From 50b8ce67474c948ecc5efad8f682584a5b6a41df Mon Sep 17 00:00:00 2001
From: Kartikay Khandelwal <kartikayk@fb.com>
Date: Fri, 26 Apr 2024 11:53:26 -0700
Subject: [PATCH] fix configs

---
 recipes/configs/phi3/mini_full.yaml            | 14 +++++---------
 recipes/configs/phi3/mini_full_low_memory.yaml | 16 ++++++----------
 2 files changed, 11 insertions(+), 19 deletions(-)

diff --git a/recipes/configs/phi3/mini_full.yaml b/recipes/configs/phi3/mini_full.yaml
index 19271778b4..756dc82ca1 100644
--- a/recipes/configs/phi3/mini_full.yaml
+++ b/recipes/configs/phi3/mini_full.yaml
@@ -1,10 +1,6 @@
 # Config for multi-device full finetuning in full_finetune_distributed.py
 # using a Phi3 Mini 4K Instruct
 #
-# This config uses hyperparameters based on small set of experiments and information
-# available on various forums. These are not meant to replicate the numbers
-# from the paper
-#
 # This config assumes that you've run the following command before launching
 # this run:
 #   tune download microsoft/Phi-3-mini-4k-instruct --output-dir ./Phi-3-mini-4k-instruct --hf-token <HF_TOKEN> --ignore-patterns ""
@@ -19,12 +15,12 @@
 #
 # This config works best when the model is being fine-tuned on 2+ GPUs.
 # Single device full finetuning requires more memory optimizations. It's
-# best to use 7B_full_single_device.yaml for those cases
+# best to use mini_low_memory.yaml for those cases
 
 # Tokenizer
 tokenizer:
   _component_: torchtune.models.phi3.phi3_tokenizer
-  path: /data/users/kartikayk/cpts/Phi-3-mini-4k-instruct/tokenizer.model
+  path: /tmp/cpts/Phi-3-mini-4k-instruct/tokenizer.model
 
 # Dataset
 dataset:
@@ -39,13 +35,13 @@ model:
 
 checkpointer:
   _component_: torchtune.utils.FullModelHFCheckpointer
-  checkpoint_dir: /data/users/kartikayk/cpts/Phi-3-mini-4k-instruct
+  checkpoint_dir: /tmp/Phi-3-mini-4k-instruct
   checkpoint_files: [
     model-00001-of-00002.safetensors,
     model-00002-of-00002.safetensors
   ]
   recipe_checkpoint: null
-  output_dir: /data/users/kartikayk/cpts/Phi-3-mini-4k-instruct
+  output_dir: /tmp/Phi-3-mini-4k-instruct
   model_type: PHI3_MINI
 resume_from_checkpoint: False
 
@@ -74,5 +70,5 @@ dtype: bf16
 metric_logger:
   _component_: torchtune.utils.metric_logging.DiskLogger
   log_dir: ${output_dir}
-output_dir: /tmp/Mistral-7B-v0.1/
+output_dir: /tmp/Phi-3-mini-4k-instruct
 log_every_n_steps: null
diff --git a/recipes/configs/phi3/mini_full_low_memory.yaml b/recipes/configs/phi3/mini_full_low_memory.yaml
index 1427b9ad63..e2c96c9658 100644
--- a/recipes/configs/phi3/mini_full_low_memory.yaml
+++ b/recipes/configs/phi3/mini_full_low_memory.yaml
@@ -1,10 +1,6 @@
 # Config for single device full finetuning in full_finetune_single_device.py
 # using a Phi3 Mini 4K Instruct
 #
-# This config uses hyperparameters based on small set of experiments and information
-# available on various forums. These are not meant to replicate the numbers
-# from the paper
-#
 # This config assumes that you've run the following command before launching
 # this run:
 #   tune download microsoft/Phi-3-mini-4k-instruct --output-dir ./Phi-3-mini-4k-instruct --hf-token <HF_TOKEN> --ignore-patterns ""
@@ -14,19 +10,19 @@
 #   pip install bitsandbytes
 #
 # To launch on a single device, run the following command from root:
-#   tune run full_finetune_single_device --config recipes/config/phi3/mini_full_low_memory.yaml
+#   tune run full_finetune_single_device --config recipes/configs/phi3/mini_full_low_memory.yaml
 #
 # You can add specific overrides through the command line. For example
 # to override the checkpointer directory while launching training
 # you can run:
-#   tune run full_finetune_single_device --config recipes/config/phi3/mini_full_low_memory.yaml checkpointer.checkpoint_dir=<YOUR_CHECKPOINT_DIR>
+#   tune run full_finetune_single_device --config recipes/configs/phi3/mini_full_low_memory.yaml checkpointer.checkpoint_dir=<YOUR_CHECKPOINT_DIR>
 #
 # This config works only for training on single device.
 
 # Tokenizer
 tokenizer:
   _component_: torchtune.models.phi3.phi3_tokenizer
-  path: /data/users/kartikayk/cpts/Phi-3-mini-4k-instruct/tokenizer.model
+  path: /tmp/Phi-3-mini-4k-instruct/tokenizer.model
 
 # Dataset
 dataset:
@@ -41,13 +37,13 @@ model:
 
 checkpointer:
   _component_: torchtune.utils.FullModelHFCheckpointer
-  checkpoint_dir: /data/users/kartikayk/cpts/Phi-3-mini-4k-instruct
+  checkpoint_dir: /tmp/Phi-3-mini-4k-instruct
   checkpoint_files: [
     model-00001-of-00002.safetensors,
     model-00002-of-00002.safetensors
   ]
   recipe_checkpoint: null
-  output_dir: /data/users/kartikayk/cpts/Phi-3-mini-4k-instruct
+  output_dir: /tmp/Phi-3-mini-4k-instruct
   model_type: PHI3_MINI
 resume_from_checkpoint: False
 
@@ -79,5 +75,5 @@ compile: False
 metric_logger:
   _component_: torchtune.utils.metric_logging.DiskLogger
   log_dir: ${output_dir}
-output_dir: /tmp/Mistral-7B-v0.1/
+output_dir: /tmp/Phi-3-mini-4k-instruct
 log_every_n_steps: null