Four distinct strategies are implemented to accommodate different distributed tuning scenarios, including BROADCAST, INDEPENDENT, MERGE, PARALLEL.
* Distributed tuning is disabled by default, with the INDEPENDENT strategy as the fallback. This conservative approach prevents unexpected behavior in standard use cases.
* Only operations with significant tuning time overhead have been assigned the PARALLEL strategy, which allows the same tensor parallelism (TP) rank to tune tactics concurrently across different ranks. This targeted approach balances performance gains with stability.
* Operations with nested tuning structures, such as NVFP4GemmUnifiedRunner, currently support only the INDEPENDENT strategy. This restriction exists because the synchronization mechanism is optimized only for leaf operations and doesn't yet handle nested hierarchies.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
Restrict tactic types to those compatible with AutoTuner cache serialization and deserialization.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
Signed-off-by: Shijie Wang <jaywan@nvidia.com>
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
Signed-off-by: Shijie <jaywan@nvidia.com>
Co-authored-by: Yukun He <23156053+hyukn@users.noreply.github.com>
The performance results of some kernels could be easily affected by the warm/cold L2 cache status. To achieve more precise profiling results, the L2 cache is cleared for every execution by the circular buffer method for better benchmarking during autotuning.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Skip the shape profile generating process if the profile has already been found in the cache under tuning mode. This is a prerequisite for nested autotuning because host overhead might be included during the profiling of the high-level op.
* Enable the profiling with CUDA graph as the default profiling method.
* Apply a heuristic method to cut off the number of repeat times of profiling according to a few-run time measurement.
Some tunable ops require a more realistic data distribution, for instance, a shape-associated tensor. Thus, a customizable pre-hook function can be declared in the tuning config to modify the input tensor before the tuning process.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
To achieve determinism for the AutoTuner profiling cache, serialization and deserialization are introduced to store the cache on disk in JSON format. Use TLLM_AUTOTUNER_CACHE_PATH to indicate the path where the cache file should be stored:
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Generalize the definition of tactics so that users can implement more customizable tactic types, making the configurations clearer for each kernel run.
* Allow the user not to specify the `gen_tuning_buckets` or the `map_to_tuning_buckets` function.
* Other code refactoring.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Change the fallback alert from DEBUG to WARNING level and only do it once.
* Add debug information for profiling cache right after the warmup phase.
* Change the level of exception message during tactic profiling from ERROR to WARNING level. All exception details are pushed to the DEBUG level.
* Other trivial refinements and cleanups.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
- Adds a new Python custom op (fp8_block_scale_moe_runner) and a FP8BlockScaleMoERunner class for autotuning.
- Updates C++ MoE and batched GEMM kernels to accept a configIndex for workspace sizing and execution.
- Extends the unit test to run both autotuned and non-autotuned code paths.
Signed-off-by: Dom Brown <3886319+DomBrown@users.noreply.github.com>
Fix AutoTuner warmup request generating.
* The current warmup phase creates one request, which is insufficient for the warmup to cover the max_num_tokens. Revise the warmup phase to a batch of requests to cover the max_num_tokens to eliminate potential fallback cases.
Refactor AutoTuner API and reduce host overhead.
Refine (min, opt, max) values of optimization profile setup for get_valid_tactics to achieve the correct canImplement definition.
* Refine cache key assembly process to reduce host overhead and simplify API.
* Fix lru_cache usage to reduce host overhead.
* Move tuning config initialization as a one-time object in tunable runner to reduce host overhead.
Improve tuning config readability.
* Use dataclass to define tuning config.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Several optimizations and fixings on the Autotuner.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Apply the new Python side Autotuner on current linear for nvFP4 data type.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Apply the new Python side Autotuner on MoE op
* Remove routers from cache key to improve inference perf
* Prevent unnecessary code profiling. Use do_preparation keyword to select which part should be executed during before evaluating any tactic.
* Remove try-catch inside moe profiling process.
* Move default tactic -1 to 0 transforms in cpp runner.
* Revise relavant tests.
* Predefined the bucketizing strategy for fused_moe
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Add specific_profile support for AutoTuner to bypass the standard cache search process for perf optimization
* Add specific_profile for moe
* Add specific profile for linear
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Fixing and revising according to reviewer's suggestions.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Use lru_cache for inference pref optimization.
* Revert gen_custom_cache_key feature
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Replace runner with runner id to achieve a serializable cache.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Code clean up and minor fixings.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Move all tunable runners and custom ops into torch_custom_ops.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Treat min_latency_mode as a independent dynamic tensor. Modify get_valid_tactics to suit for it.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
---------
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
