8000 [tiny] Pin flash-attn version by wizeng23 · Pull Request #1775 · oumi-ai/oumi · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

[tiny] Pin flash-attn version #1775

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.

By clicking “Sign up for GitHub”, you agree to our terms of s 8000 ervice and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Jun 19, 2025
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
2 changes: 1 addition & 1 deletion configs/examples/grpo_verl_countdown/gcp_job.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -34,7 +34,7 @@ envs:
setup: |
set -e
pip install uv && uv pip install oumi[gpu]
# 2.8.0.post2 has installation issues.
# TODO: OPE-1336 - Remove version pin when error with later versions is fixed.
pip install -U "flash-attn==2.7.4.post1" --no-build-isolation

run: |
Expand Down
1 change: 1 addition & 0 deletions configs/examples/grpo_verl_geometry3k/gcp_job.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -41,6 +41,7 @@ setup: |
# In the meantime, we need to use this specific commit to support vLLM 0.8.3:
# https://github.com/volcengine/verl/pull/912
pip install git+https://github.com/volcengine/verl.git@1ee730163f6326e9679644db62eb32c8d1947c7f
# TODO: OPE-1336 - Remove version pin when error with later versions is fixed.
pip install -U "flash-attn==2.7.4.post1" --no-build-isolation

run: |
Expand Down
3 changes: 2 additions & 1 deletion configs/recipes/vision/phi3/sft/full/oumi_gcp_job.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -45,7 +45,8 @@ setup: |
# downloading the model during training.
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download microsoft/Phi-3-vision-128k-instruct

pip install -U flash-attn --no-build-isolation
# TODO: OPE-1336 - Remove version pin when error with later versions is fixed.
pip install -U "flash-attn==2.7.4.post1" --no-build-isolation

run: |
set -e # Exit if any command failed.
Expand Down
3 changes: 2 additions & 1 deletion configs/recipes/vision/phi3/sft/full/trl_gcp_job.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -45,7 +45,8 @@ setup: |
# downloading the model during training.
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download microsoft/Phi-3-vision-128k-instruct

pip install -U flash-attn --no-build-isolation
# TODO: OPE-1336 - Remove version pin when error with later versions is fixed.
pip install -U "flash-attn==2.7.4.post1" --no-build-isolation

run: |
set -e # Exit if any command failed.
Expand Down
3 changes: 2 additions & 1 deletion configs/recipes/vision/phi3/sft/lora/gcp_job.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -45,7 +45,8 @@ setup: |
# downloading the model during training.
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download microsoft/Phi-3-vision-128k-instruct

pip install -U flash-attn --no-build-isolation
# TODO: OPE-1336 - Remove version pin when error with later versions is fixed.
pip install -U "flash-attn==2.7.4.post1" --no-build-isolation

run: |
set -e # Exit if any command failed.
Expand Down
3 changes: 2 additions & 1 deletion configs/recipes/vision/phi4/sft/full/gcp_job.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,8 @@ setup: |
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download microsoft/Phi-4-multimodal-instruct

# The model requires flash_attention_2! Install it here.
pip install -U flash-attn --no-build-isolation
# TODO: OPE-1336 - Remove version pin when error with later versions is fixed.
pip install -U "flash-attn==2.7.4.post1" --no-build-isolation


run: |
Expand Down
3 changes: 2 additions & 1 deletion configs/recipes/vision/phi4/sft/lora/gcp_job.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,8 @@ setup: |
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download microsoft/Phi-4-multimodal-instruct

# The model requires flash_attention_2! Install it here.
pip install -U flash-attn --no-build-isolation
# TODO: OPE-1336 - Remove version pin when error with later versions is fixed.
pip install -U "flash-attn==2.7.4.post1" --no-build-isolation


run: |
Expand Down
3 changes: 2 additions & 1 deletion configs/recipes/vision/qwen2_5_vl_3b/sft/full/gcp_job.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,8 @@ setup: |
# Also, if you want to try it with a more efficient attention implementation,
# you can install the `flash_attention_2` package and set `attn_implementation:
# "flash_attention_2"` in the model config.
# pip install -U flash-attn --no-build-isolation
# # TODO: OPE-1336 - Remove version pin when error with later versions is fixed.
pip install -U "flash-attn==2.7.4.post1" --no-build-isolation

run: |
set -e # Exit if any command failed.
Expand Down
3 changes: 2 additions & 1 deletion configs/recipes/vision/qwen2_5_vl_3b/sft/lora/gcp_job.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,8 @@ setup: |
# Also, if you want to try it with a more efficient attention implementation,
# you can install the `flash_attention_2` package and set `attn_implementation:
# "flash_attention_2"` in the model config.
# pip install -U flash-attn --no-build-isolation
# TODO: OPE-1336 - Remove version pin when error with later versions is fixed.
# pip install -U "flash-attn==2.7.4.post1" --no-build-isolation

run: |
set -e # Exit if any command failed.
Expand Down
5 changes: 2 additions & 3 deletions configs/recipes/vision/qwen2_vl_2b/evaluation/gcp_job.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -38,7 +38,8 @@ envs:
setup: |
set -e
pip install uv && uv pip install oumi[gpu] hf_transfer
pip install -U flash-attn --no-build-isolation
# TODO: OPE-1336 - Remove version pin when error with later versions is fixed.
pip install -U "flash-attn==2.7.4.post1" --no-build-isolation

# Install model from HF Hub. This tool increases download speed compared to
# downloading the model during eval.
Expand All @@ -53,8 +54,6 @@ run: |
exit 1
fi

pip install -U flash-attn --no-build-isolation

echo "Starting evaluation for ${MODEL_CHECKPOINT_DIR} ..."

set -x
Expand Down
3 changes: 2 additions & 1 deletion configs/recipes/vision/smolvlm/sft/full/gcp_job.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -41,7 +41,8 @@ setup: |
# downloading the model during training.
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download HuggingFaceTB/SmolVLM-Instruct --exclude "onnx/*" "runs/*"

pip install -U flash-attn --no-build-isolation
# TODO: OPE-1336 - Remove version pin when error with later versions is fixed.
pip install -U "flash-attn==2.7.4.post1" --no-build-isolation

run: |
set -e # Exit if any command failed.
Expand Down
3 changes: 2 additions & 1 deletion configs/recipes/vision/smolvlm/sft/lora/gcp_job.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -41,7 +41,8 @@ setup: |
# downloading the model during training.
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download HuggingFaceTB/SmolVLM-Instruct --exclude "onnx/*" "runs/*"

pip install -U flash-attn --no-build-isolation
# TODO: OPE-1336 - Remove version pin when error with later versions is fixed.
pip install -U "flash-attn==2.7.4.post1" --no-build-isolation

run: |
set -e # Exit if any command failed.
Expand Down
Loading
0