From 2b824ed81380972690a7a3477d7e9c3eef569c55 Mon Sep 17 00:00:00 2001 From: 51616 Date: Tue, 29 Jul 2025 07:00:03 +0000 Subject: [PATCH] scripts --- scripts/short_ctx/gemma_qa_short_ctx_perceiver.sh | 4 ++-- scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l.sh | 4 ++-- .../gemma_qa_short_ctx_perceiver_l2l_max_qas_per_sample_1.sh | 4 ++-- 3 files changed, 6 insertions(+), 6 deletions(-) diff --git a/scripts/short_ctx/gemma_qa_short_ctx_perceiver.sh b/scripts/short_ctx/gemma_qa_short_ctx_perceiver.sh index 76617d5..16a6994 100644 --- a/scripts/short_ctx/gemma_qa_short_ctx_perceiver.sh +++ b/scripts/short_ctx/gemma_qa_short_ctx_perceiver.sh @@ -13,9 +13,9 @@ echo "Using port: $port" uv run accelerate launch --main_process_port $port \ --num_processes=4 --gpu_ids all intx_sft.py $1 \ --model_name_or_path=google/gemma-2-2b-it \ ---max_steps=10_000 \ +--max_steps=30_000 \ --per_device_train_batch_size=-1 \ ---gradient_accumulation_steps=2 \ +--gradient_accumulation_steps=4 \ --per_device_eval_batch_size=64 \ --target_modules=down_proj \ --num_blocks=8 \ diff --git a/scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l.sh b/scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l.sh index 0540c7f..a495011 100644 --- a/scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l.sh +++ b/scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l.sh @@ -13,9 +13,9 @@ echo "Using port: $port" uv run accelerate launch --main_process_port $port \ --num_processes=4 --gpu_ids all intx_sft.py $1 \ --model_name_or_path=google/gemma-2-2b-it \ ---max_steps=10_000 \ +--max_steps=30_000 \ --per_device_train_batch_size=-1 \ ---gradient_accumulation_steps=4 \ +--gradient_accumulation_steps=8 \ --per_device_eval_batch_size=32 \ --target_modules=down_proj \ --num_blocks=8 \ diff --git a/scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l_max_qas_per_sample_1.sh b/scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l_max_qas_per_sample_1.sh index 06b898a..fb08d80 100644 --- a/scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l_max_qas_per_sample_1.sh +++ b/scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l_max_qas_per_sample_1.sh @@ -13,9 +13,9 @@ echo "Using port: $port" uv run accelerate launch --main_process_port $port \ --num_processes=4 --gpu_ids all intx_sft.py $1 \ --model_name_or_path=google/gemma-2-2b-it \ ---max_steps=10_000 \ +--max_steps=30_000 \ --per_device_train_batch_size=-1 \ ---gradient_accumulation_steps=4 \ +--gradient_accumulation_steps=8 \ --per_device_eval_batch_size=32 \ --target_modules=down_proj \ --num_blocks=8 \