9cec5ec0e
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1cf16e690eeffb43383faf73f756188bbebff6588262d0db28fa1d7b96ca6295
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..170b5987a28fb4e7cab2960bf71ba734be9344eb
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9f31e54ec0920367b706db3a8e1445f3ee46711a3d016d5daf918494d1df671
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..35d32f0afa4ba6108075a086faf2930b30261d60
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:577d9454462418df632a4cab79b9fb3a3bb5b86cfefa8a0a3ce677ef0580d3c4
+size 37776772
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c1701914e1ace47a7152137261aeccaf76f9fd6c
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc8e4564f8f7380e3e854ed91fbc151072708d2e7a50085fa82955263aae7470
+size 50496
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8b5c9472e245656ae06d84945bd104fdda7cc3d3
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b92f0c920b3caef6e2c2a1fdfe50a2b3f778e016bb695a1f29aa5d0f37c9ca6
+size 25895
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a78c87f652bcb11ea5be4e43bd2539d5029ec8d7
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e53b98b4dcf4cb9e55d6a3d6f63b08ddaf7fdc882518706a1e079c1f7d510796
+size 25806
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5e14f88f00789679b69e2d70bfb29498728780e6
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ddc28da4e17129129e02f3d15c688f1fa0b8883c80ead512a0792aa3f0b24e0c
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..efacbcb059263c6b60e16faa415792f53ed65a76
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:58dec78d2a3a59e1bac9267dfe535c2565b2cc06fec24a2d88b9dc4b4fdb11a9
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0a09a626222d99ff9a01c16fd4976b3a7699548d
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba0df4366de4e66d99128d297c258c0bf364cfc46d633630126016ec2f847497
+size 37776772
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..451cee0df6b16fc03c62ca86b9725d96da374a9c
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5922e70e2b102fbc26489cdb9b281378328860c35dd59703e23dfbc0880bc0af
+size 50496
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..357f1ccf08d080b1a35d2bab745cf405c71b5191
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:378b6cc250369ed6412661eb1cdc86c94eefe050f4815d89850b4636cc145b67
+size 25895
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9323c6b4efc5b7d8a8fbd016092fbb8649d11bfd
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80a7ea91566bccdd1aed0be8e66d222babb369a2374b9365ca098b08fce6dd3e
+size 25806
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a7a851a4825ae50a53d54773cf04f652864f702f
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:705cf32c5ffed1fa029fadccc31144c3b065c65ba4d1a9cd3ae101fe7591aace
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1a6753d13b9d07855387c8b19f3db473d228e5a7
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2eb530a312a7bc635d5fb7e1bdcb27a0831c3162bb1874dafa056ca9f8b54a5b
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..965765416d3471367819e1389b48d98b00fc77eb
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e63cfecdbae0eec01d03a236ed7134afff2b1d592ca7eb70805d447db79ee07
+size 37776772
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2bb7e1919b5761894788a3de682d079eea64a238
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6baf27c05a02d9926e421d4e925aa2491300440e9515ba3e41b1083efbc0262c
+size 50496
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a69af564047654f82604a9d24c279fe89716330a
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e30090006b8144ecef74e54aa64e7f6e3aa60c39a8c9605b6d4ca7d40d5b65a
+size 25895
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..70050c5a1c1f3d08b0629079a2e963130b986218
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f78d1bf04ff339d324d6220c0d4133bf37fe26461d300bb854f084caea132dc1
+size 25806
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..06e43a75979b4ec4ecd3097efb7395e2ac1937e8
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c6dcf650040fa02d7415d79280f227fef5e062317cdb18f827eadf2ce427dde
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..ec9a90d5e175ab644bef3767994777f39e0010e4
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6930c0b42152f23ab69ae79ac2f490841390b8fe5af1c844c56cac2bbe8576f0
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5c14b283bd9a0f906bb14a19169a4d75978f7e06
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:879335407787b46ad5ef2261f4e3d70b689909d37bab63bd791f2cdd996abe4e
+size 37776772
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f40b6f26582395f945dbd68cbdf39ab6995a9e02
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f76d236ed285400eccc7992f34d6869c3b5b1ac4bd7d63a99b1ceae406efaac
+size 50496
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..dfe2afb3952c8ab8cdcd5e53e0986e576f869646
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7dc688e4e5e4f666f8d0c4c9fa5c0aa59326f6fd5f16013005200f7e0f159480
+size 25895
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..bff49005e16a6e82d95d7844b6b545094536d73c
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02997030e0a93e3e674e7eb1c71f83b199967cc58ac937361a76650ce244cbb4
+size 25806
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..22e9326643311fd84b1ff86642c077cde054610c
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f45dc1b920e221769c6a7c4cc8505b3707c160b4a9d4a8f047ba7b267ccedaa
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e1f435800b0776f736aba804fabe3a79fb7754aa
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bdcd7cfd4eec01f84e170816e8bcd1fe4424e008d6db7e3ab5d8421194d2966f
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..48aa6af253ebdb92aa249302b6a08e8c28a7d3a5
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1ad441f5dba17c0f742d7c5b1e632a7594942d723c4c1c104aabc81be73440e
+size 37776772
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d0f9dfbcd8cc934aac76f212495b4710986d9100
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87866010fc4d7ecf663a863eb6267346f85f8156c895fe074405992d2f8af607
+size 50496
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..bd2f17c0bc5698bc98380a365d97403fc54ef999
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e262c158adbefaa51d9454582617e6fa99f4ebb6de8e3a51654f046fe6e5009
+size 25895
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d50466dc360dd44d389eae12a33446184ac0506e
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9cb16de06dcdc1064777294059985cad82bbdf1f5f9c66c65330638d74391f27
+size 25806
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..81728efc47eac806ebadb16e03e7a55f8c10a324
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40b4b98c670c1e52e8e1b5f84098504635ff69d9d47f8c4cbed56d14d34b5dd4
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..df1f681edc8dfddd9fc3b1d94327f39725b1a182
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd04ea3497508245a532660199aeb94f01f05b787821ace6500c277c1e54d828
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/model.safetensors b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0e0e65f6601030547e3c58a5f0313288049af12a
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2de741ab1e750f939c59bf324a563b40e34a124a27e232fccb467ec637c23c4e
+size 37776772
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3d662814a86043f145d9f8ae83a79e03e1a42513
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0bbf69a4d8bc51d6c8e1146778c6bf06a88bdbd7a5d000ab8f557f3b0c749246
+size 50496
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8328dec230df0278c6d9b74701fd89cd180d836d
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:50f2bfeb467d90492e88039f68ce07bf6cbe02712571316501edc3c4fb06dbff
+size 25895
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..13c1529e644c14e998c34732e7c4be46c1ed9cc4
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1781fd1b5ef69e558800d9322aa8e689857188d950366783bfcff7418a78664d
+size 25806
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..61183d11e8962a560472a762983b7536c88bf434
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ea52cacafdfe034489e96166c3eef310015381383c12389a5288d22a270fed1
+size 25811
diff --git a/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4fe29fa81a1d3f54c5adc849527b3e65ecc4b4d7
--- /dev/null
+++ b/lm/pythia-160m_layer8/batchtopk/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a0626d483f07b09afc4dcf7cc3e57ebd477ac8e362a6e1481acde941586568d
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3a894fee0576ed540c074eef03c647191a6dca2e
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:181d188e32c929156b190682a27ad02a72a99b5f911e6dd8e7f8bd7456656d11
+size 75550084
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c7e96ab58e309de6f0f776a5257c253699a02c90
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d5f2ba8b6378e609a1d093626a81a990161e70e8ec2ba79ea1281dde0b80961
+size 99648
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8f5e499fe862c3617da75fa10c61ba56ab089b0b
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f0e3f90147b15aeaba28670b4d4c6a0bcdf41905986777ef1cfa83a38afa845
+size 50471
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..cef80bdc082e8eaebdff4dcb6f60b2f5cd42dd27
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14b90674568ea11ad593b645b1a25cd4665dce1c4a2acfa7b93b44fb66067f4d
+size 50382
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8624bfcec70830b843c2451e3b5d49f2590a9e69
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f60ec3463cb7e3eb063e8f0b1c66dac95494d2c8066cdb0e51cfc5d92d38195
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9d596421c6955b5f5cb7b40bdf337c500da566c7
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3bc1ba06ff394386d7560e0f76c1ac244cebdc6010807d9103b4b4e9dd914f40
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e380847894a15dddcc262299659feb9f7cd82c3f
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a3908cc9836d9b6c6a653c6f421a056b762015a00b8f94e249b000895a4411d
+size 75550084
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4bfab344cc084ccf4289b36541380ec3a9a09190
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5fa03c6cd74a21d4993c6bead0a3bf73f5749f3cfad8eb0f9c1bb435c96958a4
+size 99648
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f712ae3b96f6d3d699761b0b34b2c84e6184969e
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:861f5d040072846124dc722a8b305396b5bd72614a61f4d8e887646e0df73720
+size 50471
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b7cd1bf4ff278db9d335e78cf76abf6876962468
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83968bb8240ec8ce404772197304253ee2420d299158be1eae3e5aa15a38e0fd
+size 50382
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7adac54275c79ea37668ef80695c1bec4c2502f4
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9283bd5bca6e614968bcd291e2a8ecd0632c61d756225293972968d78125df36
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..da47210d9a0149ea530625c25bfe71df06cb60ce
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21a2df44922fbdef866c5a01dd49936049059e2624d770f15df2fdec98621cf1
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e80aa45d6430eceb05a4850d1767a3562456b240
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76881c623866c8dfdd4b0a73801864c24c744b13aba31c9344568e392d019497
+size 75550084
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d73537ec769f8fd36252e2b5cabd50379945b6b6
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:194ddd3b6306e95a18bc817419c0c1ebc57f5eeac7584aacdb847da021e88bbd
+size 99648
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a873ccc1f9339128a8caea52c097b53179087486
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f359e87a62ffc748b18756a210895844c0d632049439bcf8049c12f6db8d17ef
+size 50471
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..18839ebbf0beece8ab3473edd9f2004acef22cb7
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:843078d201877e5d17a7dface8aaa4dc2dc35062647f440ec3f2d70985dd7aa3
+size 50382
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..da59074b042cdcbd1a33fc71e93134fb35375afc
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b35678fb605c23a10b1f2164f386bba76866770dcc6f1dede764e6f07d29e359
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3fe6dcf4c16ca0ef1955a597df6b792e5842a03e
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8ace24806a70f68b541946cb8f17ce8e83133eea33a6f5fe65b60dbc6ef25b8
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1126568eb92123808c1feafdc3ffef9133d73e65
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce5f9662bed4cd785092f591a616a584289ada57b69d3774c1bb023ce6571331
+size 75550084
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e8f42f27de68cca0524cf675ceb016b41ffb3fa4
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8bbe055dc63f27954ea7026ec83dbc18313bd411b4c07499fbe32cf90d69b1a1
+size 99648
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d22a1f5a1b0f18c891ae4da9db320f8c48885778
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b1ed1f7e8b53d5a34f69c518679c15d7b67869429ad3a84772ea74f33a32f31
+size 50471
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..eac6bf59e708abb7e4743c1a96ffa450885a3ddb
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94e4deb87526cbf44fea1cdae8fa05ba1599833f23a8ad8c3acf62feddd7e7a5
+size 50382
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d6018807622000a8fb05b5388a59c9c0a9cd272f
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:817db733c3e486844d17dc05716c91646a355ff52b2fe182a28b957043f5e2a6
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fee498ae4310639f87a14fc7fde5524246bb066d
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5cd994217a966e159c736a69f2b0e582c12814938c9e65343484cda778e0a9f
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8965f7a0f776f82db37bf2a49c42785e4891c646
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1302d52903c512140e1f5c40394f89d0983761e9859111903607b4d4b355b26d
+size 75550084
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..ff14e4ec5f265559196b4e4164507bcfb86315b0
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9f0fad56afc71aa1e63cbde0501e42859a5dbc0b098617da1b313ea45337506
+size 99648
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f0e16bcfe8889f6d1986f1923b87957bccafc1e2
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc98e9c7be6bba4feac43576d171fce374b78459a325e25856361af8036f3c33
+size 50471
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3e35c9a911c326c12a4bca84b2c5e1a68c48fac4
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94b05420b0e35162b666b0edf183df924504268ae0b994090360f6ebb021fed3
+size 50382
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f8860b02d6a0ad58f38bf9059343a9180fe0e2d1
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42ce25371e1ccf1e14c6343a04a83350726be13d397abc59cc7ff2e95f56dde5
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c4ad9b2280558b4903470920e8c923f5d2d9fc0c
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe966fa845e6f3db479522a0365bea5bb776cf40d3e33d1bd226f51370904383
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3b64bcf8943991bca72b9ccff9930fe0885d3880
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:487541a9aa3d3989a0eb90db847eced3d1e3e99acfe0e90f5da5ada2ffb97487
+size 75550084
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d05c629dee499ff21902a6d411ae39713a03c44b
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a7ecc63f545b9f9f10a2a49d75b6f0c2308d0a1408a83888f7e3f358957941b
+size 99648
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9a4a5731a7a3bfcd4646d015a2bd7d3bd42f059f
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9edfc5154a87d4c64f89fb84f2f57423d08ed5d1e0f674c43c7422169e7d017d
+size 50471
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a348da340c9573a87077d703c7e3c40192884993
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af4505fc118a9317ff56d1650b6868d69dca7fcbedd5c439a8595c541119d4ec
+size 50382
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1de4367d9a91e31a832f0047374338cab5738718
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b57d9fca72b21a1cfdb691c3b7681a2f32b1531eb539a80b5ee6186c8e903dc
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7d96487fc94857dc0f60cbfbe0870ed500b350e3
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b0c8ef5e5a7daa2dad9f2e19cfca825afbb2d73933aac44e93bba9b4a87bf72
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e5d7dbfb40ec38996508404f4844c2bd17d365ef
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:62d5f429b718b168772d037856a39e50bad8ef3ebc606c507f3d7b92cae8072e
+size 75550084
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c7878a9c50d0144ea1e6062a618da3218fd39092
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28bdad413e70b2d857a10cd05d474a243afffcf9dcd480b4517ef6338c6f041e
+size 99648
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..25260054bb382a4e7d48d0f27972cb3362456261
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f9f3b8bf4b7e5dd2e27079c6aa71583fe2c04b72c0b8993cf1fa4eb2fa9fdc3
+size 50471
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2189cbd5d785b14e82b1fd28e11900b01eee2257
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:136e34fc2164b32e659a88d8fc0b8d5972b32500fa1db662e269b6ee887df5a4
+size 50382
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a93b5d6fbde67180909606fa6e782d8ed8751e2c
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff8622bdb5fa6abc5c57fbebc62ae21182de146c18600d1b85cc7500b1abb596
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fa657c7f4eca4355447ac1987d04c65cf3d390fb
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:370f0b3886042df9d86aafd055fc6705e8ca2f9130ffaa3f463f08e5e02e35f2
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..816fe2936b8da261ac55f21ca3d3c577f74edde4
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d87a81dfcc9c3fb2872abf86192cefed54265f0c32e32d30d22a6f99fbce2bfe
+size 75550084
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d26698d08851de6a40220d27eacfc4e2cd4c23fc
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c46f1d4be33408f29fc90c4c22925fc3671c7ff3c9b8d4f8cc05bad165195f74
+size 99648
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..77fb1b4922c64d40176cbeb6a35aabbf95f8ec10
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad123568fe13dda0d14daa9bba50e882d2a3593e1bfabf2d371a75b5d9ab3b57
+size 50471
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e186d2a5ad09334551d39e6cb796c64c44fbb1c5
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c325a441dac0192f637e0b1f467779f74a03a3a45a4a620cc3b851cd9d5ebb9
+size 50382
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1c6109f8ffa099f26e61e4f2ffac923e2863d495
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb1c11de61aba4656a355319ee72752b8c11f2bb483ec6ab35f56ec05809c680
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4b20214e83c05a72fd4e76cc6f1f49d5e0604077
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3ce837f2c372a5080517f772e87609279a365492184b04f952e641279b63c1f
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b443aa4347406329362de860bd13c0c4259e7ecf
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3ec5ca7694f8e86806c99bd76597abd05a076b3b6d84b0996619ef1dfdf5d8e
+size 75550084
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..519c6feabd4d56b8d03a060f37a38d8bce1f3f8e
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ddf9821da9672dab4703845994845418c349fccc76e33ed5b0c094d6a454642
+size 99648
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..6106e8b7ad2121fe4afcba864f5b95aca5a04226
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2125fe0bf158d4f7953446936b090dcca296ef2f1bd35e8d3aa86b7bbfaace01
+size 50471
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b6e8219b3ac5c707b37fcf9018adfb36915dcba6
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:17960f5b6a09ed1f1f64287d6797d1e3bcf6387fd8f5b05473ce94095d1c8cff
+size 50382
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5288543cf5d30b8f250582471d34fa8ae39e0504
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b90270dcafb36fe85f6c55e15fa3d9d78d588212e75b245da98b30bba05b0a9
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5b64ac5059fd4ee7eb0c382320eb6b6e920a9a8f
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x16/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c681092d6160811728014495156eaf90fab7e8831708c2c6a0e07bdeb314dc1b
+size 50387
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8172bdd50a13be2205a0d97b6d1061820e726724
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f1765122f42a5713c2fee48ba37284d5033babe434b6136df45053f80898436
+size 18890116
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0cc2ca8e95947c231b39bdc7e02beb88277f6dbe
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:22e3be92374681bd2b35a69f3cda0b7c2353b07cbce81427e2aa69ca7c423057
+size 25920
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..05eda085c4c6bdbbafde58dbc1d545f7867d5ef0
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ae79996fdce9877435643035375b327829ca5562678269c5e978971b248317b
+size 13607
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c6d1806021254b0b905ca864e748185e80ac7088
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c534ec87c51d66f9b899dc75bff9624310d151564de86feb661f1c15e3a7eef6
+size 13518
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..516dec8faf42b3b52c7bed73db04cc1344ae7f83
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7dfd9fb4efe5ec9c6945dd52f325508bbd0ab2f29fc0d562444dc746c39fdf6f
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..07f7e6771d8e2669176ff8ebd863cdb3b7180b17
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2c797d2f1f48909e3977a43a5f526811f181ef95822996b26498385d3321b23
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..10e03a9852cb1fac82b67e0b136376f70f739664
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ecf44745e4819deac4ad7a0bcbfb9ebbfa1b4bd6676ba3f596e36be64eb2d49
+size 18890116
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1d7e53702ed80b9c680bf931a1471fcf4ad281e4
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac2d12989cfcaa8623e93099ea2f1c0df223945d08e2579a277bcaacb0783316
+size 25920
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c02d24c47e29d7581096b888b162a143d4a6eaf3
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6ab9448b5a0ac3d70691d626c997e60e6be441cd63a9714cee24cefbb3388d3
+size 13607
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..13fa95e100da93f08305d704885752ba9885d07a
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12d8c5b814d6a27a8cfdcd82efbdcab852c81712d7fb99c70b0caa3ce2d63ad9
+size 13518
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..56f470b9703241fa21d8d84675f59bfd6133d01b
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37b63891ecfefa8dabbdbf4e278e41011c6d2dd2921e85e1682414ec100027eb
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..cb84b482961c50663f285b2b2362ef6b35c9a4f3
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8dd6e222443f6eba2a4a2d55b78b89972d1910da3f882a49448e622b0f9573c
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1d4d7a15fc39862f4420675a4a19b93714053372
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a146064d848161c51ce77efa1778b42f3f4efd3faeef7334d27fcc3fb846802
+size 18890116
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a2e3ed5b2e0cc4128c43aefdbec501874e66156a
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f64ed8b3164a5b6b93a0ab47babb95fe70d5e4a2c97208ba9f982a218e4c317
+size 25920
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0480eabd58b9ec7ae0cb606a6500f152a799ae89
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb414e98dca71d9ba9c54e56373cbcbf50ed86d08bd55e8dc0a27e18a3acc758
+size 13607
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..62e2307fc1209951b8aa2dd4ef711831e45d9c13
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23f5ab367d2a49a6f6867baca1a89fe9efe4990490a6df8189ee433612b14165
+size 13518
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fe3b892f2fdffa7a78b1640506a00daf3006fbb9
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3360dacc1981518643751f9a6bb93360aa8c67adb0f1dd0de575d9008dd3c060
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..6f8c659bffe7ebd950c0ae5bff87f6da64c24692
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc74f3ef91ef8f5dc1132705188c53db59c5027f4de7a728f68326249689f825
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f21a5fe0f79fa21ba5b4f32522cb6d6d6351db4a
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ddd2d4247b3fefb70ba7a71a755e1c6c5fce0df79dd4cbf887b67b2de122fd9
+size 18890116
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..6226b1b0ad204267a81d2f4416d60210b29713d0
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24857c3abe671087f390221d193e21655151b467915558a44a499fa423512b36
+size 25920
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0bf7c142a4af39fb645d116038a2e836936d65ec
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75f0df55ad70ddeba479a0135e8d7422de2e19fbf344b6acd86d1970aea648b7
+size 13607
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..deef7d12ab7c419be3aa6396db15275aeee0ae58
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ca64410b18d9d81e3792151ab3768b5ba3510fadfd0b3cefdef4058c8e80abc
+size 13518
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fe7e927d9701f4532a95c7d7fd1ea8a5015bedc9
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a55b9c56246079293f4f9b54cf02968d1c461e93fdc9086ce94ae4d2e4e25922
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0fecc3fbb4cde5911422e54f46f4e8978de6c322
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0215cdb66a3eca9b3d39c36a3da4c20625d8dd13ecc6ca9a190a630240d60b1
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a070e2ef70d8435da1d61956d5a481635e1a4c98
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c47f30b9a408e13159e9fd0355f9010e3473b87538afc872a6605891b63bb4f4
+size 18890116
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..870e91c72c82155def32845919b3b278b86c4b56
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5fd58f8099c5db7c30d3b1374707745015392dfe6724ef58935d4820c6f425f
+size 25920
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0cbae46f481f03a71ce51defd3855b0701b3b469
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f52ed0d44384f54e5be59077d672a8eb530d0c95b44ba88e13a684e06fff5d4
+size 13607
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..382cdc3e9cada84a67a4b74489f5a7e9fe100f22
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b195bbc1c5240377359a3a8e4b1311f4a3e080dfdc01b958fb173cb08a537bc3
+size 13518
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..88e4b829ff6c4c1de6035cc9156cf4dbe2398f12
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53f0692de0d40f3f1ba2eefb6f3de5f19f7c4d767e60fe75f53f87358741adc9
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..550dcccc89aa534621fbb31b3a287fdf791a4ca5
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43199ec03b87d216dd8ee2b4aa036eb69cdead5468b2d3de9583679b279e9969
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..2d14064dc54ea9cbc9da405c51576354f911e040
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:444f1581458bec43c888c96ca0d72d90f958cccc78ebf960e8ea9f30bfd63352
+size 18890116
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9984197723442a2d52596e1450b198aa075ecaf9
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb9738cc5a7e80a158799ab1cc43d8b83439ca99aba08031736e9cb04d97da4e
+size 25920
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..ce676b6f145de6103fc047fe58e4af9a6f6b751d
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ddf2cd7d4cef30613dc21af7acb7df5442a2d7a9019dd125f5e4e2b6d21cd0d7
+size 13607
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9c7511d9377cb2aa9f1e94ac6f312fac398804df
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:980ad81fda1da447a801c41c3941baa45002de05c8290e6719267057cb33c1fb
+size 13518
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5f8bccd7e39a826c73403b9997c68db3e5a2e0fa
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c25fd0a720a7f7ab70db74437ce30c7f41487e421a0d4006447ae3862640509
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..201cd1d2322112eb9ad124294d93fc9da2997845
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb9a48d62dc24be2dc36396c3caf6634246ab5aee61710eda83ca51e6b1bd525
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7433241b4a15de253b699e5020fa6fbb2d88ebbe
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:796b8e91d9fb25aefcdff0f3b315d864a59c6ab0929e64035a0c2aef759e08e4
+size 18890116
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1add1675e5c4a0d37b1c0c51c5579c72472cf9f0
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da50eaebf453c272d217dd8624f0d7edcefdae7c61d1653d00782c6d3cabb5e4
+size 25920
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4449bde241b7ece98132680e9ea6b109e0e8d25e
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3055abbcba1d05420331f4de739022d2d4eebdc9e4d7b527c1100919136bb30c
+size 13607
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c5e90f3676705c491dcadb825db1b287168b5b37
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2cb9bf0a2654bd766e35dbc9b5dc3c43b8f408412554c76f77e2da820992fa83
+size 13518
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..ba54adb335b87ba5d5e092b546fb17c882bb4657
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1f4060eb13949528d43348381dc3a69eb86dfda3452e41dbf27215da4b05490
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0afe31cf561e8926124cb4853acba0dc083c2bb2
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f1b471c961045b01f62432565b09a1d762be21a8194d8e4ba28ca63c9344f2d
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1f916822771fc4050c836159d53b8f7ec27794fc
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31342c528dcf2840d0cad79edd6e792faa6961ebe288a4949d8062cf249e5ce9
+size 18890116
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..18094c99508fcf001a15b7ce02c4e6f2f66f1190
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51c301d523ad4003185a9d80eae14eea9eaa004695e77358d0518afd288d58c1
+size 25920
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a823a8a93f6839e16ed2312009257eafeb06d3a9
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5fd9c62457c5d7206337149e1b2a882e1aa87e5d73bf51e9adebc69a97f935a6
+size 13607
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b7a8132ecd5c72d1de83fddbde2a6be618495662
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d86a46e4641a6a22ffcc5b8911aaf5417f95d5a83f573f3bbe68a47101badde
+size 13518
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..272068a021dcea279819e92497002224a5ff2680
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5292d4fc9684c4f716108a5d4322e12bba47d2109f8146c1ea1688ef0fedd053
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f8a9707428aba67aa08cc28ccc8004b3a483ece3
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad0f84a105ee5c8fac3a5c6bb50a950948bb58121818e0b03455f8d45bfc19c9
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9aceffa4d25dbc0ce265be76b2f28ea9317ef673
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:64c7e483bfb52d3d35ad1517ad13accd14e682af7e98500a935015f7cadb0bdc
+size 18890116
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..22ef05c1764a980f1573d50e01f0036fdcd34c20
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23aee838e7e15c33f037da54f4d478ad5c601aaa62016f140c6465b612a2acf3
+size 25920
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c20214db643684e9c51c911caeb23587c60f28e7
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e8114cf705b4f48b6859d3e4155dbf596bf2bb77a31c562a567a0c5522e7db66
+size 13607
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..765231244614315d0660e64d2a1e711c5ea61c30
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:59ad168745f9a71d5e5d684e7264ce3ab4f758f6b1c6aec4b948f848c7f2b267
+size 13518
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0c571e6a09ce414afd002b5308222c9dcdfef259
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28f8a640cf8f84f4264208a2fef4003d299767337818215e77a07182776266e2
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..26178fbabe7359e3b9ab428a458ee9ef9f42cfff
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x4/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:662104f853b51338f67a5f75f019c06d43b63e2bcf2b8e2ea4c83d8e0c62a274
+size 13523
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..62e3be8d98df1de2660773ad11d861274ac2ee40
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63c60ec62b580158f7bd335cf045c7ef4fedeefff878a8445b9509311ced082a
+size 37776772
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4b8ba6c35ba73c09ce82f913d50bf15088c18100
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:59819512776dfe5710b17640e447695a6e5a30c32c818284119451cb58443900
+size 50496
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..508c53145d5dd0745bdf431cf27916dfefeb5fd6
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a778870ce1036d3e83b275978c0c2b510d1774a6e7caf4a6aaacd8d237d09962
+size 25895
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..73ffd38752a3a7763366807f956e0b42e3428823
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:60ea9a49de44bf039861ab3dbdfde3b167ffbe6613f39848310164ddbb6e0e70
+size 25806
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c691d5fe0ab7f7997e2b40e295c0fffba313f11e
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:34c5741a090a261ac09f5de4e89e9c8d3933287fe6b331670f0ef79d8b876262
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2189292bb0beb0f4e4defa28bb14f2f3ab04accc
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be3ae89a9cbaaaf6a9d40c4458f8f9d374267487777976184e1e851effc080e8
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9d7b9a515d8ebc9de855cf775ea0264db8ef5e58
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da80a161f759e3b9e6b9145aef7030bba0e9fbc7889c73c153e3191e47d4c7f8
+size 37776772
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..37dfc142bca465a1c10340a4535d8e7fe0caf13a
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e1aef766b8f4e6179b1c06141c5dd4bc745eeafde2b886a154c4f21b00932165
+size 50496
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..bc208402ef54b6a406edb57c99522e6db6b38497
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d8e466877fd9c4871bf54c5d59a8122e226860c0e109ac914fddbbe440965b9
+size 25895
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..ecac52b37d4cb783b002646d6f633219346cd74f
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efa94645e0902773f2cf2263c82fd53ec72ed2529b8d1dee298118e1c144b5b3
+size 25806
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..0ae814213ab0b3fa644af64184e6e6535b1617d5
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a713601ed085062e1d78443034986346cb4b3995561f621990ab6f76c5507d3
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..909d2fa46ae83b347220c0d9c9bf6b2db7be3c7e
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d47e763dc026bb11c98420f712dede20b9a9b568e5830acc7f5226ae691fb3ae
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..24d3747f81e17ea9a9cc90e6593a75750d4815ca
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a4db30cecb76382c75364a49b30258f5e0b27a135e06f6b2ab63911c6288ee3
+size 37776772
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b1d153ed38c203251186ab20f7ab53a7406b71bb
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0be70dedd670bbe2499d9a11f64ad09e42c28737f8747bed7a2677ab9ea09cde
+size 50496
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..6cf944bad521f683ec7e4a0fe55c3a4ae2e21710
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5dfa591ec4b8a897bbbf7061ba75919bedd9342a0f2c039209e9c32a3ad20de
+size 25895
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..86d838ed6bb8237b91d2b35eaab0f8699dcf56cb
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0e44e457d413a025ce1602dcb6bc9af0a32995c8557610523ac9c5c92065d1d
+size 25806
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..87031169db24ed11d87d866b36829836fbe263c0
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f77542a530dda7de93d001ef862538a616d221fb886a1ac178384d4ced7f3175
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1287e94c8dd93bcb3e5184f24cb2e6ff0b22fc8b
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d08477a88a74d3f30f2ea9bf7eac6871aa1acc5a7080b3727744cf77382af9c
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7e49a486ae8e8298d4e7f279f975474406640576
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ddda51abf430665b50f6bfa0d49052311292ec821ce324ae9d6634bd4fbd00c
+size 37776772
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9128c39d530b06426f8d6160239c47e0142a9eff
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49977abb50b8ed9fd136d19e8653b8bf50dd446283ba98b868b667ed6a882857
+size 50496
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..33a3344b064fc31a3ce5cca0f6d1fcc13b97eeb5
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91fc383c091ec0afa601934b8a81ede13d95c2dd605e98dca67ac2fbf9dc2ad8
+size 25895
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5ce806350e7b7af7e07f7cbffc721d7cd4448077
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a434a7a65c78a3e398c47934f724712d8caec2a2bf68b772fe8060e3d4c8bf6
+size 25806
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fc666cc8af01db4e97ef7ef317663e23b6ec2ef5
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e0be0757a4c0157d146229244212ea722085e87d16dcc73a25efb2ad14b4195
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..eafd61d29281eb44b104c52dd7abd36ee0165b3c
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2aafafab3d0ef57d5a5f3cca7ed6bcbc59d7c210dd11b82913bb3a2f0aac5d8b
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..36bcf3e6a72837a62525adc64236a4a9edcfb2f7
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae9324ec4e1ab7159440dd70a1e39084de6847a7917ab476fa2cc74f6c2851f8
+size 37776772
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4d1bd05d284706b9c793627e4d538027453d90c2
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef47331550380778ed4b204023d1bba6f04000053e91f3acf84e5bab24e20965
+size 50496
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c36fc3614ddd1f36d3684f1c21e0aa10b2897df0
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f72e72baa2a3f1cef4d0d30106a8e5915500e2b1db6c90751d41433b17913674
+size 25895
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d4beb92d9f5d1fe23bf5f69e7ea33160099a2880
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4fafbda38b9acda22b8d3f7bdafcb8bd04ca25490cd7624a3ec594d2748eb7c
+size 25806
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..bc22dd4a86732394d812c7bb4791cb79e2086758
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef8bab7320a1f1cfdb365f59ab2778fb92ad546f09a374c869fddda7e0aff337
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..70adaf9816c7ca19625d5f23f42b913bd5dd44d7
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db1b15f37475d26ea7b700424f6e660bda122609fc492d90f75989b77b55f1b9
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..24bf07fdc27a3ae9e39b2279866e0b5f026ac360
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa57f7ea4dad753b9df75cf2aa2ac6450587278c1843b6d2e5e82a29a6fe304b
+size 37776772
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9395107b1b2ef83521522936425ad0e39b9f3891
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c0c92864827a98745a4cbc37d43e698292cc54d77dd26316c926a0f3c2c1b34c
+size 50496
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7be088d834662a5611571fc2ee1b1f635c3f65a8
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75ceada0b43d451df50cb7fddb37aaa6f87c3d31cfaeaa583b17018ef0aadf2a
+size 25895
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8d0cc864c68fe1d09b11d3bfcb40528b5434f211
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e05f18113050c16128892fdb9b5493f45f6b30da648babb17ff70b63e2b703e
+size 25806
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7bfa940c46274594e417bdf6d53dd36e9208c63c
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0cb3de757f77bc56cfebb4e54c72694076930a4b1e1544b978077d93bb987e0a
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f47a584e2f0203de122c99df2ae40d6a1a899e92
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0fb5d2a8e2f32e024db76e0b372ea23b67558d5fa3686983f75b5b08a3b620c1
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..de03dc44a633edac0fe77fa2d09cb4b50d7594cb
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:990b4180a19e230b4266b02e42e0f3b10f8c9d379d53dc12b517cc3d8ba4eaca
+size 37776772
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1a66af5b86149310d8616a8575bad8b1095ea07f
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a4a9725e260361bb97f9e8fb4267ec32d7a6a1bbff9575bd64a89d47dd0c85d
+size 50496
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..6cd7c5836b211c0ebabd6d3c1b49e16c9cbde248
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c9f4bdda1ba7166e90ac7e41664043b4648a5ace764aac3f5d448936a461bed
+size 25895
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1faa5a90849d35fedf4f140dc18a2e6e33528989
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4b4228bbddd885a54763e55f9a5ef1e07a5f8e889a78d4c4b95082ff021a65e
+size 25806
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e18799c5fde9735acc9951c3101b0a9adeb12289
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d2b37b465708e9995dbd5cea43c9113100c35b5121f62e7e8e80a8cd63bb7c55
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..139e6398d90cd06d9bcf8886f6f8f8b7cd861bef
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31bdd3fbcfad6b092a95a2b5f6468dc5c59cbce145f440eca5de1bc51f6eb5c6
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c0fd420b70d2114bbcaeb1d63343a2533d9ef550
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf6971795b0c1d41b4a536b34fbb4a7bc9fec48c0fec8c26ee9ace5cad1932cc
+size 37776772
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..cf0215487d32947f1c4d6ca813162c2e24f1718c
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b389adbe33310771fa5f2df47d65a081c2bf041231a3fe67a2a093cecb123dd3
+size 50496
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..804ff42a060559eabddd654a96550db12a6d03df
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:420dc7aaafc3ab30d6bf836b627b69e4921e49a085842f57e02e50b51306bc1a
+size 25895
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..ae9ca5be517410b03e6886b8154be7b3ed92dbef
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee9e0e6297b3a2030e764703dabe5bf1eea6190b08d87ea5a9d19548ec17321e
+size 25806
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..036bc19359a20be3db50537efbfce416bb0458df
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6947d950d2bc44fc3708f1971b6d89453494264fac86a5003cfe0ea3feda708
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7623f95f7bb4815b6609b1b2128d54b0b329752c
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb310a337ee2e7828417a62e3a088e7e5276889296e9cba6560ac816172bd872
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/model.safetensors b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..aac2d455d05a8b80f758cc5f3a4186481ac01756
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dcc7702ec0a06810f8deaab2d802730123641794e3081c36e6659beaac2191a2
+size 37776772
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..7cfe312430fc865b95449ac337ce1ae00e08bbae
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bba19864ab7fca9c9254fe2fdb36ed206c0a6b537ba7e2fc8048473f7e288fe9
+size 50496
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f077afd30f2e6d2e7f5dba2995492d2d38ba9822
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc082f22df9d4e36223355a73841dee42f9dbe055cef3905a2c6ce6e02bab94c
+size 25895
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..01da5e9276fe61a8bf5f956dfe9ac228eb26b203
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f8cded22c0c250c0b54026daec44adb18f92075e3dd65bfdf9ac6ad0a43ffce6
+size 25806
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..369c28adc6bcde137df0f9b0bedc94c3f9a38cf0
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f7aa28af3fbe48925eaf201b84927ba6b767ff56f562949011f393dfe822d54
+size 25811
diff --git a/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5f19ba08517d18120577fe678b914c8a29feff1b
--- /dev/null
+++ b/lm/pythia-160m_layer8/matryoshka/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c1d363cd2e76b3f384c336b306c1adefcd6a2bf07ab78c95da202b19973e9bb
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/model.safetensors b/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..dd7933ff962a5fa02be9d42117ce79497f0c88f0
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca64e94dbaabcf1081884f451f89519d09f7b15d1837ccabd210e31c7e4fb057
+size 37776696
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..41710b36b10b32f764aae8d1c23bf64a36ac9569
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e69d55dbb5f6d9d90f4c76ccec7e4d4ef2ebac10e619b22b6134764777dac47
+size 50496
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b7022032b283526539026ff53a7a7a7bc03c59c3
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bdf15050e2e41564182e44057e990fb0e00872cfbe99021a3b8fd2587c799c78
+size 25895
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5125731dd6a71667ec2253c2619ef7b3e747eecf
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7105f8e98b1e5e99a0c68482ea81be15c3cddb43224349f9af248d60d9f5171c
+size 25806
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..58a3f83f9a51a51b7665eb71eb7c3861c743ae7e
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:815e0c40df2f1b706b9ee9931d7fbf51b15b2a1cecfd04f67efcef8a07778634
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..075659c3da23766ee21d15b8f5661519e60a9407
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6703c721c0b768e3dffb31fa03561f6d9d008523c9a09002b137920d765b28ae
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/model.safetensors b/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8b181993d57775f5578508e962b71e68c8ffd56c
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:71f3b96c1a21125185a9b119b50e76511f5dbd408c625d3349ff6edc6e57572f
+size 37776696
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..36c63246b2ae223845ed4e89f64455e133443738
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6c749828d6f22fc58a3475b22859c67b8f82121694433253aa46ea46ceb7e54
+size 50496
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..834c56b452e150e660d09415db7e026bc28b7d68
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:90077ff318bc0b5b7d7252049b1582c06f7c79ff8d3c631dcf44ddca98b88ba7
+size 25895
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..224a8862028e7bf77cead6df30fe37289d6cc601
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:555b438fab49eb49890be31265deb3dfb5880cf1dae4350a2ca0b09c33751ffc
+size 25806
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..57cf5641a36482c4797d5592c724214ae209dd48
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1f2d5b8361cd6ddd45ae562f080d612ea49c8aea789e2601bf83a88f3790d02
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e0174a3b23d88d911649a5160e017c7ce745c7e8
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37f002bbf07ccd4f82c86c11f76bc8da44d34188f196d6385438c1ce393bb167
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/model.safetensors b/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..38e9b209c2d565e70b0fc94d44eec79d8f715acc
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f653d36d73c7debf7176b3dd8e57a8a3ab6ce7bd76c6877e5eb2be2148d8ea5
+size 37776696
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9002c60b10eb68c05542da1ab04b894ed86f9a24
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:466fa3c5e72eb9dfde96a0c04522f2e3023d30b4bd546eda610448271ca35621
+size 50496
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8f5a202ed8d858d37d078d954a91e75f063cd6e6
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6bef34dc8c4421d39f4f831be01235b001733f4706c5c146ef1a1fc9cafe6bb
+size 25895
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9f43286a6766c51fbd44b167b06c2a02ac382ca2
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1fca7a52d05ec0c06c81607f7841b8250d5d2942aa1c9fa00653a1e8b9365ade
+size 25806
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1b6270af1a345bd4afdfd3884a600f363033ae3b
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:236be1bdee9f399a0e4a97a2e809d06344dff16592178a8a96b32c569b6ddaad
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..05996d051ea9f2b8ad14bbe420e6da3dd1f0c2a8
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_128/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:491046902064c271d1864cefa61f33c2341c643fc64122f84267d30a182dcfb9
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/model.safetensors b/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..dd7933ff962a5fa02be9d42117ce79497f0c88f0
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca64e94dbaabcf1081884f451f89519d09f7b15d1837ccabd210e31c7e4fb057
+size 37776696
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..41710b36b10b32f764aae8d1c23bf64a36ac9569
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e69d55dbb5f6d9d90f4c76ccec7e4d4ef2ebac10e619b22b6134764777dac47
+size 50496
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b7022032b283526539026ff53a7a7a7bc03c59c3
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bdf15050e2e41564182e44057e990fb0e00872cfbe99021a3b8fd2587c799c78
+size 25895
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5125731dd6a71667ec2253c2619ef7b3e747eecf
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7105f8e98b1e5e99a0c68482ea81be15c3cddb43224349f9af248d60d9f5171c
+size 25806
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..58a3f83f9a51a51b7665eb71eb7c3861c743ae7e
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:815e0c40df2f1b706b9ee9931d7fbf51b15b2a1cecfd04f67efcef8a07778634
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..075659c3da23766ee21d15b8f5661519e60a9407
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6703c721c0b768e3dffb31fa03561f6d9d008523c9a09002b137920d765b28ae
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/model.safetensors b/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8b181993d57775f5578508e962b71e68c8ffd56c
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:71f3b96c1a21125185a9b119b50e76511f5dbd408c625d3349ff6edc6e57572f
+size 37776696
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..36c63246b2ae223845ed4e89f64455e133443738
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6c749828d6f22fc58a3475b22859c67b8f82121694433253aa46ea46ceb7e54
+size 50496
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..834c56b452e150e660d09415db7e026bc28b7d68
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:90077ff318bc0b5b7d7252049b1582c06f7c79ff8d3c631dcf44ddca98b88ba7
+size 25895
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..224a8862028e7bf77cead6df30fe37289d6cc601
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:555b438fab49eb49890be31265deb3dfb5880cf1dae4350a2ca0b09c33751ffc
+size 25806
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..57cf5641a36482c4797d5592c724214ae209dd48
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1f2d5b8361cd6ddd45ae562f080d612ea49c8aea789e2601bf83a88f3790d02
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e0174a3b23d88d911649a5160e017c7ce745c7e8
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37f002bbf07ccd4f82c86c11f76bc8da44d34188f196d6385438c1ce393bb167
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/model.safetensors b/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..38e9b209c2d565e70b0fc94d44eec79d8f715acc
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f653d36d73c7debf7176b3dd8e57a8a3ab6ce7bd76c6877e5eb2be2148d8ea5
+size 37776696
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9002c60b10eb68c05542da1ab04b894ed86f9a24
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:466fa3c5e72eb9dfde96a0c04522f2e3023d30b4bd546eda610448271ca35621
+size 50496
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8f5a202ed8d858d37d078d954a91e75f063cd6e6
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6bef34dc8c4421d39f4f831be01235b001733f4706c5c146ef1a1fc9cafe6bb
+size 25895
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9f43286a6766c51fbd44b167b06c2a02ac382ca2
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1fca7a52d05ec0c06c81607f7841b8250d5d2942aa1c9fa00653a1e8b9365ade
+size 25806
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1b6270af1a345bd4afdfd3884a600f363033ae3b
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:236be1bdee9f399a0e4a97a2e809d06344dff16592178a8a96b32c569b6ddaad
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..05996d051ea9f2b8ad14bbe420e6da3dd1f0c2a8
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_32/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:491046902064c271d1864cefa61f33c2341c643fc64122f84267d30a182dcfb9
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/model.safetensors b/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..dd7933ff962a5fa02be9d42117ce79497f0c88f0
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca64e94dbaabcf1081884f451f89519d09f7b15d1837ccabd210e31c7e4fb057
+size 37776696
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..41710b36b10b32f764aae8d1c23bf64a36ac9569
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e69d55dbb5f6d9d90f4c76ccec7e4d4ef2ebac10e619b22b6134764777dac47
+size 50496
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..b7022032b283526539026ff53a7a7a7bc03c59c3
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bdf15050e2e41564182e44057e990fb0e00872cfbe99021a3b8fd2587c799c78
+size 25895
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..5125731dd6a71667ec2253c2619ef7b3e747eecf
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7105f8e98b1e5e99a0c68482ea81be15c3cddb43224349f9af248d60d9f5171c
+size 25806
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..58a3f83f9a51a51b7665eb71eb7c3861c743ae7e
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:815e0c40df2f1b706b9ee9931d7fbf51b15b2a1cecfd04f67efcef8a07778634
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..075659c3da23766ee21d15b8f5661519e60a9407
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed0/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6703c721c0b768e3dffb31fa03561f6d9d008523c9a09002b137920d765b28ae
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..224a8862028e7bf77cead6df30fe37289d6cc601
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:555b438fab49eb49890be31265deb3dfb5880cf1dae4350a2ca0b09c33751ffc
+size 25806
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..57cf5641a36482c4797d5592c724214ae209dd48
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1f2d5b8361cd6ddd45ae562f080d612ea49c8aea789e2601bf83a88f3790d02
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e0174a3b23d88d911649a5160e017c7ce745c7e8
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed1/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37f002bbf07ccd4f82c86c11f76bc8da44d34188f196d6385438c1ce393bb167
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/model.safetensors b/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..38e9b209c2d565e70b0fc94d44eec79d8f715acc
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f653d36d73c7debf7176b3dd8e57a8a3ab6ce7bd76c6877e5eb2be2148d8ea5
+size 37776696
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9002c60b10eb68c05542da1ab04b894ed86f9a24
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/active_count_hist_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:466fa3c5e72eb9dfde96a0c04522f2e3023d30b4bd546eda610448271ca35621
+size 50496
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt b/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
new file mode 100644
index 0000000000000000000000000000000000000000..8f5a202ed8d858d37d078d954a91e75f063cd6e6
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/firing_rate_per_latent.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6bef34dc8c4421d39f4f831be01235b001733f4706c5c146ef1a1fc9cafe6bb
+size 25895
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..9f43286a6766c51fbd44b167b06c2a02ac382ca2
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/max_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1fca7a52d05ec0c06c81607f7841b8250d5d2942aa1c9fa00653a1e8b9365ade
+size 25806
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1b6270af1a345bd4afdfd3884a600f363033ae3b
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/mean_act_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:236be1bdee9f399a0e4a97a2e809d06344dff16592178a8a96b32c569b6ddaad
+size 25811
diff --git a/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt b/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
new file mode 100644
index 0000000000000000000000000000000000000000..05996d051ea9f2b8ad14bbe420e6da3dd1f0c2a8
--- /dev/null
+++ b/lm/pythia-160m_layer8/mp/x8/l0_64/seed2/stats.fineweb.split_train.seed_no/sparsity_per_token.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:491046902064c271d1864cefa61f33c2341c643fc64122f84267d30a182dcfb9
+size 25811