AppleSwing
commited on
Commit
•
a3cdaa8
1
Parent(s):
c2dbb45
fix generation issue
Browse files
src/backend/hflm_with_measurement.py
CHANGED
@@ -294,7 +294,7 @@ class HFLMWithMeasurement(HFLM):
|
|
294 |
|
295 |
return re_ord.get_original(res)
|
296 |
|
297 |
-
def _model_generate(self, context,
|
298 |
# temperature = 0.0 if not set
|
299 |
# if do_sample is false and temp==0.0:
|
300 |
# remove temperature, as do_sample=False takes care of this
|
@@ -302,7 +302,7 @@ class HFLMWithMeasurement(HFLM):
|
|
302 |
generation_kwargs["temperature"] = generation_kwargs.get("temperature", 0.0)
|
303 |
do_sample = generation_kwargs.get("do_sample", None)
|
304 |
|
305 |
-
is_gsm8k = generation_kwargs.get("is_gsm8k", False)
|
306 |
|
307 |
# The temperature has to be a strictly positive float -- if it is 0.0, use greedy decoding strategies
|
308 |
if generation_kwargs.get("temperature") == 0.0 and do_sample is None:
|
@@ -311,12 +311,12 @@ class HFLMWithMeasurement(HFLM):
|
|
311 |
if do_sample is False and generation_kwargs.get("temperature") == 0.0:
|
312 |
generation_kwargs.pop("temperature")
|
313 |
|
314 |
-
if is_gsm8k:
|
315 |
-
|
316 |
|
317 |
context_length = context.shape[1]
|
318 |
model_config = self.model.config
|
319 |
-
|
320 |
if not self.precision:
|
321 |
if model_config.quantization_config._load_in_4bit:
|
322 |
self.precision = "4bit"
|
@@ -325,38 +325,21 @@ class HFLMWithMeasurement(HFLM):
|
|
325 |
else:
|
326 |
raise ValueError("Unknown precision")
|
327 |
|
328 |
-
|
329 |
-
|
330 |
-
|
331 |
-
|
332 |
-
|
333 |
-
|
334 |
-
|
335 |
-
|
336 |
-
|
337 |
-
|
338 |
-
|
339 |
-
|
340 |
-
|
341 |
-
|
342 |
-
|
343 |
-
**generation_kwargs,
|
344 |
-
)
|
345 |
-
end = time()
|
346 |
-
else:
|
347 |
-
# print("Using GSM8K")
|
348 |
-
stop_watch = StopWatch(self.tokenizer)
|
349 |
-
start = time()
|
350 |
-
res = self.model.generate(
|
351 |
-
input_ids=context,
|
352 |
-
max_length=max_length,
|
353 |
-
eos_token_id=stop,
|
354 |
-
pad_token_id=self.tokenizer.pad_token_id,
|
355 |
-
use_cache=True,
|
356 |
-
streamer=stop_watch,
|
357 |
-
**generation_kwargs,
|
358 |
-
)
|
359 |
-
end = time()
|
360 |
|
361 |
batch_size = context.shape[0]
|
362 |
output_length = stop_watch.decoding_iterations
|
@@ -498,15 +481,18 @@ class HFLMWithMeasurement(HFLM):
|
|
498 |
f"Expected `kwargs` to be of type `dict` but got {type(gen_kwargs)}"
|
499 |
)
|
500 |
# add EOS token to stop sequences
|
501 |
-
eos =
|
502 |
if not until:
|
503 |
until = [eos]
|
504 |
else:
|
505 |
until.append(eos)
|
506 |
|
507 |
-
is_gsm8k = kwargs.get("is_gsm8k", False)
|
508 |
-
if is_gsm8k:
|
509 |
-
|
|
|
|
|
|
|
510 |
|
511 |
if "max_gen_toks" in kwargs.keys():
|
512 |
max_gen_toks = kwargs.pop("max_gen_toks")
|
@@ -532,8 +518,8 @@ class HFLMWithMeasurement(HFLM):
|
|
532 |
context_enc = context_enc.to(self.device)
|
533 |
attn_masks = attn_masks.to(self.device)
|
534 |
|
535 |
-
if "
|
536 |
-
kwargs["
|
537 |
|
538 |
# perform batched generation
|
539 |
cont, end_to_end_time, prefilling_time, token_per_sec, mfu, mbu = self._model_generate(
|
@@ -551,17 +537,16 @@ class HFLMWithMeasurement(HFLM):
|
|
551 |
cont_toks = cont_toks[context_enc.shape[1] :]
|
552 |
|
553 |
s = self.tok_decode(cont_toks)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
554 |
|
555 |
# print(s)
|
556 |
-
|
557 |
-
# use secondary stop seqs to cut off should-have-been-stopped content post-hoc
|
558 |
-
if not is_gsm8k:
|
559 |
-
for term in until:
|
560 |
-
if len(term) > 0:
|
561 |
-
# ignore '' separator,
|
562 |
-
# for seq2seq case where self.tok_decode(self.eot_token_id) = ''
|
563 |
-
s = s.split(term)[0]
|
564 |
-
|
565 |
res.append((s, end_to_end_time, prefilling_time, token_per_sec, mfu, mbu))
|
566 |
|
567 |
self.cache_hook.add_partial("generate_until", (context, gen_kwargs), s)
|
|
|
294 |
|
295 |
return re_ord.get_original(res)
|
296 |
|
297 |
+
def _model_generate(self, context, max_tokens, stop, **generation_kwargs):
|
298 |
# temperature = 0.0 if not set
|
299 |
# if do_sample is false and temp==0.0:
|
300 |
# remove temperature, as do_sample=False takes care of this
|
|
|
302 |
generation_kwargs["temperature"] = generation_kwargs.get("temperature", 0.0)
|
303 |
do_sample = generation_kwargs.get("do_sample", None)
|
304 |
|
305 |
+
# is_gsm8k = generation_kwargs.get("is_gsm8k", False)
|
306 |
|
307 |
# The temperature has to be a strictly positive float -- if it is 0.0, use greedy decoding strategies
|
308 |
if generation_kwargs.get("temperature") == 0.0 and do_sample is None:
|
|
|
311 |
if do_sample is False and generation_kwargs.get("temperature") == 0.0:
|
312 |
generation_kwargs.pop("temperature")
|
313 |
|
314 |
+
# if is_gsm8k:
|
315 |
+
# generation_kwargs.pop("is_gsm8k")
|
316 |
|
317 |
context_length = context.shape[1]
|
318 |
model_config = self.model.config
|
319 |
+
|
320 |
if not self.precision:
|
321 |
if model_config.quantization_config._load_in_4bit:
|
322 |
self.precision = "4bit"
|
|
|
325 |
else:
|
326 |
raise ValueError("Unknown precision")
|
327 |
|
328 |
+
stopping_criteria = stop_sequences_criteria(
|
329 |
+
self.tokenizer, stop, context.shape[1], context.shape[0]
|
330 |
+
)
|
331 |
+
stop_watch = StopWatch(self.tokenizer)
|
332 |
+
start = time()
|
333 |
+
res = self.model.generate(
|
334 |
+
input_ids=context,
|
335 |
+
max_new_tokens=max_tokens,
|
336 |
+
stopping_criteria=stopping_criteria,
|
337 |
+
pad_token_id=self.tokenizer.pad_token_id,
|
338 |
+
use_cache=True,
|
339 |
+
streamer=stop_watch,
|
340 |
+
**generation_kwargs,
|
341 |
+
)
|
342 |
+
end = time()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
343 |
|
344 |
batch_size = context.shape[0]
|
345 |
output_length = stop_watch.decoding_iterations
|
|
|
481 |
f"Expected `kwargs` to be of type `dict` but got {type(gen_kwargs)}"
|
482 |
)
|
483 |
# add EOS token to stop sequences
|
484 |
+
eos = "<|eot_id|>"
|
485 |
if not until:
|
486 |
until = [eos]
|
487 |
else:
|
488 |
until.append(eos)
|
489 |
|
490 |
+
# is_gsm8k = kwargs.get("is_gsm8k", False)
|
491 |
+
# if is_gsm8k:
|
492 |
+
# until = ["Question:", "Question", "</s>"]
|
493 |
+
# eos_ids = [self.tokenizer.eos_token_id,
|
494 |
+
# self.tokenizer.convert_tokens_to_ids("<|eot_id|>")]
|
495 |
+
|
496 |
|
497 |
if "max_gen_toks" in kwargs.keys():
|
498 |
max_gen_toks = kwargs.pop("max_gen_toks")
|
|
|
518 |
context_enc = context_enc.to(self.device)
|
519 |
attn_masks = attn_masks.to(self.device)
|
520 |
|
521 |
+
if "max_tokens" not in kwargs:
|
522 |
+
kwargs["max_tokens"] = max_gen_toks
|
523 |
|
524 |
# perform batched generation
|
525 |
cont, end_to_end_time, prefilling_time, token_per_sec, mfu, mbu = self._model_generate(
|
|
|
537 |
cont_toks = cont_toks[context_enc.shape[1] :]
|
538 |
|
539 |
s = self.tok_decode(cont_toks)
|
540 |
+
|
541 |
+
# # use secondary stop seqs to cut off should-have-been-stopped content post-hoc
|
542 |
+
# if not is_gsm8k:
|
543 |
+
for term in until:
|
544 |
+
if len(term) > 0:
|
545 |
+
# ignore '' separator,
|
546 |
+
# for seq2seq case where self.tok_decode(self.eot_token_id) = ''
|
547 |
+
s = s.split(term)[0]
|
548 |
|
549 |
# print(s)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
550 |
res.append((s, end_to_end_time, prefilling_time, token_per_sec, mfu, mbu))
|
551 |
|
552 |
self.cache_hook.add_partial("generate_until", (context, gen_kwargs), s)
|
src/backend/tasks/gsm8k/gsm8k-custom.yaml
CHANGED
@@ -22,18 +22,21 @@ metric_list:
|
|
22 |
- "\\.$"
|
23 |
generation_kwargs:
|
24 |
until:
|
25 |
-
- "
|
|
|
|
|
|
|
26 |
do_sample: false
|
27 |
temperature: 0.0
|
28 |
-
is_gsm8k: true
|
29 |
repeats: 1
|
30 |
num_fewshot: 5
|
31 |
filter_list:
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
- name: "flexible-extract"
|
38 |
filter:
|
39 |
- function: "regex"
|
|
|
22 |
- "\\.$"
|
23 |
generation_kwargs:
|
24 |
until:
|
25 |
+
- "Question:"
|
26 |
+
- "Question"
|
27 |
+
- "</s>"
|
28 |
+
- "<|im_end|>"
|
29 |
do_sample: false
|
30 |
temperature: 0.0
|
31 |
+
# is_gsm8k: true
|
32 |
repeats: 1
|
33 |
num_fewshot: 5
|
34 |
filter_list:
|
35 |
+
- name: "strict-match"
|
36 |
+
filter:
|
37 |
+
- function: "regex"
|
38 |
+
regex_pattern: "#### (\\-?[0-9\\.\\,]+)"
|
39 |
+
- function: "take_first"
|
40 |
- name: "flexible-extract"
|
41 |
filter:
|
42 |
- function: "regex"
|