…uantization.

Merge the warpper modification with the v1

…gs and pages; fix on the previous quantization implementaion, with lanuch_graph dtype set to the quant type

… (naive sparse attention, flash sparse attention, flashmoba)

- Introduced a comprehensive benchmarking suite for TopK kernel variants, measuring kernel-level latency.
- Added scripts for offline calibration of TopK mapping modes, including:# 0: None           — original fp16 bit-pattern bucketing
# 1: LUT CDF        — LUT-based CDF equalization (calibrated)
# 2: Quantile       — piecewise-linear quantile mapping (calibrated)
# 3: Power          — y = sign(x) * |x|^p
# 4: Log            — y = sign(x) * log(|x| + 1)
# 5: Index Cache    — reuse previous layer's indices
# 6: Asinh          — y = asinh(beta * x)
# 7: Log1p          — y = sign(x) * log1p(alpha * |x|)
# 8: Trunc8         — bf16 upper-8-bit bucketing
-  Adding various remap functions for the bucket sort in sglang topk kernel, with evaluation and visualization scripts.
- Implemented analysis tools for TopK distribution profiling.

- Removed outdated GPU architecture flags from setup.py.
- Added new mapping modes (Erf, Tanh, Subtract) to analyze_topk_distribution.py and bench_topk.py.
- Updated functions to handle new modes and added support for noscale parameters in autotune and benchmark scripts.
- Enhanced the TopK kernel with additional profiling metrics and improved handling of kernel arguments.
- Updated example scripts to reflect new modes and parameters for distribution analysis.

…cripts to reflect changes in histogram calibration and TopK mapping parameters.

- Added ExpStretch and TopkWindow modes to analyze_topk_distribution.py and bench_topk.py.
- Introduced topk_output_sglang_ori function for original sglang kernel in vortex_torch_C.
- Updated autotune and benchmark scripts to include new modes and original kernel.
- Modified example scripts to reflect changes in histogram calibration and TopK mapping parameters.

- Added new  file for enhanced profiling capabilities.
- Updated  to include the new profiling source file.
- Modified  to expand the sweep grid for parameter tuning.
- Refactored  to improve handling of hyperparameters and added subprocess profiling for large TopK values.
- Enhanced  and  to support new parameters for profiling.
- Updated example scripts to reflect changes in TopK parameters and profiling options.

… and usability

- Updated autotune_topk_mapping.py to optimize hyperparameter tuning based on kernel latency.
- Simplified the sweep grid and improved documentation for usage.
- Enhanced bench_topk.py to expose public helpers and added CLI modes for benchmarking.
- Introduced new remap functions and improved kernel integration for profiling.
- Added watchdog timeout option in calibrate_topk.py for SGLang scheduler.
- Removed outdated greedy_layer_search.py as part of code cleanup.

… and usability

- Updated autotune_topk_mapping.py to optimize hyperparameter tuning based on kernel latency.
- Simplified the sweep grid and improved documentation for usage.
- Enhanced bench_topk.py to expose public helpers and added CLI modes for benchmarking.
- Introduced new remap functions and improved kernel integration for profiling.
- Added watchdog timeout option in calibrate_topk.py for SGLang scheduler.
- Removed outdated greedy_layer_search.py as part of code cleanup.

…kernel in vortex_torch_C.

- Updated setup.py to include the new source file for the original kernel.
- Enhanced autotune_topk_mapping.py and bench_topk.py to support new mapping modes and original kernel integration.
- Expanded the sweep grid in autotune_topk_mapping.py for improved hyperparameter tuning.
- Added a new command-line argument in calibrate_topk.py for maximum total tokens to manage KV pool size.
- Removed outdated remap_function_bench.sh script as part of code cleanup.