siddharthmb commited on
Commit
acbbbdc
·
verified ·
1 Parent(s): a52894a

Upload probes, handles, and attention diagnostics

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. README.md +78 -0
  2. checksums.json +0 -0
  3. exp2/handles/expression_dir_l31_lenient.json +17 -0
  4. exp2/handles/expression_dir_l31_lenient.pt +3 -0
  5. exp2/handles/expression_dir_l31_strict.json +17 -0
  6. exp2/handles/expression_dir_l31_strict.pt +3 -0
  7. exp2/probes/dense/metrics.parquet +3 -0
  8. exp2/probes/h1/bow_metrics.parquet +3 -0
  9. exp2/probes/h1/bow_scores.parquet +3 -0
  10. exp2/probes/h1/layer_03/lenient/probe_final.pt +3 -0
  11. exp2/probes/h1/layer_03/lenient/standardizer.pt +3 -0
  12. exp2/probes/h1/layer_03/strict/probe_final.pt +3 -0
  13. exp2/probes/h1/layer_03/strict/standardizer.pt +3 -0
  14. exp2/probes/h1/layer_07/lenient/probe_final.pt +3 -0
  15. exp2/probes/h1/layer_07/lenient/standardizer.pt +3 -0
  16. exp2/probes/h1/layer_07/strict/probe_final.pt +3 -0
  17. exp2/probes/h1/layer_07/strict/standardizer.pt +3 -0
  18. exp2/probes/h1/layer_11/lenient/probe_final.pt +3 -0
  19. exp2/probes/h1/layer_11/lenient/standardizer.pt +3 -0
  20. exp2/probes/h1/layer_11/strict/probe_final.pt +3 -0
  21. exp2/probes/h1/layer_11/strict/standardizer.pt +3 -0
  22. exp2/probes/h1/layer_15/lenient/probe_final.pt +3 -0
  23. exp2/probes/h1/layer_15/lenient/standardizer.pt +3 -0
  24. exp2/probes/h1/layer_15/strict/probe_final.pt +3 -0
  25. exp2/probes/h1/layer_15/strict/standardizer.pt +3 -0
  26. exp2/probes/h1/layer_19/lenient/probe_final.pt +3 -0
  27. exp2/probes/h1/layer_19/lenient/standardizer.pt +3 -0
  28. exp2/probes/h1/layer_19/strict/probe_final.pt +3 -0
  29. exp2/probes/h1/layer_19/strict/standardizer.pt +3 -0
  30. exp2/probes/h1/layer_23/lenient/probe_final.pt +3 -0
  31. exp2/probes/h1/layer_23/lenient/standardizer.pt +3 -0
  32. exp2/probes/h1/layer_23/strict/probe_final.pt +3 -0
  33. exp2/probes/h1/layer_23/strict/standardizer.pt +3 -0
  34. exp2/probes/h1/layer_27/lenient/probe_final.pt +3 -0
  35. exp2/probes/h1/layer_27/lenient/standardizer.pt +3 -0
  36. exp2/probes/h1/layer_27/strict/probe_final.pt +3 -0
  37. exp2/probes/h1/layer_27/strict/standardizer.pt +3 -0
  38. exp2/probes/h1/layer_31/lenient/probe_final.pt +3 -0
  39. exp2/probes/h1/layer_31/lenient/standardizer.pt +3 -0
  40. exp2/probes/h1/layer_31/strict/probe_final.pt +3 -0
  41. exp2/probes/h1/layer_31/strict/standardizer.pt +3 -0
  42. exp2/probes/h1/layer_35/lenient/probe_final.pt +3 -0
  43. exp2/probes/h1/layer_35/lenient/standardizer.pt +3 -0
  44. exp2/probes/h1/layer_35/strict/probe_final.pt +3 -0
  45. exp2/probes/h1/layer_35/strict/standardizer.pt +3 -0
  46. exp2/probes/h1/metrics.parquet +3 -0
  47. exp2/probes/h1/metrics_l11.parquet +3 -0
  48. exp2/probes/h1/metrics_l15.parquet +3 -0
  49. exp2/probes/h1/metrics_l19.parquet +3 -0
  50. exp2/probes/h1/metrics_l23.parquet +3 -0
README.md ADDED
@@ -0,0 +1,78 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ language:
4
+ - en
5
+ tags:
6
+ - interpretability
7
+ - probes
8
+ - steering
9
+ - multi-agent
10
+ - qwen3
11
+ library_name: pytorch
12
+ ---
13
+
14
+ # Multi-Agent Verification Probes & Direction Handles
15
+
16
+ Trained linear probes, difference-in-means direction handles, and attention
17
+ diagnostics from an interpretability study of failure modes in multi-agent fact
18
+ verification (Qwen3-32B verifier + Qwen3-8B subagents on AVeriTeC claims).
19
+
20
+ These artifacts were trained on the residual-stream activations released in the
21
+ companion dataset repo:
22
+ [siddharthmb/multiagent-verification-failure-modes](https://huggingface.co/datasets/siddharthmb/multiagent-verification-failure-modes).
23
+ Code: [Sid-MB/mats-gf-multiagent-failure-modes](https://github.com/Sid-MB/mats-gf-multiagent-failure-modes).
24
+
25
+ ## Contents
26
+
27
+ - `exp2/` — transmission-loss interpretability (Qwen3-8B subagent / Qwen3-32B verifier):
28
+ - `probes/h1/` — per-layer (3–35, step 4) linear probes predicting
29
+ gold-evidence-in-context at the subagent message-final token position, at two
30
+ transmission-loss thresholds, plus bag-of-words shortcut baselines
31
+ (`bow_*.parquet`) and per-layer metrics.
32
+ - `probes/h2/` — per-(layer, round) 4-class verdict-commitment probes at the
33
+ Qwen3-32B verifier think-close position (layers 3–63, step 4), with
34
+ round-shuffled and label-shuffled controls and row-level predictions.
35
+ - `probes/dense/` — metrics for the dense tagged-position probe sweeps.
36
+ - `handles/` — conveyed-minus-unconveyed difference-in-means "expression"
37
+ directions at Qwen3-8B layer 31 (strict and lenient thresholds), as `.pt`
38
+ with a JSON manifest. Held-out conveyed-vs-lost AUC 0.765. Note: steering
39
+ with this direction did **not** causally force conveyance (a null result);
40
+ it is a predictive direction, not a validated intervention handle.
41
+ - `exp5/` — trust interpretability (Qwen3-32B verifier):
42
+ - `probes/` — content-quality and source-tier probes per layer x round-context,
43
+ in four families (`gold_first_blind`, `gold_first_told`, `strong_first_blind`,
44
+ `strong_first_told`), with standardizers.
45
+ - `handles/` — told-minus-blind disclosure directions per layer (3–63, step 4)
46
+ and round-context, with `manifest.json` (per-direction norm, matched-group
47
+ counts, separation AUC). The verdict-context **layer-51** direction predicts
48
+ tier-tracking (AUC 0.82) and causally shifted single verdicts in both
49
+ directions vs a norm-matched sham when steered.
50
+ - `exp6/` — attention-pickup diagnostics (Qwen3-8B subagent):
51
+ - `attention/span_mass.parquet` — teacher-forced attention mass on the gold
52
+ evidence span per (turn, layer, phase) for 830 fact-check turns, comparing
53
+ failed vs conveyed transmission-loss turns.
54
+ - `attention/exemplars/*.npz` — 20 full attention maps for exemplar turns.
55
+ - exp6 saved no direction handles of its own (its steering interventions reused
56
+ exp2's expression direction), so there is no `exp6/handles/`.
57
+
58
+ ## Formats and loading
59
+
60
+ - Probe checkpoints (`*.pt`) are [goodfire-core](https://github.com/goodfire-ai)
61
+ `LinearProbe` checkpoints (a state dict plus normalization statistics);
62
+ `torch.load(path, map_location="cpu")` shows the raw contents, or use
63
+ `LinearProbe.from_checkpoint` with goodfire-core installed. Inputs are
64
+ standardized residual-stream activations at the position/layer named in the file.
65
+ - Direction handles (`*.pt`) are plain tensors (residual-stream-dimensional
66
+ vectors) with accompanying JSON manifests describing layer, context, norm,
67
+ and evaluation numbers.
68
+ - `attention/exemplars/*.npz` are numpy archives of per-head attention maps.
69
+
70
+ `checksums.json` at the repo root maps every file to its sha256 and size.
71
+
72
+ ## Provenance
73
+
74
+ Trained on our own harvested activations of Qwen/Qwen3-8B and Qwen/Qwen3-32B
75
+ (Apache 2.0) replayed over the episode corpus in the companion dataset repo.
76
+ Weights released under Apache 2.0. Note the underlying episode *text* corpus is
77
+ CC BY-NC 4.0 (AVeriTeC-derived); that license applies to the dataset repo, not
78
+ to these probe weights.
checksums.json ADDED
The diff for this file is too large to render. See raw diff
 
exp2/handles/expression_dir_l31_lenient.json ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "Qwen/Qwen3-8B",
3
+ "layer": 31,
4
+ "position": "subagent_msg_last",
5
+ "threshold": "lenient",
6
+ "construction": "difference-in-means conveyed-minus-unconveyed, train claims",
7
+ "n_train_conveyed": 266,
8
+ "n_train_tl": 1123,
9
+ "direction_norm_raw": 24.333595275878906,
10
+ "heldout_auc_conveyed_vs_tl": 0.6725786328315735,
11
+ "heldout_mean_gap": 19.81165313720703,
12
+ "bow_matched_pairs_frac_separated": 0.6289592760180995,
13
+ "bow_matched_pairs_mean_similarity": 0.13923797817248934,
14
+ "n_matched_pairs": 442,
15
+ "within_conveyed_control_frac": 0.3333333333333333,
16
+ "note": "matched-pairs frac near heldout separation => not lexical; control frac near 0.5 => no spurious within-class separation"
17
+ }
exp2/handles/expression_dir_l31_lenient.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5df6c8b24a0294e004120287014ee22cf9379b060d66e0b9ac492ba53cb3268e
3
+ size 35517
exp2/handles/expression_dir_l31_strict.json ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "Qwen/Qwen3-8B",
3
+ "layer": 31,
4
+ "position": "subagent_msg_last",
5
+ "threshold": "strict",
6
+ "construction": "difference-in-means conveyed-minus-unconveyed, train claims",
7
+ "n_train_conveyed": 266,
8
+ "n_train_tl": 301,
9
+ "direction_norm_raw": 39.27574920654297,
10
+ "heldout_auc_conveyed_vs_tl": 0.7649123072624207,
11
+ "heldout_mean_gap": 43.508705139160156,
12
+ "bow_matched_pairs_frac_separated": 0.7684210526315789,
13
+ "bow_matched_pairs_mean_similarity": 0.1278183632599857,
14
+ "n_matched_pairs": 95,
15
+ "within_conveyed_control_frac": 0.3333333333333333,
16
+ "note": "matched-pairs frac near heldout separation => not lexical; control frac near 0.5 => no spurious within-class separation"
17
+ }
exp2/handles/expression_dir_l31_strict.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9414a1e4e064a94e0b4fef035b4c46a839eec1adf76a500d668533b26290310f
3
+ size 35509
exp2/probes/dense/metrics.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:88afa8792899ae38ee1b30fc2438bf47742fef5686cbee7cb9889d163b8a0fcd
3
+ size 8495
exp2/probes/h1/bow_metrics.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f2c1b8f5138f4e35146d4da44bd980d8b68d3ef351610f030b5b7517197ece5d
3
+ size 2496
exp2/probes/h1/bow_scores.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:39cc8e983d392e11cc29b253e34719ae3601ad5277bbc8084e9c46eba321c245
3
+ size 348604
exp2/probes/h1/layer_03/lenient/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b3b457d5e0de26d107a7907f78f9e28f21aaef146b622c4f81251cdda6c9d952
3
+ size 36613
exp2/probes/h1/layer_03/lenient/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b5fb2f6f491a3d8529f5850a3719642bf7ba197920bd1b019ef8ccdc0f9c4c72
3
+ size 34637
exp2/probes/h1/layer_03/strict/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:410f2492d005071ed7c8330ad362de1e7b977716818e663f5a8be2717759988b
3
+ size 36613
exp2/probes/h1/layer_03/strict/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:23fe0b259122e3ad7f794b20b8c130b7aacb50c115bfdadc79e7ceeb04610c6f
3
+ size 34637
exp2/probes/h1/layer_07/lenient/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b99361924c1e4595b84725befaeff077573c3f7efdcce461e43d81ec769cf9dd
3
+ size 36613
exp2/probes/h1/layer_07/lenient/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dde7b43d527fb46dd5126a0d94f94607525f5b2ebf1912c1eae3e09929cfd394
3
+ size 34637
exp2/probes/h1/layer_07/strict/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ffc5cf52aadbfa3fb000c47ececf7ea4ec4435227355f6dd49dd48595eba255
3
+ size 36613
exp2/probes/h1/layer_07/strict/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:76ef594fafe0e860d1e33d87fd80c8901193b74f61c752da2f5c698f3f3fdc8f
3
+ size 34637
exp2/probes/h1/layer_11/lenient/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7389e9ee7c596bd9f292a88cf1266eee38afc9e5ff3b6bfc2e0b7c28ff6b4368
3
+ size 36613
exp2/probes/h1/layer_11/lenient/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8cafd178dae4dbb4bf77008e8dadf7fc8a03d65e7e07213b49e3e8894974f8bf
3
+ size 34637
exp2/probes/h1/layer_11/strict/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c98cbe1c63bafa7db7e420aa3327a1b66036c62632835b5e98ded877fd5fe6e6
3
+ size 36613
exp2/probes/h1/layer_11/strict/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3ebf9691c688525e39cfa05400856481ec985035b63d77ec5fb141cb22e04739
3
+ size 34637
exp2/probes/h1/layer_15/lenient/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:22d426ca96baeea4e4cc147549304bf5cd94f771959574c2c2995a54d5377629
3
+ size 36613
exp2/probes/h1/layer_15/lenient/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:15f327ce94d59a553a03e523908fc8ab8e03f4930e245c596fd557592f046a1f
3
+ size 34637
exp2/probes/h1/layer_15/strict/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8c86788ed187646c30267b8432d58024bfd7dc42e851ec28f6fe7e3e3369cafa
3
+ size 36613
exp2/probes/h1/layer_15/strict/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:27b2830d61dea7a9f33474d5d8d3fdfde20131af4b372ede2a4396acbb3918bb
3
+ size 34637
exp2/probes/h1/layer_19/lenient/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6641713aceb91c3c033ced75bf61089a427ed31d5b5986c562b920c0e4b921a3
3
+ size 36613
exp2/probes/h1/layer_19/lenient/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e70038730b511f7944b4c210c2377551b0ba2f3981cd1fcec7b5a8b44f2d2cb2
3
+ size 34637
exp2/probes/h1/layer_19/strict/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:92544bee882cfc7adbd54d659f7393730f1c9942c7288e59291d0db95ffdb3b9
3
+ size 36613
exp2/probes/h1/layer_19/strict/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5482411370cee4530e4640d7772074ff0549d2710572b4b674c63378397278d3
3
+ size 34637
exp2/probes/h1/layer_23/lenient/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3e4968afa9a3e1ab4a27dd37b9e2c0de8db2377ffa4a9a7b65c4d71d3a53be38
3
+ size 36613
exp2/probes/h1/layer_23/lenient/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:55a82ff27b60db4a4a1d3327c5192d8437209d37f54fa0d6bf06b237edda5aee
3
+ size 34637
exp2/probes/h1/layer_23/strict/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a8103cd0ebe8938b2f78682d4e11814681c76516f3c9efae2af76f77e8aad76e
3
+ size 36613
exp2/probes/h1/layer_23/strict/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d0c25955c88673cf47628f2b639e4abeb0345d2176dce5ea786579bf1f19b1d0
3
+ size 34637
exp2/probes/h1/layer_27/lenient/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9e832f9483011a7e0755613c31763676a38541b5095a879f9f92bddb36e6261e
3
+ size 36613
exp2/probes/h1/layer_27/lenient/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3903e0e35fde782eb89ab704daa15d89b1f98c7eed4d685c4555322c57cf7e56
3
+ size 34637
exp2/probes/h1/layer_27/strict/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61aab03c459a4f30b74ef7aca51470bf6cf87b0540bc82dabc1eda121a3d27dd
3
+ size 36613
exp2/probes/h1/layer_27/strict/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:191a3a6d029b27f0847e8517b275acc3daa6136ba15c3a2234b0af7746cf5f11
3
+ size 34637
exp2/probes/h1/layer_31/lenient/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9c280c3da29aa4433800bf28f4f636c310fe9852c520867d61648e970a68a7f1
3
+ size 36613
exp2/probes/h1/layer_31/lenient/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5bbc666638983f6fc5b54a2c05d5f607c95f59aa1a3880ea3fd098c423d39fbc
3
+ size 34637
exp2/probes/h1/layer_31/strict/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6f7bca883d4832efec26c3a9b0887f454f501dc08082f04244831035b126e051
3
+ size 36613
exp2/probes/h1/layer_31/strict/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:38065c8661d9b7c3b93dddbca68ab913852a544ad9fa5f41d9e32e30c5b35092
3
+ size 34637
exp2/probes/h1/layer_35/lenient/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9f4564e0c776fa820fbc4db26d2cbce85a1cf16cd0d97067b67753e14e4ea710
3
+ size 36613
exp2/probes/h1/layer_35/lenient/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5b21cbc2ce96455d7307a4ce20e26617af79e3f1189b41bec374324e6895e506
3
+ size 34637
exp2/probes/h1/layer_35/strict/probe_final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9456cec468c1b8b511cc1dc25310be49d1dd5bc8d3bd1b5ef80d893b3ff818f7
3
+ size 36613
exp2/probes/h1/layer_35/strict/standardizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d2aca8826f27676122aee21a2d2eccf305ec3b47e83721f304c01e8c9d106b95
3
+ size 34637
exp2/probes/h1/metrics.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8fb489063b02ba5b1dc6b999196741792a396ef4fb557bc00ef0c4e3263039c7
3
+ size 6432
exp2/probes/h1/metrics_l11.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2c897a6a6c1e703e76f90c272043bd335e8403341bd28fe91f44a4ad56891a93
3
+ size 4794
exp2/probes/h1/metrics_l15.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a18b10af2f7c15256990387512a1a0a17bef5e49ee343ddbd1173d5a164516c6
3
+ size 4789
exp2/probes/h1/metrics_l19.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4a49755109d622e531c6ad5096488625961798a661466587764eac42cdee4852
3
+ size 4781
exp2/probes/h1/metrics_l23.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6cf8a3485c4489cd626bccda693f9f7a7f4bfcec6f6aad8ed6c0a71a0a87c212
3
+ size 4781