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
..
test_flash_llama_grammar_json.json	Auto max prefill (#2797)	2024-12-06 05:52:00 +01:00
test_flash_llama_grammar_load.json	Auto max prefill (#2797)	2024-12-06 05:52:00 +01:00
test_flash_llama_grammar_regex.json	Auto max prefill (#2797)	2024-12-06 05:52:00 +01:00
test_flash_llama_grammar_single_load_instance.json	fix: correctly index into mask when applying grammar (#1618)	2024-03-01 18:22:01 +01:00
test_flash_llama_grammar.json	Auto max prefill (#2797)	2024-12-06 05:52:00 +01:00