Skip to content

Optimize statistical unigram tokenizer decode_forward #63

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Optimize statistical unigram tokenizer decode_forward #63

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 13 additions & 3 deletions src/statistical/unigram.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,14 @@ structure, To hold unknown token index and map of vocabulary to log probability
struct SentencePieceModel
vocab_map::Dict{String, Tuple{Float64, Int}}
unk_id::Int
# used to make decoding more efficient
max_vocab_codeunit_len::Int

function SentencePieceModel(vocab_map, unk_id)
num_codeunits(idx) = isassigned(vocab_map.keys, idx) ? ncodeunits(vocab_map.keys[idx]) : 0
max_vocab_codeunit_len, _ = findmax(num_codeunits.(eachindex(vocab_map.keys)))
new(vocab_map, unk_id, max_vocab_codeunit_len)
end
end

"""
Expand Down Expand Up @@ -97,14 +105,16 @@ function decode_forward(sp::SentencePieceModel, text::String)
scores = fill(-Inf, lastindex(text))
scores[1] = 0
for char_end in eachindex(text)
for char_start in eachindex(text)
char_start > char_end && break
min_start = max(firstindex(text), char_end - sp.max_vocab_codeunit_len + 1)
candidate_text_substr = SubString(text, thisind(text, min_start), char_end)
for relative_idx in eachindex(candidate_text_substr)
char_start = candidate_text_substr.offset + relative_idx
subtoken = SubString(text, char_start:char_end)
if haskey(sp.vocab_map, subtoken)
subtokenid = sp.vocab_map[subtoken][2]
local_score = scores[char_start] + sp.vocab_map[subtoken][1]
if local_score > scores[char_end]
results[char_end] = Nodes(SubString(text, char_start:char_end), local_score, subtokenid, char_start, char_end)
results[char_end] = Nodes(subtoken, local_score, subtokenid, char_start, char_end)
scores[char_end] = local_score
end
end
Expand Down