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

torcho gemlite integration #2498

Closed
wants to merge 1 commit into from
+23 −0
Closed

torcho gemlite integration #2498

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
8 changes: 8 additions & 0 deletions python/sglang/bench_one_batch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -385,6 +385,14 @@ def latency_test(
8, # shorter decoding to speed up the warmup
server_args.device,
)

try:
from gemlite.core import GemLiteLinearTriton
import os, pwd
GemLiteLinearTriton.cache_config(f"/tmp/{pwd.getpwuid(os.getuid()).pw_gecos}_gemlite.json")
except ImportError:
pass

rank_print("Benchmark ...")

# Run the sweep
Expand Down
15 changes: 15 additions & 0 deletions python/sglang/srt/layers/torchao_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,6 +25,7 @@ def apply_torchao_config_to_model(
int8_weight_only,
quantize_,
)
from torchao.quantization.quant_api import gemlite_uintx_weight_only
from torchao.quantization.observer import PerRow, PerTensor

if filter_fn is None:
Expand All @@ -47,6 +48,20 @@ def filter_fn(module, fqn):
256,
], f"int4wo groupsize needs to be one of [32, 64, 128, 256] but got {group_size}"
quantize_(model, int4_weight_only(group_size=group_size), filter_fn=filter_fn)
elif "gemlite" in torchao_config:
_quant_args = torchao_config.split("-")
bit_width = int(_quant_args[-2])
group_size = None if _quant_args[-1] == 'None' else int(_quant_args[-1])
packing_bitwidth = int(_quant_args[-3])
quantize_(model, gemlite_uintx_weight_only(group_size, bit_width, packing_bitwidth), filter_fn=filter_fn)

import pwd
import os
from gemlite.core import GemLiteLinearTriton
try:
GemLiteLinearTriton.load_config(f"/tmp/{pwd.getpwuid(os.getuid()).pw_gecos}_gemlite.json")
except:
pass
elif "fp8wo" in torchao_config:
# this requires newer hardware
# [rank0]: AssertionError: fp8e4nv data type is not supported on CUDA arch < 89
Expand Down
Loading