Upload code/gpt2_align.py with huggingface_hub
Browse files- code/gpt2_align.py +15 -0
code/gpt2_align.py
CHANGED
|
@@ -123,3 +123,18 @@ def align_full(sd_a, sd_b, d, nh, acts_a=None, acts_b=None, method="permutation"
|
|
| 123 |
hp = head_match(sd_a, sd, d, nh)
|
| 124 |
sd, ok = keep(apply_head(sd, hp, d, nh), "heads"); info["heads"] = len(hp) if ok else 0
|
| 125 |
return sd, info
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 123 |
hp = head_match(sd_a, sd, d, nh)
|
| 124 |
sd, ok = keep(apply_head(sd, hp, d, nh), "heads"); info["heads"] = len(hp) if ok else 0
|
| 125 |
return sd, info
|
| 126 |
+
|
| 127 |
+
|
| 128 |
+
# --------------------------------------------------------------- embedding-row Procrustes
|
| 129 |
+
def emb_procrustes(sd_a, sd_b, tok_a, tok_b, key="transformer.wte.weight", max_anchors=None):
|
| 130 |
+
"""Fit the residual-basis rotation from the EMBEDDING ROWS of shared surface forms, data-free.
|
| 131 |
+
|
| 132 |
+
Rows of the two embedding tables that spell the same string are the one correspondence two
|
| 133 |
+
monolingual tokenizers genuinely share, so `min_R ||E_a[anchors] - E_b[anchors] R||` is a
|
| 134 |
+
legitimate estimate of the basis map with no corpus and no forward pass. Returns (R, n_anchors).
|
| 135 |
+
"""
|
| 136 |
+
anchors = AL.vocab_anchors(tok_a, tok_b, max_anchors)
|
| 137 |
+
ia = [i for i, _ in anchors]; ib = [j for _, j in anchors]
|
| 138 |
+
Ea = np.asarray(sd_a[key], float)[ia]
|
| 139 |
+
Eb = np.asarray(sd_b[key], float)[ib]
|
| 140 |
+
return AL.procrustes_align(Ea, Eb)["R"], len(anchors)
|