Update instruction_template_retriever.py
Browse files
instruction_template_retriever.py
CHANGED
@@ -1,5 +1,7 @@
|
|
1 |
import itertools
|
2 |
import json
|
|
|
|
|
3 |
|
4 |
from datasets import load_dataset
|
5 |
import faiss
|
@@ -168,7 +170,9 @@ def unuse_gaussian_coverage_pooling(m):
|
|
168 |
|
169 |
class InstructionTemplateRetriever:
|
170 |
FINETEMPLATES_REVISION = "831ab22c90f9da011bd972585afdf609f40fa54b"
|
171 |
-
RETRIEVAL_EMBEDDING_NAME =
|
|
|
|
|
172 |
RETRIEVAL_EMBEDDING_REVISION = "db4efbde126216250ffa5a356663fc7da3bf7856"
|
173 |
|
174 |
def __init__(
|
@@ -222,6 +226,21 @@ class InstructionTemplateRetriever:
|
|
222 |
elif torch.backends.mps.is_available():
|
223 |
self.m = self.m.to("mps")
|
224 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
225 |
def _filter_rows(self, rows, filter_string):
|
226 |
if not rows:
|
227 |
return []
|
@@ -233,7 +252,14 @@ class InstructionTemplateRetriever:
|
|
233 |
return rows
|
234 |
|
235 |
def search(
|
236 |
-
self,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
237 |
):
|
238 |
"""
|
239 |
Given a document
|
@@ -246,6 +272,31 @@ class InstructionTemplateRetriever:
|
|
246 |
deduplicate (bool): Deduplicate results between coverage sections.
|
247 |
"""
|
248 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
249 |
# Search FAISS index
|
250 |
vecs = self.m.encode([document], normalize_embeddings=False).reshape(
|
251 |
-1, self.m[0].auto_model.config.hidden_size
|
@@ -284,7 +335,7 @@ class InstructionTemplateRetriever:
|
|
284 |
"score": s.item(),
|
285 |
**d_in_mem[i.item()],
|
286 |
}
|
287 |
-
for i, s in zip(indices, scores)
|
288 |
]
|
289 |
for chunk_idx, (indices, scores) in enumerate(
|
290 |
zip(indices_per_input, scores_per_input)
|
|
|
1 |
import itertools
|
2 |
import json
|
3 |
+
import pickle
|
4 |
+
from random import Random
|
5 |
|
6 |
from datasets import load_dataset
|
7 |
import faiss
|
|
|
170 |
|
171 |
class InstructionTemplateRetriever:
|
172 |
FINETEMPLATES_REVISION = "831ab22c90f9da011bd972585afdf609f40fa54b"
|
173 |
+
RETRIEVAL_EMBEDDING_NAME = (
|
174 |
+
"fineinstructions/instruction_template_retrieval_embedding"
|
175 |
+
)
|
176 |
RETRIEVAL_EMBEDDING_REVISION = "db4efbde126216250ffa5a356663fc7da3bf7856"
|
177 |
|
178 |
def __init__(
|
|
|
226 |
elif torch.backends.mps.is_available():
|
227 |
self.m = self.m.to("mps")
|
228 |
|
229 |
+
with open(
|
230 |
+
hf_hub_download(
|
231 |
+
"fineinstructions/finetemplates",
|
232 |
+
"faiss_index/reweighting_stats.pkl",
|
233 |
+
revision=FINETEMPLATES_REVISION,
|
234 |
+
repo_type="dataset",
|
235 |
+
),
|
236 |
+
"rb",
|
237 |
+
) as reweighting_stats_fp:
|
238 |
+
reweighting_stats = pickle.load(reweighting_stats_fp)
|
239 |
+
self.resampling_weights = reweighting_stats["resampling_weights"]
|
240 |
+
self.template_variable_count_mapping = reweighting_stats[
|
241 |
+
"template_variable_count_mapping"
|
242 |
+
]
|
243 |
+
|
244 |
def _filter_rows(self, rows, filter_string):
|
245 |
if not rows:
|
246 |
return []
|
|
|
252 |
return rows
|
253 |
|
254 |
def search(
|
255 |
+
self,
|
256 |
+
document,
|
257 |
+
filters="",
|
258 |
+
search_k=20000,
|
259 |
+
max_results=250,
|
260 |
+
deduplicate=True,
|
261 |
+
reweight=False,
|
262 |
+
reweighting_epsilon=True,
|
263 |
):
|
264 |
"""
|
265 |
Given a document
|
|
|
272 |
deduplicate (bool): Deduplicate results between coverage sections.
|
273 |
"""
|
274 |
|
275 |
+
def _reweight(inp, k=None):
|
276 |
+
if reweight:
|
277 |
+
inp0, inp = itertools.tee(inp)
|
278 |
+
first_row = next(inp0)
|
279 |
+
r = Random(first_row[1].item())
|
280 |
+
epsilon = reweighting_epsilon
|
281 |
+
bucket = first_row[1]
|
282 |
+
items = []
|
283 |
+
weights = []
|
284 |
+
for i, s in inp:
|
285 |
+
if abs(bucket - s.item()) <= epsilon:
|
286 |
+
items.append((i, s))
|
287 |
+
weights.append(
|
288 |
+
self.resampling_weights[
|
289 |
+
self.template_variable_count_mapping[i.item()]
|
290 |
+
]
|
291 |
+
)
|
292 |
+
else:
|
293 |
+
break
|
294 |
+
return r.choices(
|
295 |
+
items, weights=weights, k=(len(items) if k is None else k)
|
296 |
+
)
|
297 |
+
else:
|
298 |
+
return inp
|
299 |
+
|
300 |
# Search FAISS index
|
301 |
vecs = self.m.encode([document], normalize_embeddings=False).reshape(
|
302 |
-1, self.m[0].auto_model.config.hidden_size
|
|
|
335 |
"score": s.item(),
|
336 |
**d_in_mem[i.item()],
|
337 |
}
|
338 |
+
for i, s in _reweight(zip(indices, scores), k=None)
|
339 |
]
|
340 |
for chunk_idx, (indices, scores) in enumerate(
|
341 |
zip(indices_per_input, scores_per_input)
|