From b6127bf969fddebd9b94e720e7f38cccbf06ff78 Mon Sep 17 00:00:00 2001 From: Masahiro Tanaka Date: Fri, 18 Apr 2025 16:23:38 +0000 Subject: [PATCH 1/2] update description of versions for deepcompile --- benchmarks/deepcompile/README.md | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) diff --git a/benchmarks/deepcompile/README.md b/benchmarks/deepcompile/README.md index 16a825199..440f2093b 100644 --- a/benchmarks/deepcompile/README.md +++ b/benchmarks/deepcompile/README.md @@ -10,7 +10,10 @@ We tested the scripts with Python 3.10.12 and CUDA 12.4. In addition, you need to install the following: - PyTorch v2.6.0 -- DeepSpeed (v0.16.6 or newer) + - For the Mixtral model, we recommend using PyTorch v2.7.0 (currently the final release candidate) due to an issue with `torch.where`. + - See details [here](https://github.com/pytorch/pytorch/issues/149278) +- DeepSpeed (v.0.16.6 or newer) + - As DeepCompile is under active development, we recommend using the latest version or installing from source. - transformers - accelerate - datasets v3.1 @@ -24,6 +27,12 @@ pip3 install transformers datasets==3.1 accelerate # Install DeepSpeed pip install deepspeed +# Or install the latest revision of DeepSpeed from source +# git clone https://github.com/deepspeedai/DeepSpeed +# cd DeepSpeed +# git pip install . +# cd .. + # Clone this repository git clone https://github.com/deepspeedai/DeepSpeedExamples cd benchmarks/deepcompile From efd404b26f3cb213dda022384ef62350e36059e0 Mon Sep 17 00:00:00 2001 From: Logan Adams Date: Fri, 18 Apr 2025 09:46:26 -0700 Subject: [PATCH 2/2] Update to match specific tag name Signed-off-by: Logan Adams --- benchmarks/deepcompile/README.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/benchmarks/deepcompile/README.md b/benchmarks/deepcompile/README.md index 440f2093b..fbcf02489 100644 --- a/benchmarks/deepcompile/README.md +++ b/benchmarks/deepcompile/README.md @@ -12,7 +12,7 @@ In addition, you need to install the following: - PyTorch v2.6.0 - For the Mixtral model, we recommend using PyTorch v2.7.0 (currently the final release candidate) due to an issue with `torch.where`. - See details [here](https://github.com/pytorch/pytorch/issues/149278) -- DeepSpeed (v.0.16.6 or newer) +- DeepSpeed (v0.16.6 or newer) - As DeepCompile is under active development, we recommend using the latest version or installing from source. - transformers - accelerate @@ -117,7 +117,7 @@ Here are some example charts: To enable DeepCompile, simply set "deepcompile": true in the compile section of your DeepSpeed configuration JSON: ```json -{ +{ … "zero_optimization": { "stage": 3,