huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-09-12 04:44:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add comment for why slot 0 is OK.

Browse Source
This commit is contained in:
Nicolas Patry 2024-09-10 11:46:45 +02:00
parent 2881edb3d7
commit ed51bb94ce
No known key found for this signature in database
GPG Key ID: 64AF4752B2967863
1 changed files with 3 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
server/text_generation_server/models/flash_causal_lm.py
View File

@ -1520,6 +1520,9 @@ class FlashCausalLM(Model):
) )
# assert block_tables.shape[0] >= slots.shape[0] # assert block_tables.shape[0] >= slots.shape[0]
cuda_graph["block_tables"][: block_tables.shape[0]] = block_tables cuda_graph["block_tables"][: block_tables.shape[0]] = block_tables
# XXX: This is working only because block 0 is reserved for the healthcheck
# so it doesn't matter if we override it with bogus values.
cuda_graph["slots"].fill_(0) cuda_graph["slots"].fill_(0)
cuda_graph["slots"][: slots.shape[0]] = slots cuda_graph["slots"][: slots.shape[0]] = slots
cuda_graph["input_lengths"].zero_() cuda_graph["input_lengths"].zero_()