Fixing top_n_tokens. (#1497)

# What does this PR do?

Superseeds #1459

The fix works as follows.
We updated next_token_chooser to return all logprbs, then
batch_top_n_tokens, now also gets accepted_ids + speculated_length (so
it knows how to interpret the flat logprobs).

We then update the code to return lists ot `Tokens` that it expects.
<!--
Congratulations! You've made it this far! You're not quite done yet
though.

Once merged, your PR is going to appear in the release notes with the
title you set, so make sure it's a great title that fully reflects the
extent of your awesome contribution.

Then, please replace this with a description of the change and which
issue is fixed (if applicable). Please also include relevant motivation
and context. List any dependencies (if any) that are required for this
change.

Once you're done, someone will review your PR shortly (see the section
"Who can review?" below to tag some potential reviewers). They may
suggest changes to make the code even better. If no one reviewed your PR
after a week has passed, don't hesitate to post a new comment
@-mentioning the same persons---sometimes notifications get lost.
-->

<!-- Remove if not applicable -->

Fixes # (issue)


## Before submitting
- [ ] This PR fixes a typo or improves the docs (you can dismiss the
other checks if that's the case).
- [ ] Did you read the [contributor
guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#start-contributing-pull-requests),
      Pull Request section?
- [ ] Was this discussed/approved via a Github issue or the
[forum](https://discuss.huggingface.co/)? Please add a link
      to it if that's the case.
- [ ] Did you make sure to update the documentation with your changes?
Here are the
[documentation
guidelines](https://github.com/huggingface/transformers/tree/main/docs),
and
[here are tips on formatting
docstrings](https://github.com/huggingface/transformers/tree/main/docs#writing-source-documentation).
- [ ] Did you write any new necessary tests?


## Who can review?

Anyone in the community is free to review the PR once the tests have
passed. Feel free to tag
members/contributors who may be interested in your PR.

<!-- Your PR will be replied to more quickly if you can figure out the
right person to tag with @


@OlivierDehaene OR @Narsil

 -->
This commit is contained in:
Nicolas Patry 2024-01-26 20:13:47 +01:00 committed by GitHub
parent c2d4a3b5c7
commit 069895b985
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
6 changed files with 142 additions and 76 deletions

View File

@ -50,19 +50,39 @@ def test_batch_top_tokens():
top_n_tokens = [0, 2, 3, 4, 5]
top_n_tokens_tensor = torch.tensor(top_n_tokens)
inp_logprobs = torch.tensor([[-1.0, -3.0, -4.0, -2.0, -3.0]] * 5)
accepted_ids = torch.ones_like(top_n_tokens_tensor)
topn_tok_ids, topn_tok_logprobs = batch_top_tokens(
top_n_tokens, top_n_tokens_tensor, inp_logprobs
top_n_tokens, top_n_tokens_tensor, inp_logprobs, accepted_ids
)
assert topn_tok_ids[0] == []
assert topn_tok_ids[1] == [0, 3]
assert topn_tok_ids[2] == [0, 3, 1, 4]
assert topn_tok_ids[3] == [0, 3, 1, 4]
assert topn_tok_ids[4] == [0, 3, 1, 4, 2]
assert topn_tok_ids[0] == [[]]
assert topn_tok_ids[1] == [[0, 3]]
assert topn_tok_ids[2] == [[0, 3, 1, 4]]
assert topn_tok_ids[3] == [[0, 3, 1, 4]]
assert topn_tok_ids[4] == [[0, 3, 1, 4, 2]]
assert topn_tok_logprobs[0] == []
assert topn_tok_logprobs[1] == [-1, -2]
assert topn_tok_logprobs[2] == [-1, -2, -3, -3]
assert topn_tok_logprobs[3] == [-1, -2, -3, -3]
assert topn_tok_logprobs[4] == [-1, -2, -3, -3, -4]
assert topn_tok_logprobs[0] == [[]]
assert topn_tok_logprobs[1] == [[-1, -2]]
assert topn_tok_logprobs[2] == [[-1, -2, -3, -3]]
assert topn_tok_logprobs[3] == [[-1, -2, -3, -3]]
assert topn_tok_logprobs[4] == [[-1, -2, -3, -3, -4]]
# Now let's make second member of the batch be speculated
inp_logprobs = torch.tensor([[-1.0, -3.0, -4.0, -2.0, -3.0]] * 5 * 2)
accepted_ids[1] = 2
topn_tok_ids, topn_tok_logprobs = batch_top_tokens(
top_n_tokens, top_n_tokens_tensor, inp_logprobs, accepted_ids
)
assert topn_tok_ids[0] == [[]]
assert topn_tok_ids[1] == [[0, 3], [0, 3]]
assert topn_tok_ids[2] == [[0, 3, 1, 4]]
assert topn_tok_ids[3] == [[0, 3, 1, 4]]
assert topn_tok_ids[4] == [[0, 3, 1, 4, 2]]
assert topn_tok_logprobs[0] == [[]]
assert topn_tok_logprobs[1] == [[-1, -2], [-1, -2]]
assert topn_tok_logprobs[2] == [[-1, -2, -3, -3]]
assert topn_tok_logprobs[3] == [[-1, -2, -3, -3]]
assert topn_tok_logprobs[4] == [[-1, -2, -3, -3, -4]]

View File

@ -580,10 +580,13 @@ class CausalLM(Model):
generations: List[Generation] = []
stopped = True
# Speculation is not active for causal
accepted_ids = torch.ones_like(batch.input_ids)[:, 0]
batch_top_token_ids, batch_top_token_logprobs = batch_top_tokens(
batch.top_n_tokens,
batch.top_n_tokens_tensor,
torch.log_softmax(logits[:, -1], -1),
accepted_ids,
)
start_decode = time.time_ns()
@ -692,6 +695,8 @@ class CausalLM(Model):
prefill_tokens = None
if top_n_tokens > 0:
all_top_tokens = []
for (top_token_ids, top_token_logprobs) in zip(top_token_ids, top_token_logprobs):
toptoken_texts = self.tokenizer.batch_decode(
top_token_ids,
clean_up_tokenization_spaces=False,
@ -706,6 +711,8 @@ class CausalLM(Model):
toptoken_texts,
special_toptokens,
)
all_top_tokens.append(top_tokens)
top_tokens = all_top_tokens
else:
top_tokens = None

View File

@ -842,6 +842,8 @@ class FlashCausalLM(Model):
else:
next_token_logits = out
speculate = get_speculate()
(
next_input_ids,
next_token_logprobs,
@ -851,16 +853,15 @@ class FlashCausalLM(Model):
) = batch.next_token_chooser(
batch.all_input_ids_tensor[:, : batch.max_seqlen],
next_token_logits,
get_speculate(),
speculate,
batch.speculative_ids,
speculative_logits,
)
batch_top_token_ids, batch_top_token_logprobs = batch_top_tokens(
batch.top_n_tokens, batch.top_n_tokens_tensor, logprobs
batch.top_n_tokens, batch.top_n_tokens_tensor, logprobs, accepted_ids
)
speculative_length = 0 if speculative_ids is None else speculative_ids.shape[1]
if prefill:
if len(batch) > 1 and prefill_logprobs:
# We create the prefill_tokens_indices tensor that will be used to gather prefill logprobs
@ -1062,6 +1063,8 @@ class FlashCausalLM(Model):
prefill_tokens = None
if top_n_tokens > 0:
all_top_tokens = []
for (top_token_ids, top_token_logprobs) in zip(top_token_ids, top_token_logprobs):
toptoken_texts = self.tokenizer.batch_decode(
top_token_ids,
clean_up_tokenization_spaces=False,
@ -1076,6 +1079,8 @@ class FlashCausalLM(Model):
toptoken_texts,
special_toptokens,
)
all_top_tokens.append(top_tokens)
top_tokens = all_top_tokens
else:
top_tokens = None

View File

@ -640,10 +640,13 @@ class Seq2SeqLM(Model):
batch.past_key_values,
)
# Speculation is not active for seq2seq
accepted_ids = torch.ones_like(batch.decoder_input_ids)[:, 0]
batch_top_token_ids, batch_top_token_logprobs = batch_top_tokens(
batch.top_n_tokens,
batch.top_n_tokens_tensor,
torch.log_softmax(logits[:, -1], -1),
accepted_ids,
)
start_decode = time.time_ns()
@ -746,6 +749,8 @@ class Seq2SeqLM(Model):
prefill_tokens = None
if top_n_tokens > 0:
all_top_tokens = []
for (top_token_ids, top_token_logprobs) in zip(top_token_ids, top_token_logprobs):
toptoken_texts = self.tokenizer.batch_decode(
top_token_ids,
clean_up_tokenization_spaces=False,
@ -760,6 +765,8 @@ class Seq2SeqLM(Model):
toptoken_texts,
special_toptokens,
)
all_top_tokens.append(top_tokens)
top_tokens = all_top_tokens
else:
top_tokens = None

View File

@ -95,5 +95,5 @@ class Generation:
generated_text=self.generated_text.to_pb()
if self.generated_text is not None
else None,
top_tokens=self.top_tokens.to_pb() if self.top_tokens is not None else None,
top_tokens=[top_tokens.to_pb() for top_tokens in self.top_tokens] if self.top_tokens is not None else None,
)

View File

@ -277,7 +277,8 @@ class HeterogeneousNextTokenChooser:
scores[:, j] = _scores
next_ids[:, j] = _next_ids
next_ids = next_ids.view(B * S)
scores = scores.view(B * S, -1)
allscores = scores.view(B * S, -1)
alllogprobs = torch.log_softmax(allscores, -1)
if speculated_ids is not None:
accepted_ids = []
@ -305,16 +306,17 @@ class HeterogeneousNextTokenChooser:
accepted_ids, device=input_ids.device, dtype=input_ids.dtype
)
next_ids = next_ids[indices]
scores = scores[indices]
logprobs = alllogprobs[indices]
indices = torch.arange(B, device=input_ids.device) * S
if speculative_scores is not None:
speculative_scores = speculative_scores[indices + accepted_ids - 1]
else:
accepted_ids = torch.ones_like(next_ids)
logprobs = alllogprobs
logprobs = torch.log_softmax(scores, -1)
next_logprobs = torch.gather(logprobs, 1, next_ids.view(-1, 1)).view(-1)
if speculate > 0:
if speculative_scores is not None:
# Medusa provided some scores
@ -327,7 +329,7 @@ class HeterogeneousNextTokenChooser:
else:
speculative_ids = None
return next_ids, next_logprobs, logprobs, accepted_ids, speculative_ids
return next_ids, next_logprobs, alllogprobs, accepted_ids, speculative_ids
def filter(self, indices):
if self.watermark_processor is not None:
@ -436,8 +438,8 @@ class HeterogeneousSampling:
def batch_top_tokens(
top_n_tokens: List[int], top_n_tokens_tensor: torch.Tensor, logprobs: torch.Tensor
) -> Tuple[List[List[int]], List[List[float]]]:
top_n_tokens: List[int], top_n_tokens_tensor: torch.Tensor, logprobs: torch.Tensor, accepted_ids: torch.Tensor
) -> Tuple[List[List[List[int]]], List[List[List[float]]]]:
"""Find the top n most likely tokens for a batch of generations.
When multiple tokens have equal probabilities and they don't all fit, the
@ -446,14 +448,19 @@ def batch_top_tokens(
max_top_n = max(top_n_tokens)
# Early exit when top_n_tokens is not used
if max_top_n == 0:
return [[]] * len(top_n_tokens), [[]] * len(top_n_tokens)
return [[[]]] * len(top_n_tokens), [[[]]] * len(top_n_tokens)
batch_size = accepted_ids.shape[0]
speculate_size = logprobs.shape[0] // batch_size
top_n_tokens_tensor = top_n_tokens_tensor.repeat_interleave(speculate_size)
# Ensure top_n doesn't exceed vocab size
top_n_tokens = [min(tok, logprobs.size(-1)) for tok in top_n_tokens]
top_n_tokens = [min(tok, logprobs.size(-1)) for tok in top_n_tokens for _ in range(speculate_size)]
# Parallel kthvalue adapted from https://discuss.pytorch.org/t/how-to-efficiently-get-the-k-th-largest-values-in-parallel/160529/2
# Sorted topk is faster than torch.sort() since we only need a small subset
sorted_top_k = torch.topk(logprobs, k=max_top_n, dim=1, sorted=True).values
sorted_top_k = torch.topk(logprobs, k=max_top_n, dim=-1, sorted=True).values
nth_highest = torch.gather(
sorted_top_k, 1, (top_n_tokens_tensor - 1).clip(min=0).unsqueeze(1)
)
@ -471,13 +478,33 @@ def batch_top_tokens(
top_indices = top_k.indices.tolist()
top_values = top_k.values.tolist()
return (
[
idxs[:n] if req_n > 0 else []
for idxs, n, req_n in zip(top_indices, top_n_ishes, top_n_tokens)
],
[
vals[:n] if req_n > 0 else []
for vals, n, req_n in zip(top_values, top_n_ishes, top_n_tokens)
],
)
batch_top_token_ids = []
batch_top_token_logprobs = []
accepted_ids_list = accepted_ids.tolist()
for i, n_accepted_ids in enumerate(accepted_ids_list):
start = speculate_size * i
stop = speculate_size * (i + 1)
_top_indices = top_indices[start: stop]
_top_values = top_values[start: stop]
_top_n_ishes = top_n_ishes[start: stop]
_top_n_tokens = top_n_tokens[start: stop]
_top_indices = _top_indices[:n_accepted_ids]
_top_values = _top_values[:n_accepted_ids]
_top_n_ishes = _top_n_ishes[:n_accepted_ids]
_top_n_tokens = _top_n_tokens[:n_accepted_ids]
row_top_token_ids = []
row_top_token_logprobs = []
for idxs, vals, n, req_n in zip(_top_indices, _top_values, _top_n_ishes, _top_n_tokens):
indices = idxs[:n] if req_n > 0 else []
values = vals[:n] if req_n > 0 else []
row_top_token_ids.append(indices)
row_top_token_logprobs.append(values)
batch_top_token_ids.append(row_top_token_ids)
batch_top_token_logprobs.append(row_top_token_logprobs)
return batch_top_token_ids, batch_top_token_logprobs