From c2fc16477988dc34a6d8cf33e8cbc66f12656b80 Mon Sep 17 00:00:00 2001
From: Zach <zanussbaum@gmail.com>
Date: Wed, 5 Apr 2023 20:42:35 +0000
Subject: [PATCH] fix: configs

---
 configs/deepspeed/ds_config_gptj.json |  2 +-
 configs/train/finetune_gptj.yaml      | 10 +++++-----
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/configs/deepspeed/ds_config_gptj.json b/configs/deepspeed/ds_config_gptj.json
index 9e7a410a..3e933966 100644
--- a/configs/deepspeed/ds_config_gptj.json
+++ b/configs/deepspeed/ds_config_gptj.json
@@ -19,7 +19,7 @@
 		"device": "none"
 	  },
 	  "offload_optimizer": {
-		"device": "cpu"
+		"device": "none"
 	  },
 	  "allgather_partitions": true,
 	  "allgather_bucket_size": 5e8,
diff --git a/configs/train/finetune_gptj.yaml b/configs/train/finetune_gptj.yaml
index 1b42d780..f37283b3 100644
--- a/configs/train/finetune_gptj.yaml
+++ b/configs/train/finetune_gptj.yaml
@@ -2,7 +2,7 @@
 model_name: "EleutherAI/gpt-j-6B"
 tokenizer_name: "EleutherAI/gpt-j-6B"
 gradient_checkpointing: true
-save_name: "nomic-ai/gpt4all-gptj-multinode"
+save_name: "nomic-ai/gpt4all-gptj-multinode-deepspeed"
 
 # dataset
 streaming: false
@@ -12,13 +12,13 @@ max_length: 1024
 batch_size: 32
 
 # train dynamics
-lr: 4.0e-5
+lr: 2.0e-5
 min_lr: 0 
 weight_decay: 0.0
-eval_every: 100
+eval_every: 500
 eval_steps: 105
-save_every: 100
-log_grads_every: 100
+save_every: 500
+log_grads_every: 500
 output_dir: "ckpts/gpt4all-gptj-multinode"
 checkpoint: null
 lora: false