Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[Misc] Load FP8 kv-cache scaling factors from checkpoints #4893

Merged
merged 6 commits into from
May 22, 2024
Merged

[Misc] Load FP8 kv-cache scaling factors from checkpoints #4893

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
575551f
load fp8 kv-cache scaling factor
comaniac May 17, 2024
e909f28
string
comaniac May 17, 2024
77c61f4
docstring
comaniac May 17, 2024
db2d034
comment
comaniac May 20, 2024
7369d7c
test
comaniac May 20, 2024
61ddb22
done
comaniac May 20, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
string
  • Loading branch information
@comaniac
comaniac committed May 20, 2024
commit e909f289c1c77857c84b54d7edad32d9f06c28f5
10 changes: 5 additions & 5 deletions benchmarks/kernels/benchmark_paged_attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -181,13 +181,13 @@ def run_cuda_benchmark(num_iters: int, profile: bool = False) -> float:
parser.add_argument("--seed", type=int, default=0)
parser.add_argument("--profile", action="store_true")
parser.add_argument(
'--kv-cache-dtype',
"--kv-cache-dtype",
type=str,
choices=['auto', 'fp8', 'fp8_e5m2', 'fp8_e4m3'],
choices=["auto", "fp8", "fp8_e5m2", "fp8_e4m3"],
default="auto",
help='Data type for kv cache storage. If "auto", will use model '
'data type. CUDA 11.8+ supports fp8 (=fp8_e4m3) and fp8_e5m2. '
'ROCm (AMD GPU) supports fp8 (=fp8_e4m3)')
help="Data type for kv cache storage. If 'auto', will use model "
"data type. CUDA 11.8+ supports fp8 (=fp8_e4m3) and fp8_e5m2. "
"ROCm (AMD GPU) supports fp8 (=fp8_e4m3)")
args = parser.parse_args()
print(args)

Expand Down
3 changes: 2 additions & 1 deletion vllm/model_executor/layers/quantization/fp8.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,7 @@
from vllm.model_executor.layers.quantization.base_config import (
QuantizationConfig, QuantizeMethodBase)
from vllm.model_executor.utils import set_weight_attrs
from vllm.utils import print_warning_once

ACTIVATION_SCHEMES = ["static", "dynamic"]

Expand Down Expand Up @@ -287,7 +288,7 @@ def process_weights_after_loading(self, layer: Module) -> None:
"for fp8 KV cache")
layer._kv_scale = kv_scales[0]
if layer._kv_scale == 1.0:
logger.warning(
print_warning_once(
"Using KV cache scaling factor 1.0 for fp8_e4m3. This may "
"cause accuracy issues. Please make sure kv-cache scaling "
"factor is available in the fp8 checkpoint.")
Expand Down