Merge pull request #17 from stackav-oss/feature/jmanning/more-vllm-0.8.5-cleanup-v3

More vLLM 0.8.5 cleanup

Author: jmanning-stackav

benchmarks/bnb_dequantize_blockwise_benchmark.py

@@ -69,47 +69,35 @@ def _to_torch_dtype(dtype_str: str) -> torch.dtype:
 )
 @click.option(
     "--compress-statistics",
-    required=False,
-    type=bool,
     is_flag=True,
-    default=False,
     help="Flag for double-quantization",
 )
 @click.option(
     "--enable-bnb",
-    required=False,
-    type=bool,
     is_flag=True,
     default=envs.CONCH_BENCH_ENABLE_ALL_REF,
     help="Flag to enable BNB reference impl",
 )
 @click.option(
-    "-i",
     "--num-iterations",
     required=False,
     type=int,
     default=100,
     help="Number of iterations",
 )
 @click.option(
-    "-w",
     "--num-warmup-iterations",
     required=False,
     type=int,
     default=10,
     help="Number of warmup iterations",
 )
 @click.option(
-    "-v",
     "--verbose",
-    required=False,
-    type=bool,
     is_flag=True,
-    default=False,
     help="Flag for printing verbose output",
 )
 @click.option(
-    "-g",
     "--gpu",
     required=False,
     type=str,
@@ -118,10 +106,7 @@ def _to_torch_dtype(dtype_str: str) -> torch.dtype:
 )
 @click.option(
     "--csv",
-    required=False,
-    type=bool,
     is_flag=True,
-    default=False,
     help="Flag for printing results in CSV format",
 )
 def main(  # noqa: PLR0913
@@ -202,7 +187,9 @@ def main(  # noqa: PLR0913
             error_msg = "bitsandbytes must be installed and enabled via CONCH_ENABLE_BNB=1"
             raise NotImplementedError(error_msg)
 
-        from bitsandbytes.functional import dequantize_4bit as bnb_dequantize_4bit  # type: ignore[import-untyped]
+        from bitsandbytes.functional import (  # type: ignore[import-not-found, import-untyped, unused-ignore]  # isort:skip
+            dequantize_4bit as bnb_dequantize_4bit,
+        )
         from bitsandbytes.functional import quantize_4bit as bnb_quantize_4bit
 
         bnb_quantized, bnb_state = bnb_quantize_4bit(

benchmarks/bnb_quantize_blockwise_benchmark.py

@@ -68,47 +68,35 @@ def _to_torch_dtype(dtype_str: str) -> torch.dtype:
 )
 @click.option(
     "--compress-statistics",
-    required=False,
-    type=bool,
     is_flag=True,
-    default=False,
     help="Flag for double-quantization",
 )
 @click.option(
     "--enable-bnb",
-    required=False,
-    type=bool,
     is_flag=True,
     default=envs.CONCH_BENCH_ENABLE_ALL_REF,
     help="Flag to enable BNB reference impl",
 )
 @click.option(
-    "-i",
     "--num-iterations",
     required=False,
     type=int,
     default=100,
     help="Number of iterations",
 )
 @click.option(
-    "-w",
     "--num-warmup-iterations",
     required=False,
     type=int,
     default=10,
     help="Number of warmup iterations",
 )
 @click.option(
-    "-v",
     "--verbose",
-    required=False,
-    type=bool,
     is_flag=True,
-    default=False,
     help="Flag for printing verbose output",
 )
 @click.option(
-    "-g",
     "--gpu",
     required=False,
     type=str,
@@ -117,10 +105,7 @@ def _to_torch_dtype(dtype_str: str) -> torch.dtype:
 )
 @click.option(
     "--csv",
-    required=False,
-    type=bool,
     is_flag=True,
-    default=False,
     help="Flag for printing results in CSV format",
 )
 def main(  # noqa: PLR0913
@@ -192,7 +177,9 @@ def main(  # noqa: PLR0913
             error_msg = "bitsandbytes must be installed and enabled via CONCH_ENABLE_BNB=1"
             raise NotImplementedError(error_msg)
 
-        from bitsandbytes.functional import quantize_4bit as bnb_quantize_4bit  # type: ignore[import-untyped]
+        from bitsandbytes.functional import (  # type: ignore[import-not-found, import-untyped, unused-ignore]  # isort:skip
+            quantize_4bit as bnb_quantize_4bit,
+        )
 
         bnb_output, bnb_state = bnb_quantize_4bit(
             x,

benchmarks/copy_blocks_benchmark.py

@@ -18,88 +18,74 @@
 
 @click.command()
 @click.option(
-    "-h",
     "--head-dim",
     required=True,
     type=int,
     default=256,
     help="Head dimension",
 )
 @click.option(
-    "-l",
     "--num-layers",
     required=True,
     type=int,
     default=8,
     help="Number of layers",
 )
 @click.option(
-    "-c",
     "--cache-block-size",
     required=True,
     type=int,
     default=32,
     help="Number of KV vectors in each cache block",
 )
 @click.option(
-    "-k",
     "--num-kv-heads",
     required=False,
     type=int,
     default=8,
     help="Number of kv heads",
 )
 @click.option(
-    "-b",
     "--num-blocks",
     required=False,
     type=int,
     default=2000,
     help="Number of blocks in cache",
 )
 @click.option(
-    "-m",
     "--num-mappings",
     required=False,
     type=int,
     default=512,
     help="Number of mappings to copy",
 )
 @click.option(
-    "-i",
     "--num-iterations",
     required=False,
     type=int,
     default=100,
     help="Number of iterations",
 )
 @click.option(
-    "-w",
     "--num-warmup-iterations",
     required=False,
     type=int,
     default=10,
     help="Number of warmup iterations",
 )
 @click.option(
-    "-a",
     "--absolute-tolerance",
     required=False,
     type=float,
     default=1e-3,
     help="Absolute tolerance to match with",
 )
 @click.option(
-    "-v",
     "--verbose",
-    required=False,
-    type=bool,
     is_flag=True,
-    default=False,
     help="Flag for printing verbose output",
 )
 @click.option(
-    "-g",
     "--gpu",
     required=False,
     type=str,
@@ -108,10 +94,7 @@
 )
 @click.option(
     "--csv",
-    required=False,
-    type=bool,
     is_flag=True,
-    default=False,
     help="Flag for printing results in CSV format",
 )
 def main(

benchmarks/fused_add_rms_norm_benchmark.py

@@ -17,48 +17,39 @@
 
 @click.command()
 @click.option(
-    "-d",
     "--hidden-size",
     required=True,
     type=int,
     default=2048,
     help="Dimension",
 )
 @click.option(
-    "-t",
     "--num-tokens",
     required=True,
     type=int,
     default=4096,
     help="Number of tokens",
 )
 @click.option(
-    "-i",
     "--num-iterations",
     required=False,
     type=int,
     default=100,
     help="Number of iterations",
 )
 @click.option(
-    "-w",
     "--num-warmup-iterations",
     required=False,
     type=int,
     default=10,
     help="Number of warmup iterations",
 )
 @click.option(
-    "-v",
     "--verbose",
-    required=False,
-    type=bool,
     is_flag=True,
-    default=False,
     help="Flag for printing verbose output",
 )
 @click.option(
-    "-g",
     "--gpu",
     required=False,
     type=str,
@@ -67,10 +58,7 @@
 )
 @click.option(
     "--csv",
-    required=False,
-    type=bool,
     is_flag=True,
-    default=False,
     help="Flag for printing results in CSV format",
 )
 def main(  # noqa: PLR0913

benchmarks/gelu_tanh_and_mul_benchmark.py

@@ -17,56 +17,46 @@
 
 @click.command()
 @click.option(
-    "-s",
     "--hidden-size",
     required=False,
     type=int,
     default=13824,
     help="Feedforward hidden size",
 )
 @click.option(
-    "-t",
     "--num-tokens",
     required=False,
     type=int,
     default=8192,
     help="Number of tokens",
 )
 @click.option(
-    "-i",
     "--num-iterations",
     required=False,
     type=int,
     default=100,
     help="Number of iterations",
 )
 @click.option(
-    "-w",
     "--num-warmup-iterations",
     required=False,
     type=int,
     default=10,
     help="Number of warmup iterations",
 )
 @click.option(
-    "-a",
     "--absolute-tolerance",
     required=False,
     type=float,
     default=1e-3,
     help="Absolute tolerance to match with",
 )
 @click.option(
-    "-v",
     "--verbose",
-    required=False,
-    type=bool,
     is_flag=True,
-    default=False,
     help="Flag for printing verbose output",
 )
 @click.option(
-    "-g",
     "--gpu",
     required=False,
     type=str,
@@ -75,10 +65,7 @@
 )
 @click.option(
     "--csv",
-    required=False,
-    type=bool,
     is_flag=True,
-    default=False,
     help="Flag for printing results in CSV format",
 )
 def main(