huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-07-04 06:50:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'habana-main' into 2.3.0

Browse Source
This commit is contained in:
Yuan Wu 2024-11-28 12:38:36 +08:00 committed by GitHub
commit b83419a769
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 5 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
server/requirements.txt
View File

@ -74,7 +74,6 @@ six==1.16.0 ; python_version >= "3.9" and python_version < "3.13"
sympy==1.12.1 ; python_version >= "3.9" and python_version < "3.13" sympy==1.12.1 ; python_version >= "3.9" and python_version < "3.13"
threadpoolctl==3.5.0 ; python_version >= "3.9" and python_version < "3.13" threadpoolctl==3.5.0 ; python_version >= "3.9" and python_version < "3.13"
tokenizers==0.20.1 ; python_version >= "3.9" and python_version < "3.13" tokenizers==0.20.1 ; python_version >= "3.9" and python_version < "3.13"
torch==2.4.0a0+git74cd574 ; python_version >= "3.9" and python_version < "3.13"
tqdm==4.66.5 ; python_version >= "3.9" and python_version < "3.13" tqdm==4.66.5 ; python_version >= "3.9" and python_version < "3.13"
transformers==4.45.2 ; python_version >= "3.9" and python_version < "3.13" transformers==4.45.2 ; python_version >= "3.9" and python_version < "3.13"
transformers[sentencepiece]==4.45.2 ; python_version >= "3.9" and python_version < "3.13" transformers[sentencepiece]==4.45.2 ; python_version >= "3.9" and python_version < "3.13"

7
server/text_generation_server/models/causal_lm.py
View File

@ -494,9 +494,12 @@ class CausalLMBatch(Batch):
inputs.append(concat_text_chunks(r.input_chunks.chunks)) inputs.append(concat_text_chunks(r.input_chunks.chunks))
top_n_tokens.append(r.top_n_tokens) top_n_tokens.append(r.top_n_tokens)
max_truncation = max(max_truncation, r.truncate) max_truncation = max(max_truncation, r.truncate)
max_new_tokens = max(r.stopping_criteria.max_new_tokens for r in requests)
max_input_length = max_truncation max_input_length = max_truncation
if max_input_length < PAD_SEQUENCE_TO_MULTIPLE_OF:
max_input_length = PAD_SEQUENCE_TO_MULTIPLE_OF
max_new_tokens = max(r.stopping_criteria.max_new_tokens for r in requests)
# TODO: by tokenizing all inputs at once we loose information on actual input lengths # TODO: by tokenizing all inputs at once we loose information on actual input lengths
# this means that we cannot shift inputs to the left after a long input sequence # this means that we cannot shift inputs to the left after a long input sequence
# was filtered out # was filtered out