From e6372c0d772031ef6c4db144cfbeab1645034c1d Mon Sep 17 00:00:00 2001 From: Felipe Mello Date: Thu, 21 Nov 2024 07:28:13 -0800 Subject: [PATCH 01/11] remove default --- torchtune/_cli/download.py | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) diff --git a/torchtune/_cli/download.py b/torchtune/_cli/download.py index 82b4935c01..490ea9a0d5 100644 --- a/torchtune/_cli/download.py +++ b/torchtune/_cli/download.py @@ -91,9 +91,7 @@ def _add_arguments(self) -> None: "--ignore-patterns", type=str, required=False, - default="*.safetensors", - help="If provided, files matching any of the patterns are not downloaded. Defaults to ignoring " - "safetensors files to avoid downloading duplicate weights.", + help="If provided, files matching any of the patterns are not downloaded. Example: '*.safetensors'", ) def _download_cmd(self, args: argparse.Namespace) -> None: From 3c010aac12ea43cf554d29719b503e8db522492f Mon Sep 17 00:00:00 2001 From: Felipe Mello Date: Thu, 21 Nov 2024 08:48:57 -0800 Subject: [PATCH 02/11] update rst --- docs/source/api_ref_models.rst | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/docs/source/api_ref_models.rst b/docs/source/api_ref_models.rst index b2d74022b1..1180ba6766 100644 --- a/docs/source/api_ref_models.rst +++ b/docs/source/api_ref_models.rst @@ -99,7 +99,7 @@ To download the Llama3.1-405B-Instruct model: tune download meta-llama/Meta-Llama-3.1-405B-Instruct --ignore-patterns "original/consolidated*" --hf-token To download the Llama3 weights of the above models, you can instead download from `Meta-Llama-3-8B-Instruct` and -`Meta-Llama-3-70B-Instruct`. +`Meta-Llama-3-70B-Instruct`, and switch to --ignore-patterns "\*.safetensors" .. autosummary:: :toctree: generated/ @@ -307,7 +307,7 @@ To download the Mistral 7B v0.1 model: .. code-block:: bash - tune download mistralai/Mistral-7B-v0.1 --output-dir /tmp/Mistral-7B-v0.1 --hf-token + tune download mistralai/Mistral-7B-v0.1 --output-dir /tmp/Mistral-7B-v0.2 --ignore-patterns "*.safetensors" --hf-token .. autosummary:: :toctree: generated/ From 7c87dab053059862011a443a9664156ce92b65c7 Mon Sep 17 00:00:00 2001 From: Felipe Mello Date: Thu, 21 Nov 2024 08:49:21 -0800 Subject: [PATCH 03/11] update mistral --- recipes/configs/mistral/7B_full.yaml | 2 +- recipes/configs/mistral/7B_full_low_memory.yaml | 2 +- recipes/configs/mistral/7B_full_ppo_low_memory.yaml | 2 +- recipes/configs/mistral/7B_lora.yaml | 2 +- recipes/configs/mistral/7B_lora_single_device.yaml | 2 +- recipes/configs/mistral/7B_qlora_single_device.yaml | 2 +- 6 files changed, 6 insertions(+), 6 deletions(-) diff --git a/recipes/configs/mistral/7B_full.yaml b/recipes/configs/mistral/7B_full.yaml index e025cee824..23c82e1d71 100644 --- a/recipes/configs/mistral/7B_full.yaml +++ b/recipes/configs/mistral/7B_full.yaml @@ -7,7 +7,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download mistralai/Mistral-7B-v0.1 --hf-token --output-dir /tmp/Mistral-7B-v0.1 +# tune download mistralai/Mistral-7B-v0.1 --output-dir /tmp/Mistral-7B-v0.1 --ignore-patterns "*.safetensors" --hf-token # # Run this config on 4 GPUs using the following: # tune run --nnodes 1 --nproc_per_node 4 full_finetune_distributed --config mistral/7B_full diff --git a/recipes/configs/mistral/7B_full_low_memory.yaml b/recipes/configs/mistral/7B_full_low_memory.yaml index dbd8a9bae5..01de2f11ea 100644 --- a/recipes/configs/mistral/7B_full_low_memory.yaml +++ b/recipes/configs/mistral/7B_full_low_memory.yaml @@ -7,7 +7,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download mistralai/Mistral-7B-v0.1 --hf-token --output-dir /tmp/Mistral-7B-v0.1 +# tune download mistralai/Mistral-7B-v0.1 --output-dir /tmp/Mistral-7B-v0.1 --ignore-patterns "*.safetensors" --hf-token # # The default config uses an optimizer from bitsandbytes. If you do not have it installed, # you can install it with diff --git a/recipes/configs/mistral/7B_full_ppo_low_memory.yaml b/recipes/configs/mistral/7B_full_ppo_low_memory.yaml index e05be85ff6..77556a6cbd 100644 --- a/recipes/configs/mistral/7B_full_ppo_low_memory.yaml +++ b/recipes/configs/mistral/7B_full_ppo_low_memory.yaml @@ -8,7 +8,7 @@ # This config assumes that you've run the following command before launching # this run: # tune download weqweasdas/RM-Mistral-7B --output-dir /tmp/RM-Mistral-7B/ --ignore-patterns None -# tune download mistralai/Mistral-7B-Instruct-v0.2 --output-dir /tmp/Mistral-7B-Instruct-v0.2/ --hf-token HF_TOKEN +# tune download mistralai/Mistral-7B-Instruct-v0.2 --output-dir /tmp/Mistral-7B-Instruct-v0.2/ --ignore-patterns "*.safetensors" --hf-token # # You'll also need to ensure that {output_dir} exists beforehand, as checkpoints for policy and value models are saved in sub-folders. # The default config uses an optimizer from bitsandbytes. If you do not have it installed, diff --git a/recipes/configs/mistral/7B_lora.yaml b/recipes/configs/mistral/7B_lora.yaml index 30919c13e8..f637240b34 100644 --- a/recipes/configs/mistral/7B_lora.yaml +++ b/recipes/configs/mistral/7B_lora.yaml @@ -7,7 +7,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download mistralai/Mistral-7B-v0.1 --hf-token --output-dir /tmp/Mistral-7B-v0.1 +# tune download mistralai/Mistral-7B-v0.1 --output-dir /tmp/Mistral-7B-v0.1 --ignore-patterns "*.safetensors" --hf-token # # Run this config on 2 GPUs using the following: # tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config mistral/7B_lora diff --git a/recipes/configs/mistral/7B_lora_single_device.yaml b/recipes/configs/mistral/7B_lora_single_device.yaml index b6d2492bf1..c11cbe1ad2 100644 --- a/recipes/configs/mistral/7B_lora_single_device.yaml +++ b/recipes/configs/mistral/7B_lora_single_device.yaml @@ -7,7 +7,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download mistralai/Mistral-7B-v0.1 --hf-token --output-dir /tmp/Mistral-7B-v0.1 +# tune download mistralai/Mistral-7B-v0.1 --output-dir /tmp/Mistral-7B-v0.1 --ignore-patterns "*.safetensors" --hf-token # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config mistral/7B_lora_single_device diff --git a/recipes/configs/mistral/7B_qlora_single_device.yaml b/recipes/configs/mistral/7B_qlora_single_device.yaml index c0252fcb32..536f2efdf1 100644 --- a/recipes/configs/mistral/7B_qlora_single_device.yaml +++ b/recipes/configs/mistral/7B_qlora_single_device.yaml @@ -7,7 +7,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download mistralai/Mistral-7B-v0.1 --hf-token --output-dir /tmp/Mistral-7B-v0.1 +# tune download mistralai/Mistral-7B-v0.1 --output-dir /tmp/Mistral-7B-v0.1 --ignore-patterns "*.safetensors" --hf-token # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config mistral/7B_qlora_single_device From 37631564e873d27fe34076c4b4c77c9881b42688 Mon Sep 17 00:00:00 2001 From: Felipe Mello Date: Thu, 21 Nov 2024 08:50:47 -0800 Subject: [PATCH 04/11] update llama 3 --- recipes/configs/llama3/8B_dora.yaml | 2 +- recipes/configs/llama3/8B_full.yaml | 2 +- recipes/configs/llama3/8B_full_single_device.yaml | 2 +- recipes/configs/llama3/8B_lora.yaml | 2 +- recipes/configs/llama3/8B_lora_single_device.yaml | 2 +- recipes/configs/llama3/8B_qat_full.yaml | 2 +- recipes/configs/llama3/8B_qdora_single_device.yaml | 2 +- recipes/configs/llama3/8B_qlora_single_device.yaml | 2 +- 8 files changed, 8 insertions(+), 8 deletions(-) diff --git a/recipes/configs/llama3/8B_dora.yaml b/recipes/configs/llama3/8B_dora.yaml index ee7a8d07f6..276f303807 100644 --- a/recipes/configs/llama3/8B_dora.yaml +++ b/recipes/configs/llama3/8B_dora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --hf-token +# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --ignore-patterns "*.safetensors" --hf-token # # To launch on 2 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama3/8B_dora diff --git a/recipes/configs/llama3/8B_full.yaml b/recipes/configs/llama3/8B_full.yaml index 9a93d59061..a065fa9ece 100644 --- a/recipes/configs/llama3/8B_full.yaml +++ b/recipes/configs/llama3/8B_full.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --hf-token +# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --ignore-patterns "*.safetensors" --hf-token # # To launch on 4 devices, run the following command from root: # tune run --nproc_per_node 4 full_finetune_distributed --config llama3/8B_full diff --git a/recipes/configs/llama3/8B_full_single_device.yaml b/recipes/configs/llama3/8B_full_single_device.yaml index b3fee76295..a63845fc30 100644 --- a/recipes/configs/llama3/8B_full_single_device.yaml +++ b/recipes/configs/llama3/8B_full_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --hf-token +# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --ignore-patterns "*.safetensors" --hf-token # # The default config uses an optimizer from bitsandbytes. If you do not have it installed, # you can install it with diff --git a/recipes/configs/llama3/8B_lora.yaml b/recipes/configs/llama3/8B_lora.yaml index 3ced0899e4..f9cb8f9d95 100644 --- a/recipes/configs/llama3/8B_lora.yaml +++ b/recipes/configs/llama3/8B_lora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --hf-token +# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --ignore-patterns "*.safetensors" --hf-token # # To launch on 2 devices, run the following command from root: # tune run --nproc_per_node 2 lora_finetune_distributed --config llama3/8B_lora diff --git a/recipes/configs/llama3/8B_lora_single_device.yaml b/recipes/configs/llama3/8B_lora_single_device.yaml index 4535758ac9..5ae3a0088a 100644 --- a/recipes/configs/llama3/8B_lora_single_device.yaml +++ b/recipes/configs/llama3/8B_lora_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --hf-token +# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --ignore-patterns "*.safetensors" --hf-token # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config llama3/8B_lora_single_device diff --git a/recipes/configs/llama3/8B_qat_full.yaml b/recipes/configs/llama3/8B_qat_full.yaml index 274ee0eae0..49c5c7ee74 100644 --- a/recipes/configs/llama3/8B_qat_full.yaml +++ b/recipes/configs/llama3/8B_qat_full.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --hf-token +# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --ignore-patterns "*.safetensors" --hf-token # # To launch on 4 devices, run the following command from root: # tune run --nproc_per_node 4 qat_distributed --config llama3/8B_qat_full diff --git a/recipes/configs/llama3/8B_qdora_single_device.yaml b/recipes/configs/llama3/8B_qdora_single_device.yaml index 8eb1b5151c..823e0f75fe 100644 --- a/recipes/configs/llama3/8B_qdora_single_device.yaml +++ b/recipes/configs/llama3/8B_qdora_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --hf-token +# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --ignore-patterns "*.safetensors" --hf-token # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config llama3/8B_qdora_single_device diff --git a/recipes/configs/llama3/8B_qlora_single_device.yaml b/recipes/configs/llama3/8B_qlora_single_device.yaml index 0c4ab423b8..76af71e432 100644 --- a/recipes/configs/llama3/8B_qlora_single_device.yaml +++ b/recipes/configs/llama3/8B_qlora_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --hf-token +# tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --ignore-patterns "*.safetensors" --hf-token # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config llama3/8B_qlora_single_device From ec217fcc77a92514b62f0e88c6dc63304d77f280 Mon Sep 17 00:00:00 2001 From: Felipe Mello Date: Thu, 21 Nov 2024 08:50:56 -0800 Subject: [PATCH 05/11] update llama 2 --- recipes/configs/llama2/13B_full.yaml | 2 +- recipes/configs/llama2/13B_lora.yaml | 2 +- recipes/configs/llama2/13B_qlora_single_device.yaml | 2 +- recipes/configs/llama2/70B_lora.yaml | 2 +- recipes/configs/llama2/70B_qlora.yaml | 2 +- recipes/configs/llama2/7B_full.yaml | 2 +- recipes/configs/llama2/7B_full_low_memory.yaml | 2 +- recipes/configs/llama2/7B_lora.yaml | 2 +- recipes/configs/llama2/7B_lora_dpo.yaml | 2 +- recipes/configs/llama2/7B_lora_dpo_single_device.yaml | 2 +- recipes/configs/llama2/7B_lora_single_device.yaml | 2 +- recipes/configs/llama2/7B_qat_full.yaml | 2 +- recipes/configs/llama2/7B_qlora.yaml | 2 +- recipes/configs/llama2/7B_qlora_single_device.yaml | 2 +- recipes/configs/llama2/generation_v2.yaml | 2 +- 15 files changed, 15 insertions(+), 15 deletions(-) diff --git a/recipes/configs/llama2/13B_full.yaml b/recipes/configs/llama2/13B_full.yaml index d577559305..fd7b7421c1 100644 --- a/recipes/configs/llama2/13B_full.yaml +++ b/recipes/configs/llama2/13B_full.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Llama-2-13b-hf --output-dir /tmp/Llama-2-13b-hf --hf-token +# tune download meta-llama/Llama-2-13b-hf --output-dir /tmp/Llama-2-13b-hf --ignore-patterns "*.safetensors" --hf-token # # To launch on 4 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 4 full_finetune_distributed --config llama2/13B_full diff --git a/recipes/configs/llama2/13B_lora.yaml b/recipes/configs/llama2/13B_lora.yaml index 7a6fa600d2..2bae98471d 100644 --- a/recipes/configs/llama2/13B_lora.yaml +++ b/recipes/configs/llama2/13B_lora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Llama-2-13b-hf --output-dir /tmp/Llama-2-13b-hf --hf-token +# tune download meta-llama/Llama-2-13b-hf --output-dir /tmp/Llama-2-13b-hf --ignore-patterns "*.safetensors" --hf-token # # To launch on 4 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 4 lora_finetune_distributed --config llama2/13B_lora diff --git a/recipes/configs/llama2/13B_qlora_single_device.yaml b/recipes/configs/llama2/13B_qlora_single_device.yaml index a10285544a..62e74c4e62 100644 --- a/recipes/configs/llama2/13B_qlora_single_device.yaml +++ b/recipes/configs/llama2/13B_qlora_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Llama-2-13b-hf --output-dir /tmp/Llama-2-13b-hf --hf-token +# tune download meta-llama/Llama-2-13b-hf --output-dir /tmp/Llama-2-13b-hf --ignore-patterns "*.safetensors" --hf-token # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config llama2/13B_qlora_single_device diff --git a/recipes/configs/llama2/70B_lora.yaml b/recipes/configs/llama2/70B_lora.yaml index a67bfc9da2..bf2a0817d0 100644 --- a/recipes/configs/llama2/70B_lora.yaml +++ b/recipes/configs/llama2/70B_lora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Llama-2-70b-hf --output-dir /tmp/Llama-2-70b-hf --hf-token +# tune download meta-llama/Llama-2-70b-hf --output-dir /tmp/Llama-2-70b-hf --ignore-patterns "*.safetensors" --hf-token # # This config needs 8 GPUs to run # # tune run --nproc_per_node 8 lora_finetune_distributed --config llama2/70B_lora diff --git a/recipes/configs/llama2/70B_qlora.yaml b/recipes/configs/llama2/70B_qlora.yaml index d04b7c6753..38444bf0c7 100644 --- a/recipes/configs/llama2/70B_qlora.yaml +++ b/recipes/configs/llama2/70B_qlora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Llama-2-70b-hf --output-dir /tmp/Llama-2-70b-hf --hf-token +# tune download meta-llama/Llama-2-70b-hf --output-dir /tmp/Llama-2-70b-hf --ignore-patterns "*.safetensors" --hf-token # # This config needs 8 GPUs to run # # tune run --nproc_per_node 8 lora_finetune_distributed --config llama2/70B_qlora diff --git a/recipes/configs/llama2/7B_full.yaml b/recipes/configs/llama2/7B_full.yaml index 3031538137..7e69c8f5a6 100644 --- a/recipes/configs/llama2/7B_full.yaml +++ b/recipes/configs/llama2/7B_full.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --hf-token +# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --ignore-patterns "*.safetensors" --hf-token # # To launch on 4 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 4 full_finetune_distributed --config llama2/7B_full diff --git a/recipes/configs/llama2/7B_full_low_memory.yaml b/recipes/configs/llama2/7B_full_low_memory.yaml index 07514959db..d7ee50898e 100644 --- a/recipes/configs/llama2/7B_full_low_memory.yaml +++ b/recipes/configs/llama2/7B_full_low_memory.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --hf-token +# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --ignore-patterns "*.safetensors" --hf-token # # The default config uses an optimizer from bitsandbytes. If you do not have it installed, # you can install it with diff --git a/recipes/configs/llama2/7B_lora.yaml b/recipes/configs/llama2/7B_lora.yaml index 8e64a3fc11..5bf21ccb2c 100644 --- a/recipes/configs/llama2/7B_lora.yaml +++ b/recipes/configs/llama2/7B_lora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --hf-token +# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --ignore-patterns "*.safetensors" --hf-token # # To launch on 2 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama2/7B_lora diff --git a/recipes/configs/llama2/7B_lora_dpo.yaml b/recipes/configs/llama2/7B_lora_dpo.yaml index f3b827ae3b..abf1b43138 100644 --- a/recipes/configs/llama2/7B_lora_dpo.yaml +++ b/recipes/configs/llama2/7B_lora_dpo.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --hf-token +# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --ignore-patterns "*.safetensors" --hf-token # # To launch on 2 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 2 lora_dpo_distributed --config llama2/7B_lora_dpo diff --git a/recipes/configs/llama2/7B_lora_dpo_single_device.yaml b/recipes/configs/llama2/7B_lora_dpo_single_device.yaml index 6483219e9b..7543cb5d6f 100644 --- a/recipes/configs/llama2/7B_lora_dpo_single_device.yaml +++ b/recipes/configs/llama2/7B_lora_dpo_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --hf-token +# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --ignore-patterns "*.safetensors" --hf-token # # To launch on a single device, run the following command from root: # tune run lora_dpo_single_device --config llama2/7B_lora_dpo_single_device diff --git a/recipes/configs/llama2/7B_lora_single_device.yaml b/recipes/configs/llama2/7B_lora_single_device.yaml index 481fed1a7e..4196cc5a59 100644 --- a/recipes/configs/llama2/7B_lora_single_device.yaml +++ b/recipes/configs/llama2/7B_lora_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --hf-token +# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --ignore-patterns "*.safetensors" --hf-token # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config llama2/7B_lora_single_device diff --git a/recipes/configs/llama2/7B_qat_full.yaml b/recipes/configs/llama2/7B_qat_full.yaml index 1208627f1c..15a3f000e4 100644 --- a/recipes/configs/llama2/7B_qat_full.yaml +++ b/recipes/configs/llama2/7B_qat_full.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --hf-token +# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --ignore-patterns "*.safetensors" --hf-token # # To launch on 4 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 4 qat_distributed --config llama2/7B_qat_full diff --git a/recipes/configs/llama2/7B_qlora.yaml b/recipes/configs/llama2/7B_qlora.yaml index 80cee9853c..667b94c376 100644 --- a/recipes/configs/llama2/7B_qlora.yaml +++ b/recipes/configs/llama2/7B_qlora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --hf-token +# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --ignore-patterns "*.safetensors" --hf-token # # To launch on 2 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama2/7B_qlora diff --git a/recipes/configs/llama2/7B_qlora_single_device.yaml b/recipes/configs/llama2/7B_qlora_single_device.yaml index b1f119d7db..028265007e 100644 --- a/recipes/configs/llama2/7B_qlora_single_device.yaml +++ b/recipes/configs/llama2/7B_qlora_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --hf-token +# tune download meta-llama/Llama-2-7b-hf --output-dir /tmp/Llama-2-7b-hf --ignore-patterns "*.safetensors" --hf-token # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config llama2/7B_qlora_single_device diff --git a/recipes/configs/llama2/generation_v2.yaml b/recipes/configs/llama2/generation_v2.yaml index 7ce4e2c43d..7a9222862d 100644 --- a/recipes/configs/llama2/generation_v2.yaml +++ b/recipes/configs/llama2/generation_v2.yaml @@ -1,7 +1,7 @@ # Config for running the InferenceRecipe in generate_V2.py to generate output from an LLM # # This config assumes that you've run the following command before launching: -# tune download meta-llama/Llama-2-7b-chat-hf --output-dir /tmp/Llama-2-7b-chat-hf +# tune download meta-llama/Llama-2-7b-chat-hf --output-dir /tmp/Llama-2-7b-chat-hf --ignore-patterns "*.safetensors" # # To launch, run the following command: # tune run dev/generate_v2 --config llama2/generation_v2 From 9ce0b12fc6afec0246354b53d7b8ec9f3a3acee3 Mon Sep 17 00:00:00 2001 From: Felipe Mello Date: Thu, 21 Nov 2024 08:53:18 -0800 Subject: [PATCH 06/11] fix --- docs/source/api_ref_models.rst | 2 +- recipes/configs/llama3/70B_full.yaml | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/docs/source/api_ref_models.rst b/docs/source/api_ref_models.rst index 1180ba6766..a571f96e30 100644 --- a/docs/source/api_ref_models.rst +++ b/docs/source/api_ref_models.rst @@ -99,7 +99,7 @@ To download the Llama3.1-405B-Instruct model: tune download meta-llama/Meta-Llama-3.1-405B-Instruct --ignore-patterns "original/consolidated*" --hf-token To download the Llama3 weights of the above models, you can instead download from `Meta-Llama-3-8B-Instruct` and -`Meta-Llama-3-70B-Instruct`, and switch to --ignore-patterns "\*.safetensors" +`Meta-Llama-3-70B-Instruct`, and remove the ignore patterns flag. .. autosummary:: :toctree: generated/ diff --git a/recipes/configs/llama3/70B_full.yaml b/recipes/configs/llama3/70B_full.yaml index 7cffc3fce9..5878b2fd95 100644 --- a/recipes/configs/llama3/70B_full.yaml +++ b/recipes/configs/llama3/70B_full.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download meta-llama/Meta-Llama-3-70B-Instruct --output-dir /tmp/Meta-Llama-3-70B-Instruct --ignore-patterns "original/consolidated*" --hf-token +# tune download meta-llama/Meta-Llama-3-70B-Instruct --output-dir /tmp/Meta-Llama-3-70B-Instruct --ignore-patterns "original/consolidated*" --hf-token # # To launch on 8 devices, run the following command from root: # tune run --nproc_per_node 8 full_finetune_distributed --config llama3/70B_full From 59dc10fb95b3ef88bc444fbbfe6546a1606e50d8 Mon Sep 17 00:00:00 2001 From: Felipe Mello Date: Thu, 21 Nov 2024 08:53:38 -0800 Subject: [PATCH 07/11] fix --- docs/source/api_ref_models.rst | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/source/api_ref_models.rst b/docs/source/api_ref_models.rst index a571f96e30..068c5334e2 100644 --- a/docs/source/api_ref_models.rst +++ b/docs/source/api_ref_models.rst @@ -307,7 +307,7 @@ To download the Mistral 7B v0.1 model: .. code-block:: bash - tune download mistralai/Mistral-7B-v0.1 --output-dir /tmp/Mistral-7B-v0.2 --ignore-patterns "*.safetensors" --hf-token + tune download mistralai/Mistral-7B-v0.1 --output-dir /tmp/Mistral-7B-v0.1 --ignore-patterns "*.safetensors" --hf-token .. autosummary:: :toctree: generated/ From de9dbf28188ceab49ef2fff7a6c975c09c2c2e85 Mon Sep 17 00:00:00 2001 From: Felipe Mello Date: Thu, 21 Nov 2024 08:57:20 -0800 Subject: [PATCH 08/11] remove default after merge --- torchtune/_cli/download.py | 1 - 1 file changed, 1 deletion(-) diff --git a/torchtune/_cli/download.py b/torchtune/_cli/download.py index ca8921bec7..52315b255e 100644 --- a/torchtune/_cli/download.py +++ b/torchtune/_cli/download.py @@ -108,7 +108,6 @@ def _add_arguments(self) -> None: "--ignore-patterns", type=str, required=False, - default="*.safetensors", help="If provided, files matching any of the patterns are not downloaded. Example: '*.safetensors'", ) self._parser.add_argument( From f33111c1a423f44ed8ef779f6da14370b57489c3 Mon Sep 17 00:00:00 2001 From: Felipe Mello Date: Thu, 21 Nov 2024 09:17:00 -0800 Subject: [PATCH 09/11] update tests --- tests/torchtune/_cli/test_download.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tests/torchtune/_cli/test_download.py b/tests/torchtune/_cli/test_download.py index 3338adc870..f89b91a845 100644 --- a/tests/torchtune/_cli/test_download.py +++ b/tests/torchtune/_cli/test_download.py @@ -34,7 +34,7 @@ def snapshot_download(self, mocker, tmpdir): def test_download_calls_snapshot(self, capsys, monkeypatch, snapshot_download): model = "meta-llama/Llama-2-7b" - testargs = f"tune download {model}".split() + testargs = f"tune download {model} --ignore-patterns *.safetensors".split() monkeypatch.setattr(sys, "argv", testargs) # Call the first time and get GatedRepoError From 2a1d0c225bcae9ff1d423c0ba4ff0ce5b77dd9d1 Mon Sep 17 00:00:00 2001 From: Felipe Mello Date: Thu, 21 Nov 2024 12:11:14 -0800 Subject: [PATCH 10/11] added comment --- torchtune/_cli/download.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/torchtune/_cli/download.py b/torchtune/_cli/download.py index 52315b255e..55f5ef4ab1 100644 --- a/torchtune/_cli/download.py +++ b/torchtune/_cli/download.py @@ -108,7 +108,8 @@ def _add_arguments(self) -> None: "--ignore-patterns", type=str, required=False, - help="If provided, files matching any of the patterns are not downloaded. Example: '*.safetensors'", + help="If provided, files matching any of the patterns are not downloaded. Example: '*.safetensors'. " + "Only supported for Hugging Face Hub models.", ) self._parser.add_argument( "--source", From b05689b73b5c2e0d5b2a46d8ab107786255af631 Mon Sep 17 00:00:00 2001 From: Felipe Mello Date: Thu, 21 Nov 2024 13:07:36 -0800 Subject: [PATCH 11/11] remove ignore patterns none --- docs/source/api_ref_models.rst | 6 +++--- recipes/configs/mistral/7B_full_ppo_low_memory.yaml | 2 +- recipes/configs/phi3/mini_full.yaml | 2 +- recipes/configs/phi3/mini_full_low_memory.yaml | 2 +- recipes/configs/phi3/mini_lora.yaml | 2 +- recipes/configs/phi3/mini_lora_single_device.yaml | 2 +- recipes/configs/phi3/mini_qlora_single_device.yaml | 2 +- recipes/configs/qwen2/0.5B_full.yaml | 2 +- recipes/configs/qwen2/0.5B_full_single_device.yaml | 2 +- recipes/configs/qwen2/0.5B_lora.yaml | 2 +- recipes/configs/qwen2/0.5B_lora_single_device.yaml | 2 +- recipes/configs/qwen2/1.5B_full.yaml | 2 +- recipes/configs/qwen2/1.5B_full_single_device.yaml | 2 +- recipes/configs/qwen2/1.5B_lora.yaml | 2 +- recipes/configs/qwen2/1.5B_lora_single_device.yaml | 2 +- recipes/configs/qwen2/7B_full.yaml | 2 +- recipes/configs/qwen2/7B_full_single_device.yaml | 2 +- recipes/configs/qwen2/7B_lora.yaml | 2 +- recipes/configs/qwen2/7B_lora_single_device.yaml | 2 +- .../configs/qwen2/knowledge_distillation_distributed.yaml | 4 ++-- .../configs/qwen2/knowledge_distillation_single_device.yaml | 4 ++-- recipes/configs/qwen2_5/0.5B_full.yaml | 2 +- recipes/configs/qwen2_5/0.5B_full_single_device.yaml | 2 +- recipes/configs/qwen2_5/0.5B_lora.yaml | 2 +- recipes/configs/qwen2_5/0.5B_lora_single_device.yaml | 2 +- recipes/configs/qwen2_5/1.5B_full.yaml | 2 +- recipes/configs/qwen2_5/1.5B_full_single_device.yaml | 2 +- recipes/configs/qwen2_5/1.5B_lora.yaml | 2 +- recipes/configs/qwen2_5/1.5B_lora_single_device.yaml | 2 +- recipes/configs/qwen2_5/14B_lora_single_device.yaml | 2 +- recipes/configs/qwen2_5/32B_lora.yaml | 2 +- recipes/configs/qwen2_5/3B_full.yaml | 2 +- recipes/configs/qwen2_5/3B_full_single_device.yaml | 2 +- recipes/configs/qwen2_5/3B_lora.yaml | 2 +- recipes/configs/qwen2_5/3B_lora_single_device.yaml | 2 +- recipes/configs/qwen2_5/72B_lora.yaml | 2 +- recipes/configs/qwen2_5/7B_full.yaml | 2 +- recipes/configs/qwen2_5/7B_full_single_device.yaml | 2 +- recipes/configs/qwen2_5/7B_lora.yaml | 2 +- recipes/configs/qwen2_5/7B_lora_single_device.yaml | 2 +- 40 files changed, 44 insertions(+), 44 deletions(-) diff --git a/docs/source/api_ref_models.rst b/docs/source/api_ref_models.rst index 068c5334e2..e658de7294 100644 --- a/docs/source/api_ref_models.rst +++ b/docs/source/api_ref_models.rst @@ -217,7 +217,7 @@ To download the Qwen2.5 1.5B model, for example: .. code-block:: bash - tune download Qwen/Qwen2.5-1.5B-Instruct --output-dir /tmp/Qwen2_5-1_5B-Instruct --ignore-patterns None + tune download Qwen/Qwen2.5-1.5B-Instruct --output-dir /tmp/Qwen2_5-1_5B-Instruct .. autosummary:: :toctree: generated/ @@ -258,7 +258,7 @@ To download the Qwen2 1.5B model, for example: .. code-block:: bash - tune download Qwen/Qwen2-1.5B-Instruct --output-dir /tmp/Qwen2-1.5B-Instruct --ignore-patterns None + tune download Qwen/Qwen2-1.5B-Instruct --output-dir /tmp/Qwen2-1.5B-Instruct .. autosummary:: :toctree: generated/ @@ -283,7 +283,7 @@ To download the Phi-3 Mini 4k instruct model: .. code-block:: bash - tune download microsoft/Phi-3-mini-4k-instruct --output-dir /tmp/Phi-3-mini-4k-instruct --ignore-patterns None --hf-token + tune download microsoft/Phi-3-mini-4k-instruct --output-dir /tmp/Phi-3-mini-4k-instruct --hf-token .. autosummary:: :toctree: generated/ diff --git a/recipes/configs/mistral/7B_full_ppo_low_memory.yaml b/recipes/configs/mistral/7B_full_ppo_low_memory.yaml index 77556a6cbd..310c9e5bcf 100644 --- a/recipes/configs/mistral/7B_full_ppo_low_memory.yaml +++ b/recipes/configs/mistral/7B_full_ppo_low_memory.yaml @@ -7,7 +7,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download weqweasdas/RM-Mistral-7B --output-dir /tmp/RM-Mistral-7B/ --ignore-patterns None +# tune download weqweasdas/RM-Mistral-7B --output-dir /tmp/RM-Mistral-7B/ # tune download mistralai/Mistral-7B-Instruct-v0.2 --output-dir /tmp/Mistral-7B-Instruct-v0.2/ --ignore-patterns "*.safetensors" --hf-token # # You'll also need to ensure that {output_dir} exists beforehand, as checkpoints for policy and value models are saved in sub-folders. diff --git a/recipes/configs/phi3/mini_full.yaml b/recipes/configs/phi3/mini_full.yaml index 1319ab816d..594ffdc916 100644 --- a/recipes/configs/phi3/mini_full.yaml +++ b/recipes/configs/phi3/mini_full.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download microsoft/Phi-3-mini-4k-instruct --output-dir /tmp/Phi-3-mini-4k-instruct --ignore-patterns None --hf-token +# tune download microsoft/Phi-3-mini-4k-instruct --output-dir /tmp/Phi-3-mini-4k-instruct --hf-token # # Run this config on 4 GPUs using the following: # tune run --nproc_per_node 4 full_finetune_distributed --config phi3/mini_full diff --git a/recipes/configs/phi3/mini_full_low_memory.yaml b/recipes/configs/phi3/mini_full_low_memory.yaml index ad7e0f4046..05c1db379a 100644 --- a/recipes/configs/phi3/mini_full_low_memory.yaml +++ b/recipes/configs/phi3/mini_full_low_memory.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download microsoft/Phi-3-mini-4k-instruct --output-dir /tmp/Phi-3-mini-4k-instruct --ignore-patterns None --hf-token +# tune download microsoft/Phi-3-mini-4k-instruct --output-dir /tmp/Phi-3-mini-4k-instruct --hf-token # # The default config uses an optimizer from bitsandbytes. If you do not have it installed, # you can install it with diff --git a/recipes/configs/phi3/mini_lora.yaml b/recipes/configs/phi3/mini_lora.yaml index 5547be21e0..0c13048119 100644 --- a/recipes/configs/phi3/mini_lora.yaml +++ b/recipes/configs/phi3/mini_lora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download microsoft/Phi-3-mini-4k-instruct --output-dir /tmp/Phi-3-mini-4k-instruct --ignore-patterns None --hf-token +# tune download microsoft/Phi-3-mini-4k-instruct --output-dir /tmp/Phi-3-mini-4k-instruct --hf-token # # To launch on 2 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config phi3/mini_lora diff --git a/recipes/configs/phi3/mini_lora_single_device.yaml b/recipes/configs/phi3/mini_lora_single_device.yaml index 533972b0e1..3aae4f2b6c 100644 --- a/recipes/configs/phi3/mini_lora_single_device.yaml +++ b/recipes/configs/phi3/mini_lora_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download microsoft/Phi-3-mini-4k-instruct --output-dir /tmp/Phi-3-mini-4k-instruct --ignore-patterns None --hf-token +# tune download microsoft/Phi-3-mini-4k-instruct --output-dir /tmp/Phi-3-mini-4k-instruct --hf-token # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config phi3/mini_lora_single_device diff --git a/recipes/configs/phi3/mini_qlora_single_device.yaml b/recipes/configs/phi3/mini_qlora_single_device.yaml index e89bd1a542..f59a68a59d 100644 --- a/recipes/configs/phi3/mini_qlora_single_device.yaml +++ b/recipes/configs/phi3/mini_qlora_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download microsoft/Phi-3-mini-4k-instruct --output-dir /tmp/Phi-3-mini-4k-instruct --ignore-patterns None --hf-token +# tune download microsoft/Phi-3-mini-4k-instruct --output-dir /tmp/Phi-3-mini-4k-instruct --hf-token # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config phi3/mini_qlora_single_device diff --git a/recipes/configs/qwen2/0.5B_full.yaml b/recipes/configs/qwen2/0.5B_full.yaml index ca5863c37c..84336894be 100644 --- a/recipes/configs/qwen2/0.5B_full.yaml +++ b/recipes/configs/qwen2/0.5B_full.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2-0.5B-Instruct --output-dir /tmp/Qwen2-0.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2-0.5B-Instruct --output-dir /tmp/Qwen2-0.5B-Instruct # # To launch on 4 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 4 full_finetune_distributed --config qwen2/0.5B_full diff --git a/recipes/configs/qwen2/0.5B_full_single_device.yaml b/recipes/configs/qwen2/0.5B_full_single_device.yaml index 7e491216c1..8b60a17090 100644 --- a/recipes/configs/qwen2/0.5B_full_single_device.yaml +++ b/recipes/configs/qwen2/0.5B_full_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2-0.5B-Instruct --output-dir /tmp/Qwen2-0.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2-0.5B-Instruct --output-dir /tmp/Qwen2-0.5B-Instruct # # To launch on a single device, run the following command from root: # tune run full_finetune_single_device --config qwen2/0.5B_full_single_device diff --git a/recipes/configs/qwen2/0.5B_lora.yaml b/recipes/configs/qwen2/0.5B_lora.yaml index 9f54c5fdbe..16e5955da3 100644 --- a/recipes/configs/qwen2/0.5B_lora.yaml +++ b/recipes/configs/qwen2/0.5B_lora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2-0.5B-Instruct --output-dir /tmp/Qwen2-0.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2-0.5B-Instruct --output-dir /tmp/Qwen2-0.5B-Instruct # # To launch on 2 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config qwen2/0.5B_lora diff --git a/recipes/configs/qwen2/0.5B_lora_single_device.yaml b/recipes/configs/qwen2/0.5B_lora_single_device.yaml index e9907ec939..e54db398fb 100644 --- a/recipes/configs/qwen2/0.5B_lora_single_device.yaml +++ b/recipes/configs/qwen2/0.5B_lora_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2-0.5B-Instruct --output-dir /tmp/Qwen2-0.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2-0.5B-Instruct --output-dir /tmp/Qwen2-0.5B-Instruct # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config qwen2/0.5B_lora_single_device diff --git a/recipes/configs/qwen2/1.5B_full.yaml b/recipes/configs/qwen2/1.5B_full.yaml index 812ea45b10..37b5c0a926 100644 --- a/recipes/configs/qwen2/1.5B_full.yaml +++ b/recipes/configs/qwen2/1.5B_full.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2-1.5B-Instruct --output-dir /tmp/Qwen2-1.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2-1.5B-Instruct --output-dir /tmp/Qwen2-1.5B-Instruct # # To launch on 4 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 4 full_finetune_distributed --config qwen2/1.5B_full diff --git a/recipes/configs/qwen2/1.5B_full_single_device.yaml b/recipes/configs/qwen2/1.5B_full_single_device.yaml index 3b7642cf24..2acdfb3810 100644 --- a/recipes/configs/qwen2/1.5B_full_single_device.yaml +++ b/recipes/configs/qwen2/1.5B_full_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2-1.5B-Instruct --output-dir /tmp/Qwen2-1.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2-1.5B-Instruct --output-dir /tmp/Qwen2-1.5B-Instruct # # The default config uses an optimizer from bitsandbytes. If you do not have it installed, # you can install it with diff --git a/recipes/configs/qwen2/1.5B_lora.yaml b/recipes/configs/qwen2/1.5B_lora.yaml index d006b29cce..aea2f79e09 100644 --- a/recipes/configs/qwen2/1.5B_lora.yaml +++ b/recipes/configs/qwen2/1.5B_lora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2-1.5B-Instruct --output-dir /tmp/Qwen2-1.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2-1.5B-Instruct --output-dir /tmp/Qwen2-1.5B-Instruct # # To launch on 2 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config qwen2/1.5B_lora diff --git a/recipes/configs/qwen2/1.5B_lora_single_device.yaml b/recipes/configs/qwen2/1.5B_lora_single_device.yaml index 1943be6cb9..2c23954be3 100644 --- a/recipes/configs/qwen2/1.5B_lora_single_device.yaml +++ b/recipes/configs/qwen2/1.5B_lora_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2-1.5B-Instruct --output-dir /tmp/Qwen2-1.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2-1.5B-Instruct --output-dir /tmp/Qwen2-1.5B-Instruct # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config qwen2/1.5B_lora_single_device diff --git a/recipes/configs/qwen2/7B_full.yaml b/recipes/configs/qwen2/7B_full.yaml index d0a6726826..20d74346e1 100644 --- a/recipes/configs/qwen2/7B_full.yaml +++ b/recipes/configs/qwen2/7B_full.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2-7B-Instruct --output-dir /tmp/Qwen2-7B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2-7B-Instruct --output-dir /tmp/Qwen2-7B-Instruct # # To launch on 4 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 4 full_finetune_distributed --config qwen2/7B_full diff --git a/recipes/configs/qwen2/7B_full_single_device.yaml b/recipes/configs/qwen2/7B_full_single_device.yaml index 25e4a1b72b..cff3244b18 100644 --- a/recipes/configs/qwen2/7B_full_single_device.yaml +++ b/recipes/configs/qwen2/7B_full_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2-7B-Instruct --output-dir /tmp/Qwen2-7B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2-7B-Instruct --output-dir /tmp/Qwen2-7B-Instruct # # The default config uses an optimizer from bitsandbytes. If you do not have it installed, # you can install it with diff --git a/recipes/configs/qwen2/7B_lora.yaml b/recipes/configs/qwen2/7B_lora.yaml index c853a7e39f..779e3fdc49 100644 --- a/recipes/configs/qwen2/7B_lora.yaml +++ b/recipes/configs/qwen2/7B_lora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2-7B-Instruct --output-dir /tmp/Qwen2-7B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2-7B-Instruct --output-dir /tmp/Qwen2-7B-Instruct # # To launch on 2 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config qwen2/7B_lora diff --git a/recipes/configs/qwen2/7B_lora_single_device.yaml b/recipes/configs/qwen2/7B_lora_single_device.yaml index 97204f8a1d..d8c576fc41 100644 --- a/recipes/configs/qwen2/7B_lora_single_device.yaml +++ b/recipes/configs/qwen2/7B_lora_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2-7B-Instruct --output-dir /tmp/Qwen2-7B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2-7B-Instruct --output-dir /tmp/Qwen2-7B-Instruct # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config qwen2/7B_lora_single_device diff --git a/recipes/configs/qwen2/knowledge_distillation_distributed.yaml b/recipes/configs/qwen2/knowledge_distillation_distributed.yaml index aad21f1e86..d94f15c54e 100644 --- a/recipes/configs/qwen2/knowledge_distillation_distributed.yaml +++ b/recipes/configs/qwen2/knowledge_distillation_distributed.yaml @@ -3,8 +3,8 @@ # # This config assumes that you've ran the following commands before launching KD: # First download the student and teacher models -# tune download Qwen/Qwen2-0.5B-Instruct --output-dir /tmp/Qwen2-0.5B-Instruct --ignore-patterns None -# tune download Qwen/Qwen2-1.5B-Instruct --output-dir /tmp/Qwen2-1.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2-0.5B-Instruct --output-dir /tmp/Qwen2-0.5B-Instruct +# tune download Qwen/Qwen2-1.5B-Instruct --output-dir /tmp/Qwen2-1.5B-Instruct # # You get better results using KD if the teacher model has already been fine-tuned on the target dataset: # tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config qwen2/1.5B_lora diff --git a/recipes/configs/qwen2/knowledge_distillation_single_device.yaml b/recipes/configs/qwen2/knowledge_distillation_single_device.yaml index 8246cf2e01..70c3496d0e 100644 --- a/recipes/configs/qwen2/knowledge_distillation_single_device.yaml +++ b/recipes/configs/qwen2/knowledge_distillation_single_device.yaml @@ -3,8 +3,8 @@ # # This config assumes that you've ran the following commands before launching KD: # First download the student and teacher models -# tune download Qwen/Qwen2-0.5B-Instruct --output-dir /tmp/Qwen2-0.5B-Instruct --ignore-patterns None -# tune download Qwen/Qwen2-1.5B-Instruct --output-dir /tmp/Qwen2-1.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2-0.5B-Instruct --output-dir /tmp/Qwen2-0.5B-Instruct +# tune download Qwen/Qwen2-1.5B-Instruct --output-dir /tmp/Qwen2-1.5B-Instruct # # You get better results using KD if the teacher model has already been fine-tuned on the target dataset: # tune run lora_finetune_single_device --config qwen2/1.5B_lora_single_device diff --git a/recipes/configs/qwen2_5/0.5B_full.yaml b/recipes/configs/qwen2_5/0.5B_full.yaml index 775e5db2d1..1298c058e9 100644 --- a/recipes/configs/qwen2_5/0.5B_full.yaml +++ b/recipes/configs/qwen2_5/0.5B_full.yaml @@ -2,7 +2,7 @@ # using a Qwen2.5 0.5B model # # This config assumes that you've run the following command before launching: -# tune download Qwen/Qwen2.5-0.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-0.5B-Instruct # # To launch on 2 devices, run the following command from root: # tune run --nproc_per_node 2 full_finetune_distributed --config qwen2_5/0.5B_full diff --git a/recipes/configs/qwen2_5/0.5B_full_single_device.yaml b/recipes/configs/qwen2_5/0.5B_full_single_device.yaml index f668f8416b..39dfb2f8a0 100644 --- a/recipes/configs/qwen2_5/0.5B_full_single_device.yaml +++ b/recipes/configs/qwen2_5/0.5B_full_single_device.yaml @@ -2,7 +2,7 @@ # using a Qwen2.5 0.5B # # This config assumes that you've run the following command before launching: -# tune download Qwen/Qwen2.5-0.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-0.5B-Instruct # # To launch on a single device, run the following command from root: # tune run full_finetune_single_device --config qwen2_5/0.5B_full_single_device diff --git a/recipes/configs/qwen2_5/0.5B_lora.yaml b/recipes/configs/qwen2_5/0.5B_lora.yaml index 3507542663..50fe1a0a28 100644 --- a/recipes/configs/qwen2_5/0.5B_lora.yaml +++ b/recipes/configs/qwen2_5/0.5B_lora.yaml @@ -2,7 +2,7 @@ # using a Qwen2.5 0.5B model # # This config assumes that you've run the following command before launching: -# tune download Qwen/Qwen2.5-0.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-0.5B-Instruct # # To launch on 2 devices, run the following command from root: # tune run --nproc_per_node 2 lora_finetune_distributed --config qwen2_5/0.5B_lora diff --git a/recipes/configs/qwen2_5/0.5B_lora_single_device.yaml b/recipes/configs/qwen2_5/0.5B_lora_single_device.yaml index 8f4e309279..fa507e3414 100644 --- a/recipes/configs/qwen2_5/0.5B_lora_single_device.yaml +++ b/recipes/configs/qwen2_5/0.5B_lora_single_device.yaml @@ -2,7 +2,7 @@ # using a Qwen2.5 0.5B model # # This config assumes that you've run the following command before launching -# tune download Qwen/Qwen2.5-0.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-0.5B-Instruct # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config qwen2_5/0.5B_lora_single_device diff --git a/recipes/configs/qwen2_5/1.5B_full.yaml b/recipes/configs/qwen2_5/1.5B_full.yaml index 8eb535df2c..e0fb09c152 100644 --- a/recipes/configs/qwen2_5/1.5B_full.yaml +++ b/recipes/configs/qwen2_5/1.5B_full.yaml @@ -2,7 +2,7 @@ # using a Qwen2.5 1.5B model # # This config assumes that you've run the following command before launching: -# tune download Qwen/Qwen2.5-1.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-1.5B-Instruct # # To launch on 2 devices, run the following command from root: # tune run --nproc_per_node 2 full_finetune_distributed --config qwen2_5/1.5B_full diff --git a/recipes/configs/qwen2_5/1.5B_full_single_device.yaml b/recipes/configs/qwen2_5/1.5B_full_single_device.yaml index b0e860548d..480249631d 100644 --- a/recipes/configs/qwen2_5/1.5B_full_single_device.yaml +++ b/recipes/configs/qwen2_5/1.5B_full_single_device.yaml @@ -2,7 +2,7 @@ # using a Qwen2.5 1.5B # # This config assumes that you've run the following command before launching: -# tune download Qwen/Qwen2.5-1.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-1.5B-Instruct # # The default config uses an optimizer from bitsandbytes. If you do not have it installed, # you can install it with: diff --git a/recipes/configs/qwen2_5/1.5B_lora.yaml b/recipes/configs/qwen2_5/1.5B_lora.yaml index c44bea6124..8d530c3670 100644 --- a/recipes/configs/qwen2_5/1.5B_lora.yaml +++ b/recipes/configs/qwen2_5/1.5B_lora.yaml @@ -2,7 +2,7 @@ # using a Qwen2.5 1.5B model # # This config assumes that you've run the following command before launching: -# tune download Qwen/Qwen2.5-1.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-1.5B-Instruct # # To launch on 2 devices, run the following command from root: # tune run --nproc_per_node 2 lora_finetune_distributed --config qwen2_5/1.5B_lora diff --git a/recipes/configs/qwen2_5/1.5B_lora_single_device.yaml b/recipes/configs/qwen2_5/1.5B_lora_single_device.yaml index 168dea810c..e784066fe0 100644 --- a/recipes/configs/qwen2_5/1.5B_lora_single_device.yaml +++ b/recipes/configs/qwen2_5/1.5B_lora_single_device.yaml @@ -2,7 +2,7 @@ # using a Qwen2.5 1.5B model # # This config assumes that you've run the following command before launching: -# tune download Qwen/Qwen2.5-1.5B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-1.5B-Instruct # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config qwen2_5/1.5B_lora_single_device diff --git a/recipes/configs/qwen2_5/14B_lora_single_device.yaml b/recipes/configs/qwen2_5/14B_lora_single_device.yaml index 002129641a..2886a56664 100644 --- a/recipes/configs/qwen2_5/14B_lora_single_device.yaml +++ b/recipes/configs/qwen2_5/14B_lora_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2.5-14B-Instruct --output-dir /tmp/Qwen2_5-14B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-14B-Instruct --output-dir /tmp/Qwen2_5-14B-Instruct # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config qwen2_5/14B_lora_single_device diff --git a/recipes/configs/qwen2_5/32B_lora.yaml b/recipes/configs/qwen2_5/32B_lora.yaml index 28cda4f662..bed3868365 100644 --- a/recipes/configs/qwen2_5/32B_lora.yaml +++ b/recipes/configs/qwen2_5/32B_lora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2.5-32B-Instruct --output-dir /tmp/Qwen2_5-32B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-32B-Instruct --output-dir /tmp/Qwen2_5-32B-Instruct # # To launch on 8 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 8 lora_finetune_distributed --config qwen2_5/32B_lora diff --git a/recipes/configs/qwen2_5/3B_full.yaml b/recipes/configs/qwen2_5/3B_full.yaml index 3fb2d23df0..7267dd5efe 100644 --- a/recipes/configs/qwen2_5/3B_full.yaml +++ b/recipes/configs/qwen2_5/3B_full.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2.5-3B-Instruct --output-dir /tmp/Qwen2_5-3B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-3B-Instruct --output-dir /tmp/Qwen2_5-3B-Instruct # # To launch on 2 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 2 full_finetune_distributed --config qwen2_5/3B_full diff --git a/recipes/configs/qwen2_5/3B_full_single_device.yaml b/recipes/configs/qwen2_5/3B_full_single_device.yaml index a5b028c659..ef8d283098 100644 --- a/recipes/configs/qwen2_5/3B_full_single_device.yaml +++ b/recipes/configs/qwen2_5/3B_full_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2.5-3B-Instruct --output-dir /tmp/Qwen2_5-3B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-3B-Instruct --output-dir /tmp/Qwen2_5-3B-Instruct # # The default config uses an optimizer from bitsandbytes. If you do not have it installed, # you can install it with diff --git a/recipes/configs/qwen2_5/3B_lora.yaml b/recipes/configs/qwen2_5/3B_lora.yaml index ffd3b6c494..6cde39b86e 100644 --- a/recipes/configs/qwen2_5/3B_lora.yaml +++ b/recipes/configs/qwen2_5/3B_lora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2.5-3B-Instruct --output-dir /tmp/Qwen2_5-3B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-3B-Instruct --output-dir /tmp/Qwen2_5-3B-Instruct # # To launch on 2 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config qwen2_5/3B_lora diff --git a/recipes/configs/qwen2_5/3B_lora_single_device.yaml b/recipes/configs/qwen2_5/3B_lora_single_device.yaml index b6c5be1a0a..bd3cb9fa68 100644 --- a/recipes/configs/qwen2_5/3B_lora_single_device.yaml +++ b/recipes/configs/qwen2_5/3B_lora_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2.5-3B-Instruct --output-dir /tmp/Qwen2_5-3B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-3B-Instruct --output-dir /tmp/Qwen2_5-3B-Instruct # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config qwen2_5/3B_lora_single_device diff --git a/recipes/configs/qwen2_5/72B_lora.yaml b/recipes/configs/qwen2_5/72B_lora.yaml index 99019e6c43..fc7ad2dc7d 100644 --- a/recipes/configs/qwen2_5/72B_lora.yaml +++ b/recipes/configs/qwen2_5/72B_lora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2.5-72B-Instruct --output-dir /tmp/Qwen2_5-72B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-72B-Instruct --output-dir /tmp/Qwen2_5-72B-Instruct # # To launch on 8 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 8 lora_finetune_distributed --config qwen2_5/72B_lora diff --git a/recipes/configs/qwen2_5/7B_full.yaml b/recipes/configs/qwen2_5/7B_full.yaml index f6bab9f108..e1de8d5584 100644 --- a/recipes/configs/qwen2_5/7B_full.yaml +++ b/recipes/configs/qwen2_5/7B_full.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2.5-7B-Instruct --output-dir /tmp/Qwen2_5-7B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-7B-Instruct --output-dir /tmp/Qwen2_5-7B-Instruct # # To launch on 2 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 2 full_finetune_distributed --config qwen2_5/7B_full diff --git a/recipes/configs/qwen2_5/7B_full_single_device.yaml b/recipes/configs/qwen2_5/7B_full_single_device.yaml index 0986591e53..3bc3428410 100644 --- a/recipes/configs/qwen2_5/7B_full_single_device.yaml +++ b/recipes/configs/qwen2_5/7B_full_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2.5-7B-Instruct --output-dir /tmp/Qwen2_5-7B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-7B-Instruct --output-dir /tmp/Qwen2_5-7B-Instruct # # The default config uses an optimizer from bitsandbytes. If you do not have it installed, # you can install it with diff --git a/recipes/configs/qwen2_5/7B_lora.yaml b/recipes/configs/qwen2_5/7B_lora.yaml index b59ac69bcd..460c67d26f 100644 --- a/recipes/configs/qwen2_5/7B_lora.yaml +++ b/recipes/configs/qwen2_5/7B_lora.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2.5-7B-Instruct --output-dir /tmp/Qwen2_5-7B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-7B-Instruct --output-dir /tmp/Qwen2_5-7B-Instruct # # To launch on 2 devices, run the following command from root: # tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config qwen2_5/7B_lora diff --git a/recipes/configs/qwen2_5/7B_lora_single_device.yaml b/recipes/configs/qwen2_5/7B_lora_single_device.yaml index a030c2fba2..5c3353f7e9 100644 --- a/recipes/configs/qwen2_5/7B_lora_single_device.yaml +++ b/recipes/configs/qwen2_5/7B_lora_single_device.yaml @@ -3,7 +3,7 @@ # # This config assumes that you've run the following command before launching # this run: -# tune download Qwen/Qwen2.5-7B-Instruct --output-dir /tmp/Qwen2_5-7B-Instruct --ignore-patterns None +# tune download Qwen/Qwen2.5-7B-Instruct --output-dir /tmp/Qwen2_5-7B-Instruct # # To launch on a single device, run the following command from root: # tune run lora_finetune_single_device --config qwen2_5/7B_lora_single_device