From 68825ee894bbf52998573c0a584f981d5d6d87c6 Mon Sep 17 00:00:00 2001 From: tjtanaa Date: Tue, 25 Feb 2025 14:12:43 +0000 Subject: [PATCH] format code; add deepseekv3 tuning config Signed-off-by: tjtanaa --- .../kernels/benchmark_mixtral_moe_rocm.py | 1 + benchmarks/test_accuracy.py | 1 + tests/kernels/test_semi_structured.py | 1 + vllm/attention/ops/paged_attn_ater.py | 1 + vllm/entrypoints/fast_sync_llm.py | 5 +- vllm/entrypoints/sync_openai/api_server.py | 1 + ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ .../E=8,N=8192,device_name=MI308X.json | 41 ++--- ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ ...,dtype=fp8_w8a8,block_shape=[128,128].json | 164 ++++++++++++++++++ 32 files changed, 4123 insertions(+), 28 deletions(-) create mode 100644 vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=36864,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=7168,K=8192,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json create mode 100644 vllm/model_executor/layers/quantization/utils/configs/N=8192,K=1536,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json diff --git a/benchmarks/kernels/benchmark_mixtral_moe_rocm.py b/benchmarks/kernels/benchmark_mixtral_moe_rocm.py index d6b3e3bdf91b..48b06aa564f0 100644 --- a/benchmarks/kernels/benchmark_mixtral_moe_rocm.py +++ b/benchmarks/kernels/benchmark_mixtral_moe_rocm.py @@ -1,3 +1,4 @@ +# SPDX-License-Identifier: Apache-2.0 import argparse import json import os diff --git a/benchmarks/test_accuracy.py b/benchmarks/test_accuracy.py index 935aee779629..0e410c729588 100644 --- a/benchmarks/test_accuracy.py +++ b/benchmarks/test_accuracy.py @@ -1,3 +1,4 @@ +# SPDX-License-Identifier: Apache-2.0 import argparse import dataclasses diff --git a/tests/kernels/test_semi_structured.py b/tests/kernels/test_semi_structured.py index 4316d6ab30e3..b019c0bb986f 100644 --- a/tests/kernels/test_semi_structured.py +++ b/tests/kernels/test_semi_structured.py @@ -1,3 +1,4 @@ +# SPDX-License-Identifier: Apache-2.0 """Tests for sparse cutlass kernels Run `pytest tests/kernels/test_semi_structured.py`. diff --git a/vllm/attention/ops/paged_attn_ater.py b/vllm/attention/ops/paged_attn_ater.py index ba90d4e47326..f0939a3b5361 100644 --- a/vllm/attention/ops/paged_attn_ater.py +++ b/vllm/attention/ops/paged_attn_ater.py @@ -1,3 +1,4 @@ +# SPDX-License-Identifier: Apache-2.0 from dataclasses import dataclass from typing import List, Optional, Tuple diff --git a/vllm/entrypoints/fast_sync_llm.py b/vllm/entrypoints/fast_sync_llm.py index 3f09b148dc92..f1d74ea84194 100644 --- a/vllm/entrypoints/fast_sync_llm.py +++ b/vllm/entrypoints/fast_sync_llm.py @@ -1,3 +1,4 @@ +# SPDX-License-Identifier: Apache-2.0 import multiprocessing as mp from queue import Empty from typing import Union @@ -85,8 +86,8 @@ def run_engine(self): try: while True: poll_interval -= 1 - if (self.input_queue.qsize() >= - envs.VLLM_SYNC_SERVER_ACCUM_REQUESTS + if (self.input_queue.qsize() + >= envs.VLLM_SYNC_SERVER_ACCUM_REQUESTS or poll_interval <= 0 or not self.llm_engine.has_unfinished_requests()): self._poll_requests() diff --git a/vllm/entrypoints/sync_openai/api_server.py b/vllm/entrypoints/sync_openai/api_server.py index 1995e71a3ec4..e511305e4b68 100644 --- a/vllm/entrypoints/sync_openai/api_server.py +++ b/vllm/entrypoints/sync_openai/api_server.py @@ -1,3 +1,4 @@ +# SPDX-License-Identifier: Apache-2.0 import asyncio import multiprocessing import re diff --git a/vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..a5cc0d8ad2d6 --- /dev/null +++ b/vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=MI308X.json b/vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=MI308X.json index c07d5fe8e8ae..f246469d3b55 100644 --- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=MI308X.json +++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=MI308X.json @@ -1,7 +1,7 @@ { "1": { "BLOCK_SIZE_M": 16, - "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_N": 32, "BLOCK_SIZE_K": 128, "GROUP_SIZE_M": 1, "num_warps": 2, @@ -19,11 +19,11 @@ "num_stages": 2, "waves_per_eu": 0, "matrix_instr_nonkdim": 16, - "kpack": 2 + "kpack": 1 }, "4": { "BLOCK_SIZE_M": 16, - "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_N": 32, "BLOCK_SIZE_K": 128, "GROUP_SIZE_M": 1, "num_warps": 2, @@ -34,7 +34,7 @@ }, "8": { "BLOCK_SIZE_M": 16, - "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_N": 32, "BLOCK_SIZE_K": 128, "GROUP_SIZE_M": 1, "num_warps": 2, @@ -45,7 +45,7 @@ }, "16": { "BLOCK_SIZE_M": 16, - "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_N": 32, "BLOCK_SIZE_K": 128, "GROUP_SIZE_M": 1, "num_warps": 2, @@ -56,7 +56,7 @@ }, "24": { "BLOCK_SIZE_M": 16, - "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_N": 32, "BLOCK_SIZE_K": 128, "GROUP_SIZE_M": 1, "num_warps": 2, @@ -78,7 +78,7 @@ }, "48": { "BLOCK_SIZE_M": 16, - "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_N": 32, "BLOCK_SIZE_K": 128, "GROUP_SIZE_M": 1, "num_warps": 2, @@ -89,8 +89,8 @@ }, "64": { "BLOCK_SIZE_M": 32, - "BLOCK_SIZE_N": 32, - "BLOCK_SIZE_K": 256, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, "GROUP_SIZE_M": 4, "num_warps": 4, "num_stages": 2, @@ -100,8 +100,8 @@ }, "96": { "BLOCK_SIZE_M": 32, - "BLOCK_SIZE_N": 32, - "BLOCK_SIZE_K": 256, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, "GROUP_SIZE_M": 4, "num_warps": 4, "num_stages": 2, @@ -123,19 +123,8 @@ "256": { "BLOCK_SIZE_M": 64, "BLOCK_SIZE_N": 64, - "BLOCK_SIZE_K": 64, - "GROUP_SIZE_M": 4, - "num_warps": 4, - "num_stages": 2, - "waves_per_eu": 0, - "matrix_instr_nonkdim": 32, - "kpack": 2 - }, - "256": { - "BLOCK_SIZE_M": 128, - "BLOCK_SIZE_N": 128, - "BLOCK_SIZE_K": 64, - "GROUP_SIZE_M": 4, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, "num_warps": 8, "num_stages": 2, "waves_per_eu": 0, @@ -151,13 +140,13 @@ "num_stages": 2, "waves_per_eu": 0, "matrix_instr_nonkdim": 16, - "kpack": 2 + "kpack": 1 }, "1024": { "BLOCK_SIZE_M": 128, "BLOCK_SIZE_N": 128, "BLOCK_SIZE_K": 64, - "GROUP_SIZE_M": 1, + "GROUP_SIZE_M": 4, "num_warps": 8, "num_stages": 2, "waves_per_eu": 0, diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..f9e0e28af302 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..cc08fc0caa6c --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..9564050b5400 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..791542c1e492 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..4b65283dcc27 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..95f8649d6f72 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..cc68f7d066b1 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..701cf818e315 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..9c5a93aabbce --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=36864,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=36864,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..7476c0fb9cb1 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=36864,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..307ce7e8ef71 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..bebcc18d2619 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..6fd0271a2470 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..ff144de6d5fe --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..fd4cc3934f46 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..0f5a96a936a3 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..9df2e35afdf1 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..d35ccee58cff --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..24a5f8658339 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..d87b089ae315 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..0b3d057bf4f2 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..d4dd91afd462 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=8192,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=8192,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..b241912bc461 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=7168,K=8192,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 16, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 8, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file diff --git a/vllm/model_executor/layers/quantization/utils/configs/N=8192,K=1536,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json b/vllm/model_executor/layers/quantization/utils/configs/N=8192,K=1536,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json new file mode 100644 index 000000000000..33f807ffd901 --- /dev/null +++ b/vllm/model_executor/layers/quantization/utils/configs/N=8192,K=1536,device_name=MI308X,dtype=fp8_w8a8,block_shape=[128,128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "8": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "16": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "24": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "32": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "48": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 32, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "64": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "96": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "128": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "GROUP_SIZE_M": 1, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "512": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1024": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "1536": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "2048": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "3072": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + }, + "4096": { + "BLOCK_SIZE_K": 128, + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "GROUP_SIZE_M": 32, + "kpack": 1, + "matrix_instr_nonkdim": 16, + "num_warps": 4 + } +} \ No newline at end of file