Datasets:
Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -10,10 +10,10 @@ configs:
|
|
| 10 |
data_files: "thinkies_v3_preview.parquet"
|
| 11 |
---
|
| 12 |
|
| 13 |
-
# thinkies v3 —
|
| 14 |
|
| 15 |
One row per short natural-language phrase, paired with the layer-42 residual-stream direction
|
| 16 |
-
that phrase induces in Qwen3.6-27B.
|
| 17 |
the **same reference frame**, so vectors from the two releases are directly comparable.
|
| 18 |
|
| 19 |
| column | type | meaning |
|
|
@@ -45,7 +45,7 @@ vector means its label, and it does not predict how uniquely the label pins the
|
|
| 45 |
|
| 46 |
Filters applied: `reliability >= 0.65`, labels of at most 8 tokens, then
|
| 47 |
activation-space dedup at cosine > 0.95 via LSH bucketing, which removed 27,366 rows
|
| 48 |
-
(
|
| 49 |
|
| 50 |
## Label length
|
| 51 |
|
|
|
|
| 10 |
data_files: "thinkies_v3_preview.parquet"
|
| 11 |
---
|
| 12 |
|
| 13 |
+
# thinkies v3 — 1,583,873 concept vectors from Qwen3.6-27B
|
| 14 |
|
| 15 |
One row per short natural-language phrase, paired with the layer-42 residual-stream direction
|
| 16 |
+
that phrase induces in Qwen3.6-27B. 2.05x the size of `ceselder/thinkies-v2` (772,199) and in
|
| 17 |
the **same reference frame**, so vectors from the two releases are directly comparable.
|
| 18 |
|
| 19 |
| column | type | meaning |
|
|
|
|
| 45 |
|
| 46 |
Filters applied: `reliability >= 0.65`, labels of at most 8 tokens, then
|
| 47 |
activation-space dedup at cosine > 0.95 via LSH bucketing, which removed 27,366 rows
|
| 48 |
+
(1.70%) — differently-worded phrases the model represents identically.
|
| 49 |
|
| 50 |
## Label length
|
| 51 |
|