hf_text-generation-inference/load_tests
Nicolas Patry 5df8059037
Auto max prefill (#2797)
* Attempt at automatic max batch prefill.

* Taking into account number of shards.

* Adding more cards.

* Adding A100 + H100

* Adding a few more cards.

* Logprobs cost too much.

* h100 better name, and keep factor of 2

* Damn inflated sparse tflops.

* Typo in h100.

* Updated the flops calculation (checked with fvcore).

* chunking by default.

* Fix prefix caching for chat completion since we removed logprobs.

* More tests.

* Dropping all the prefill logprobs.

* Add a flag that enables users to get logprobs back.

* Repairing prompt token counting.

* Fixing a few tests.

* Remove some scaffolding.

* Attempting to reduces the issues (workarounds for now).
2024-12-06 05:52:00 +01:00
..
Makefile Auto max prefill (#2797) 2024-12-06 05:52:00 +01:00
benchmarks.py chore: prepare 2.4.1 release (#2773) 2024-11-22 17:26:15 +00:00
common.js Auto max prefill (#2797) 2024-12-06 05:52:00 +01:00
filter.py Auto max prefill (#2797) 2024-12-06 05:52:00 +01:00
long.js Auto max prefill (#2797) 2024-12-06 05:52:00 +01:00
long.py Auto max prefill (#2797) 2024-12-06 05:52:00 +01:00
long_prompt2.py Auto max prefill (#2797) 2024-12-06 05:52:00 +01:00
orca.py Auto max prefill (#2797) 2024-12-06 05:52:00 +01:00
poetry.lock feat: Add automatic nightly benchmarks (#2591) 2024-11-21 17:11:42 +00:00
pyproject.toml chore: prepare 2.4.1 release (#2773) 2024-11-22 17:26:15 +00:00