add Ulysses Attention, Ring Attention, Unified Attention, and Ulysses Anything Attention

This commit is contained in:
SHYuanBest
2026-03-08 06:35:45 +00:00
parent 7903ec31b5
commit c4ecf8e0d0
11 changed files with 42 additions and 7 deletions
+1
View File
@@ -1,6 +1,7 @@
# Example: Running inference with 2-GPU parallelism
# CUDA_VISIBLE_DEVICES=0,1 torchrun --nproc_per_node 2 infer_helios.py \
# --enable_parallelism \
# --cp_backend "ulysses" \ # ["ring", "ulysses", "unified", "ulysses_anything"]
CUDA_VISIBLE_DEVICES=0 python infer_helios.py \
--base_model_path "BestWishYsh/Helios-Base" \
+1
View File
@@ -1,6 +1,7 @@
# Example: Running inference with 2-GPU parallelism
# CUDA_VISIBLE_DEVICES=0,1 torchrun --nproc_per_node 2 infer_helios.py \
# --enable_parallelism \
# --cp_backend "ulysses" \ # ["ring", "ulysses", "unified", "ulysses_anything"]
CUDA_VISIBLE_DEVICES=0 python infer_helios.py \
--base_model_path "BestWishYsh/Helios-Base" \
+1
View File
@@ -1,6 +1,7 @@
# Example: Running inference with 2-GPU parallelism
# CUDA_VISIBLE_DEVICES=0,1 torchrun --nproc_per_node 2 infer_helios.py \
# --enable_parallelism \
# --cp_backend "ulysses" \ # ["ring", "ulysses", "unified", "ulysses_anything"]
CUDA_VISIBLE_DEVICES=0 python infer_helios.py \
--base_model_path "BestWishYsh/Helios-Base" \
@@ -1,6 +1,7 @@
# Example: Running inference with 2-GPU parallelism
# CUDA_VISIBLE_DEVICES=0,1 torchrun --nproc_per_node 2 infer_helios.py \
# --enable_parallelism \
# --cp_backend "ulysses" \ # ["ring", "ulysses", "unified", "ulysses_anything"]
CUDA_VISIBLE_DEVICES=0 python infer_helios.py \
--base_model_path "BestWishYsh/Helios-Distilled" \
@@ -1,6 +1,7 @@
# Example: Running inference with 2-GPU parallelism
# CUDA_VISIBLE_DEVICES=0,1 torchrun --nproc_per_node 2 infer_helios.py \
# --enable_parallelism \
# --cp_backend "ulysses" \ # ["ring", "ulysses", "unified", "ulysses_anything"]
CUDA_VISIBLE_DEVICES=0 python infer_helios.py \
--base_model_path "BestWishYsh/Helios-Distilled" \
@@ -1,6 +1,7 @@
# Example: Running inference with 2-GPU parallelism
# CUDA_VISIBLE_DEVICES=0,1 torchrun --nproc_per_node 2 infer_helios.py \
# --enable_parallelism \
# --cp_backend "ulysses" \ # ["ring", "ulysses", "unified", "ulysses_anything"]
CUDA_VISIBLE_DEVICES=0 python infer_helios.py \
--base_model_path "BestWishYsh/Helios-Distilled" \
+1
View File
@@ -1,6 +1,7 @@
# Example: Running inference with 2-GPU parallelism
# CUDA_VISIBLE_DEVICES=0,1 torchrun --nproc_per_node 2 infer_helios.py \
# --enable_parallelism \
# --cp_backend "ulysses" \ # ["ring", "ulysses", "unified", "ulysses_anything"]
CUDA_VISIBLE_DEVICES=0 python infer_helios.py \
--base_model_path "BestWishYsh/Helios-Mid" \
+1
View File
@@ -1,6 +1,7 @@
# Example: Running inference with 2-GPU parallelism
# CUDA_VISIBLE_DEVICES=0,1 torchrun --nproc_per_node 2 infer_helios.py \
# --enable_parallelism \
# --cp_backend "ulysses" \ # ["ring", "ulysses", "unified", "ulysses_anything"]
CUDA_VISIBLE_DEVICES=0 python infer_helios.py \
--base_model_path "BestWishYsh/Helios-Mid" \
+1
View File
@@ -1,6 +1,7 @@
# Example: Running inference with 2-GPU parallelism
# CUDA_VISIBLE_DEVICES=0,1 torchrun --nproc_per_node 2 infer_helios.py \
# --enable_parallelism \
# --cp_backend "ulysses" \ # ["ring", "ulysses", "unified", "ulysses_anything"]
CUDA_VISIBLE_DEVICES=0 python infer_helios.py \
--base_model_path "BestWishYsh/Helios-Mid" \