Universal Assisted Generation: Assisted generation with any assistant model (by Intel Labs) #33383
Update candidate_generator.py
e5c6bdb3
Update utils.py
e02174bd
add lookbehind params to _get_candidate_generator
9a43708e
make fixup
50fa7ea5
add unit tests
f2664256
danielkorat
changed the title Speculative Decoding with any draft model (by Intel AI) Speculative Decoding with any draft model (by Intel Labs) 1 year ago
fix failing tests
0d1b1e90
add docstrings
865f3d80
fix docstrings; remove non-optimized AnyTokenizer
3a4b2432
danielkorat
marked this pull request as ready for review 1 year ago
danielkorat
changed the title Speculative Decoding with any draft model (by Intel Labs) Speculative Decoding with any assistant model (by Intel Labs) 1 year ago
added any tokenizer generation correctness test
97c16941
Merge pull request #2 from danielkorat/sd-any-tokenizer-generate-test…
2b5e641b
Merge branch 'huggingface:main' into sd-any-tokenizer
ff3fc819
make fixup
822cae78
fix assertion syntax
8a811514
Merge branch 'huggingface:main' into sd-any-tokenizer
a15d9ddb
gante
commented
on 2024-09-20
PR review fixes
ae03d6e3
address additional PR comments
afdeb1bb
fix tests
f2d58324
remove stropping criteria arg
509bd2f8
make fixup
209f3993
danielkorat
changed the title Speculative Decoding with any assistant model (by Intel Labs) Assisted Generation with any assistant model (by Intel Labs) 1 year ago
danielkorat
changed the title Assisted Generation with any assistant model (by Intel Labs) Assisted generation with any assistant model (by Intel Labs) 1 year ago
add AssistantConfig
d395c24e
fix prev_tokens branching
9796f308
pass tokenizers through `generate()`kwargs
3d88817b
fix lookbehind values; tokenizer params WIP
d085eabe
fixup
596c4bf4
AssistantConfig
a1f45840
gante
commented
on 2024-10-04
remove AssistantConfig; apply PR suggestions
cb46b85a
restructure tests
3d121f4b
fixup
657d5db6
fix assistant_tokenizer arg validation
545fa33b
Merge branch 'main' into sd-any-tokenizer
1812bb96
fixup
b2dbebe7
fix tests in TestAssistedCandidateGeneratorDifferentTokenizers
552cf314
danielkorat
changed the title Assisted generation with any assistant model (by Intel Labs) Universal Assisted Generation: Assisted generation with any assistant model (by Intel Labs) 1 year ago
gante
approved these changes
on 2024-10-07
Merge branch 'huggingface:main' into sd-any-tokenizer
33ce386b
fix class docstring
c6ea99c2
PR suggestions
65528a15
doc
6eddc608
doc update and improvements to `_validate_assistant()`
c29ff190
Merge branch 'huggingface:main' into sd-any-tokenizer
81d63062
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub