short ctx exp scripts

scripts/short_ctx/gemma_qa_short_ctx.sh

@@ -1,7 +1,7 @@
 #!/bin/bash
 #SBATCH --job-name=ctxlora
 #SBATCH --nodes=1
-#SBATCH --partition=sakura-gpu
+#SBATCH --partition=a3
 #SBATCH --gpus=4
 #SBATCH --output=slurm_logs/%x-%j.out
 #SBATCH --error=slurm_logs/%x-%j.out
@@ -31,4 +31,4 @@ uv run accelerate launch --num_processes=4 --gradient_accumulation_steps=2 --gra
 --per_rank_gen=True \
 --per_layer_processing=True \
 --gen_lora_l1_reg_coef=0.1 \
---logging_steps=50
+--logging_steps=50

scripts/short_ctx/gemma_qa_short_ctx_exp_default.sh

@@ -15,7 +15,7 @@ uv run accelerate launch --main_process_port $port \
 --model_name_or_path=google/gemma-2-2b-it \
 --max_steps=12_000 \
 --per_device_train_batch_size=-1 \
---gradient_accumulation_steps=2 \
+--gradient_accumulation_steps=5 \
 --per_device_eval_batch_size=64 \
 --target_modules=down_proj \
 --num_blocks=8 \
@@ -23,6 +23,7 @@ uv run accelerate launch --main_process_port $port \
 --n_latent_queries=208 \
 --num_pre_head_layers=1 \
 --lora_r=8 \
+--eval_strategy=no \
 --eval_steps=1000 \
 --save_steps=1000 \
 --learning_rate=4e-5 \

scripts/short_ctx/gemma_qa_short_ctx_exp_default_max_qas_per_sample_1.sh

@@ -0,0 +1,42 @@
+#!/bin/bash
+#SBATCH --job-name=ctxlora
+#SBATCH --nodes=1
+#SBATCH --partition=a3
+#SBATCH --gpus=4
+#SBATCH --output=slurm_logs/%x-%j.out
+#SBATCH --error=slurm_logs/%x-%j.out
+
+port=$((10000 + ($SLURM_JOBID % 50000)))
+echo "Using port: $port"
+
+# --gradient_accumulation_steps=8 --gradient_clipping=1.0
+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=12_000 \
+--per_device_train_batch_size=-1 \
+--gradient_accumulation_steps=5 \
+--per_device_eval_batch_size=64 \
+--target_modules=down_proj \
+--num_blocks=8 \
+--num_self_attn_per_block=0 \
+--n_latent_queries=208 \
+--num_pre_head_layers=1 \
+--lora_r=8 \
+--eval_strategy=no \
+--eval_steps=1000 \
+--save_steps=1000 \
+--learning_rate=4e-5 \
+--lora_dropout=0.0 \
+--neftune_noise_alpha=5 \
+--add_negative_prompt=False \
+--add_repeat_prompt=False \
+--use_sequence_packing=True \
+--max_qas_per_sample=1 \
+--max_qas_len=2048 \
+--max_packed_inp_len=8192 \
+--max_packed_ctx_len=16384 \
+--per_rank_gen=True \
+--per_layer_processing=True \
+--gen_lora_l1_reg_coef=0.1 \
+--logging_steps=50

scripts/short_ctx/gemma_qa_short_ctx_exp_default_more_prehead.sh

@@ -0,0 +1,40 @@
+#!/bin/bash
+#SBATCH --job-name=ctxlora
+#SBATCH --nodes=1
+#SBATCH --partition=a3
+#SBATCH --gpus=4
+#SBATCH --output=slurm_logs/%x-%j.out
+#SBATCH --error=slurm_logs/%x-%j.out
+
+port=$((10000 + ($SLURM_JOBID % 50000)))
+echo "Using port: $port"
+
+# --gradient_accumulation_steps=8 --gradient_clipping=1.0
+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=12_000 \
+--per_device_train_batch_size=-1 \
+--gradient_accumulation_steps=5 \
+--per_device_eval_batch_size=64 \
+--target_modules=down_proj \
+--num_blocks=8 \
+--num_self_attn_per_block=0 \
+--n_latent_queries=208 \
+--num_pre_head_layers=4 \
+--lora_r=8 \
+--eval_steps=1000 \
+--save_steps=1000 \
+--learning_rate=4e-5 \
+--lora_dropout=0.0 \
+--neftune_noise_alpha=5 \
+--add_negative_prompt=False \
+--add_repeat_prompt=False \
+--use_sequence_packing=True \
+--max_qas_len=2048 \
+--max_packed_inp_len=8192 \
+--max_packed_ctx_len=16384 \
+--per_rank_gen=True \
+--per_layer_processing=True \
+--gen_lora_l1_reg_coef=0.1 \
+--logging_steps=50

scripts/short_ctx/gemma_qa_short_ctx_exp_default_more_prehead_max_qas_per_sample_1.sh

@@ -0,0 +1,41 @@
+#!/bin/bash
+#SBATCH --job-name=ctxlora
+#SBATCH --nodes=1
+#SBATCH --partition=a3
+#SBATCH --gpus=4
+#SBATCH --output=slurm_logs/%x-%j.out
+#SBATCH --error=slurm_logs/%x-%j.out
+
+port=$((10000 + ($SLURM_JOBID % 50000)))
+echo "Using port: $port"
+
+# --gradient_accumulation_steps=8 --gradient_clipping=1.0
+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=12_000 \
+--per_device_train_batch_size=-1 \
+--gradient_accumulation_steps=5 \
+--per_device_eval_batch_size=64 \
+--target_modules=down_proj \
+--num_blocks=8 \
+--num_self_attn_per_block=0 \
+--n_latent_queries=208 \
+--num_pre_head_layers=4 \
+--lora_r=8 \
+--eval_steps=1000 \
+--save_steps=1000 \
+--learning_rate=4e-5 \
+--lora_dropout=0.0 \
+--neftune_noise_alpha=5 \
+--add_negative_prompt=False \
+--add_repeat_prompt=False \
+--use_sequence_packing=True \
+--max_qas_per_sample=1 \
+--max_qas_len=2048 \
+--max_packed_inp_len=8192 \
+--max_packed_ctx_len=16384 \
+--per_rank_gen=True \
+--per_layer_processing=True \
+--gen_lora_l1_reg_coef=0.1 \
+--logging_steps=50

scripts/short_ctx/gemma_qa_short_ctx_old_conf.sh

@@ -0,0 +1,42 @@
+#!/bin/bash
+#SBATCH --job-name=ctxlora
+#SBATCH --nodes=1
+#SBATCH --partition=a3
+#SBATCH --gpus=4
+#SBATCH --output=slurm_logs/%x-%j.out
+#SBATCH --error=slurm_logs/%x-%j.out
+
+port=$((10000 + ($SLURM_JOBID % 50000)))
+echo "Using port: $port"
+
+# --gradient_accumulation_steps=8 --gradient_clipping=1.0
+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 \
+--num_train_epochs=5 \
+--per_device_train_batch_size=-1 \
+--gradient_accumulation_steps=8 \
+--per_device_eval_batch_size=64 \
+--target_modules=down_proj \
+--num_blocks=9 \
+--num_self_attn_per_block=0 \
+--n_latent_queries=208 \
+--num_pre_head_layers=1 \
+--lora_r=8 \
+--eval_strategy=no \
+--eval_steps=1000 \
+--save_steps=1000 \
+--learning_rate=4e-5 \
+--lora_dropout=0.0 \
+--neftune_noise_alpha=5 \
+--add_negative_prompt=False \
+--add_repeat_prompt=False \
+--use_sequence_packing=True \
+--max_qas_len=2048 \
+--max_qas_per_sample=1 \
+--max_packed_inp_len=8192 \
+--max_packed_ctx_len=16384 \
+--per_rank_gen=True \
+--per_layer_processing=True \
+--gen_lora_l1_reg_coef=0.1 \
+--logging_steps=50

scripts/short_ctx/gemma_qa_short_ctx_old_conf_8_gpus.sh

@@ -0,0 +1,42 @@
+#!/bin/bash
+#SBATCH --job-name=ctxlora
+#SBATCH --nodes=1
+#SBATCH --partition=a3
+#SBATCH --gpus=8
+#SBATCH --output=slurm_logs/%x-%j.out
+#SBATCH --error=slurm_logs/%x-%j.out
+
+port=$((10000 + ($SLURM_JOBID % 50000)))
+echo "Using port: $port"
+
+# --gradient_accumulation_steps=8 --gradient_clipping=1.0
+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 \
+--num_train_epochs=5 \
+--per_device_train_batch_size=-1 \
+--gradient_accumulation_steps=4 \
+--per_device_eval_batch_size=64 \
+--target_modules=down_proj \
+--num_blocks=8 \
+--num_self_attn_per_block=0 \
+--n_latent_queries=208 \
+--num_pre_head_layers=4 \
+--lora_r=8 \
+--eval_strategy=no \
+--eval_steps=1000 \
+--save_steps=1000 \
+--learning_rate=4e-5 \
+--lora_dropout=0.0 \
+--neftune_noise_alpha=5 \
+--add_negative_prompt=False \
+--add_repeat_prompt=False \
+--use_sequence_packing=True \
+--max_qas_len=2048 \
+--max_qas_per_sample=1 \
+--max_packed_inp_len=8192 \
+--max_packed_ctx_len=16384 \
+--per_rank_gen=True \
+--per_layer_processing=True \
+--gen_lora_l1_reg_coef=0.1 \
+--logging_steps=50

scripts/short_ctx/gemma_qa_short_ctx_old_conf_more_grad_acc.sh

@@ -0,0 +1,41 @@
+#!/bin/bash
+#SBATCH --job-name=ctxlora
+#SBATCH --nodes=1
+#SBATCH --partition=a3
+#SBATCH --gpus=4
+#SBATCH --output=slurm_logs/%x-%j.out
+#SBATCH --error=slurm_logs/%x-%j.out
+
+port=$((10000 + ($SLURM_JOBID % 50000)))
+echo "Using port: $port"
+
+# --gradient_accumulation_steps=8 --gradient_clipping=1.0
+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=12_000 \
+--per_device_train_batch_size=-1 \
+--gradient_accumulation_steps=16 \
+--per_device_eval_batch_size=64 \
+--target_modules=down_proj \
+--num_blocks=8 \
+--num_self_attn_per_block=0 \
+--n_latent_queries=208 \
+--num_pre_head_layers=4 \
+--lora_r=8 \
+--eval_steps=1000 \
+--save_steps=1000 \
+--learning_rate=4e-5 \
+--lora_dropout=0.0 \
+--neftune_noise_alpha=5 \
+--add_negative_prompt=False \
+--add_repeat_prompt=False \
+--use_sequence_packing=True \
+--max_qas_len=2048 \
+--max_qas_per_sample=1 \
+--max_packed_inp_len=8192 \
+--max_packed_ctx_len=16384 \
+--per_rank_gen=True \
+--per_layer_processing=True \
+--gen_lora_l1_reg_coef=0.1 \
+--logging_steps=50

scripts/short_ctx/gemma_qa_short_ctx_perceiver.sh

@@ -13,7 +13,7 @@ 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=12_000 \
+--max_steps=10_000 \
 --per_device_train_batch_size=-1 \
 --gradient_accumulation_steps=2 \
 --per_device_eval_batch_size=64 \
@@ -23,6 +23,7 @@ uv run accelerate launch --main_process_port $port \
 --n_latent_queries=208 \
 --num_pre_head_layers=1 \
 --lora_r=8 \
+--eval_strategy=no \
 --eval_steps=1000 \
 --save_steps=1000 \
 --learning_rate=4e-5 \

scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l.sh

@@ -13,7 +13,7 @@ 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=12_000 \
+--max_steps=10_000 \
 --per_device_train_batch_size=-1 \
 --gradient_accumulation_steps=4 \
 --per_device_eval_batch_size=32 \
@@ -37,5 +37,5 @@ uv run accelerate launch --main_process_port $port \
 --max_packed_ctx_len=8192 \
 --per_rank_gen=True \
 --per_layer_processing=True \
---gen_lora_l1_reg_coef=0.1 \
+--gen_lora_l1_reg_coef=0.0 \
 --logging_steps=50

scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l_128_max_qas_len.sh

@@ -0,0 +1,41 @@
+#!/bin/bash
+#SBATCH --job-name=ctxlora
+#SBATCH --nodes=1
+#SBATCH --partition=a3
+#SBATCH --gpus=4
+#SBATCH --output=slurm_logs/%x-%j.out
+#SBATCH --error=slurm_logs/%x-%j.out
+
+port=$((10000 + ($SLURM_JOBID % 50000)))
+echo "Using port: $port"
+
+# --gradient_accumulation_steps=8 --gradient_clipping=1.0
+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=50_000 \
+--per_device_train_batch_size=-1 \
+--gradient_accumulation_steps=4 \
+--per_device_eval_batch_size=32 \
+--target_modules=down_proj \
+--num_blocks=8 \
+--num_self_attn_per_block=3 \
+--ctx_encoder_type=per_layer_activations \
+--n_latent_queries=8 \
+--num_pre_head_layers=1 \
+--lora_r=8 \
+--eval_steps=1000 \
+--save_steps=1000 \
+--learning_rate=4e-5 \
+--lora_dropout=0.0 \
+--neftune_noise_alpha=5 \
+--add_negative_prompt=False \
+--add_repeat_prompt=False \
+--use_sequence_packing=True \
+--max_qas_len=128 \
+--max_packed_inp_len=4096 \
+--max_packed_ctx_len=4096 \
+--per_rank_gen=True \
+--per_layer_processing=True \
+--gen_lora_l1_reg_coef=0.0 \
+--logging_steps=50

scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l_32_max_qas_len.sh

@@ -0,0 +1,41 @@
+#!/bin/bash
+#SBATCH --job-name=ctxlora
+#SBATCH --nodes=1
+#SBATCH --partition=a3
+#SBATCH --gpus=4
+#SBATCH --output=slurm_logs/%x-%j.out
+#SBATCH --error=slurm_logs/%x-%j.out
+
+port=$((10000 + ($SLURM_JOBID % 50000)))
+echo "Using port: $port"
+
+# --gradient_accumulation_steps=8 --gradient_clipping=1.0
+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=50_000 \
+--per_device_train_batch_size=-1 \
+--gradient_accumulation_steps=4 \
+--per_device_eval_batch_size=32 \
+--target_modules=down_proj \
+--num_blocks=8 \
+--num_self_attn_per_block=3 \
+--ctx_encoder_type=per_layer_activations \
+--n_latent_queries=8 \
+--num_pre_head_layers=1 \
+--lora_r=8 \
+--eval_steps=1000 \
+--save_steps=1000 \
+--learning_rate=4e-5 \
+--lora_dropout=0.0 \
+--neftune_noise_alpha=5 \
+--add_negative_prompt=False \
+--add_repeat_prompt=False \
+--use_sequence_packing=True \
+--max_qas_len=32 \
+--max_packed_inp_len=4096 \
+--max_packed_ctx_len=4096 \
+--per_rank_gen=True \
+--per_layer_processing=True \
+--gen_lora_l1_reg_coef=0.0 \
+--logging_steps=50

scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l_512_max_qas_len.sh

@@ -0,0 +1,41 @@
+#!/bin/bash
+#SBATCH --job-name=ctxlora
+#SBATCH --nodes=1
+#SBATCH --partition=a3
+#SBATCH --gpus=4
+#SBATCH --output=slurm_logs/%x-%j.out
+#SBATCH --error=slurm_logs/%x-%j.out
+
+port=$((10000 + ($SLURM_JOBID % 50000)))
+echo "Using port: $port"
+
+# --gradient_accumulation_steps=8 --gradient_clipping=1.0
+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=50_000 \
+--per_device_train_batch_size=-1 \
+--gradient_accumulation_steps=4 \
+--per_device_eval_batch_size=32 \
+--target_modules=down_proj \
+--num_blocks=8 \
+--num_self_attn_per_block=3 \
+--ctx_encoder_type=per_layer_activations \
+--n_latent_queries=8 \
+--num_pre_head_layers=1 \
+--lora_r=8 \
+--eval_steps=1000 \
+--save_steps=1000 \
+--learning_rate=4e-5 \
+--lora_dropout=0.0 \
+--neftune_noise_alpha=5 \
+--add_negative_prompt=False \
+--add_repeat_prompt=False \
+--use_sequence_packing=True \
+--max_qas_len=512 \
+--max_packed_inp_len=4096 \
+--max_packed_ctx_len=4096 \
+--per_rank_gen=True \
+--per_layer_processing=True \
+--gen_lora_l1_reg_coef=0.1 \
+--logging_steps=50

scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l_64_max_qas_len.sh

@@ -0,0 +1,41 @@
+#!/bin/bash
+#SBATCH --job-name=ctxlora
+#SBATCH --nodes=1
+#SBATCH --partition=a3
+#SBATCH --gpus=4
+#SBATCH --output=slurm_logs/%x-%j.out
+#SBATCH --error=slurm_logs/%x-%j.out
+
+port=$((10000 + ($SLURM_JOBID % 50000)))
+echo "Using port: $port"
+
+# --gradient_accumulation_steps=8 --gradient_clipping=1.0
+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=50_000 \
+--per_device_train_batch_size=-1 \
+--gradient_accumulation_steps=4 \
+--per_device_eval_batch_size=32 \
+--target_modules=down_proj \
+--num_blocks=8 \
+--num_self_attn_per_block=3 \
+--ctx_encoder_type=per_layer_activations \
+--n_latent_queries=8 \
+--num_pre_head_layers=1 \
+--lora_r=8 \
+--eval_steps=1000 \
+--save_steps=1000 \
+--learning_rate=4e-5 \
+--lora_dropout=0.0 \
+--neftune_noise_alpha=5 \
+--add_negative_prompt=False \
+--add_repeat_prompt=False \
+--use_sequence_packing=True \
+--max_qas_len=64 \
+--max_packed_inp_len=4096 \
+--max_packed_ctx_len=4096 \
+--per_rank_gen=True \
+--per_layer_processing=True \
+--gen_lora_l1_reg_coef=0.0 \
+--logging_steps=50

scripts/short_ctx/gemma_qa_short_ctx_perceiver_l2l_max_qas_per_sample_1.sh

@@ -0,0 +1,42 @@
+#!/bin/bash
+#SBATCH --job-name=ctxlora
+#SBATCH --nodes=1
+#SBATCH --partition=a3
+#SBATCH --gpus=4
+#SBATCH --output=slurm_logs/%x-%j.out
+#SBATCH --error=slurm_logs/%x-%j.out
+
+port=$((10000 + ($SLURM_JOBID % 50000)))
+echo "Using port: $port"
+
+# --gradient_accumulation_steps=8 --gradient_clipping=1.0
+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 \
+--per_device_train_batch_size=-1 \
+--gradient_accumulation_steps=4 \
+--per_device_eval_batch_size=32 \
+--target_modules=down_proj \
+--num_blocks=8 \
+--num_self_attn_per_block=3 \
+--ctx_encoder_type=per_layer_activations \
+--n_latent_queries=8 \
+--num_pre_head_layers=1 \
+--lora_r=8 \
+--eval_steps=1000 \
+--save_steps=1000 \
+--learning_rate=4e-5 \
+--lora_dropout=0.0 \
+--neftune_noise_alpha=5 \
+--add_negative_prompt=False \
+--add_repeat_prompt=False \
+--use_sequence_packing=True \
+--max_qas_len=2048 \
+--max_qas_per_sample=1 \
+--max_packed_inp_len=4096 \
+--max_packed_ctx_len=8192 \
+--per_rank_gen=True \
+--per_layer_processing=True \
+--gen_lora_l1_reg_coef=0.0 \
+--logging_steps=50

scripts/short_ctx/gemma_qa_short_ctx_perceiver_max_qas_per_sample_1.sh

@@ -0,0 +1,41 @@
+#!/bin/bash
+#SBATCH --job-name=ctxlora
+#SBATCH --nodes=1
+#SBATCH --partition=a3
+#SBATCH --gpus=4
+#SBATCH --output=slurm_logs/%x-%j.out
+#SBATCH --error=slurm_logs/%x-%j.out
+
+port=$((10000 + ($SLURM_JOBID % 50000)))
+echo "Using port: $port"
+
+# --gradient_accumulation_steps=8 --gradient_clipping=1.0
+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 \
+--per_device_train_batch_size=-1 \
+--gradient_accumulation_steps=2 \
+--per_device_eval_batch_size=64 \
+--target_modules=down_proj \
+--num_blocks=8 \
+--num_self_attn_per_block=3 \
+--n_latent_queries=208 \
+--num_pre_head_layers=1 \
+--lora_r=8 \
+--eval_steps=1000 \
+--save_steps=1000 \
+--learning_rate=4e-5 \
+--lora_dropout=0.0 \
+--neftune_noise_alpha=5 \
+--add_negative_prompt=False \
+--add_repeat_prompt=False \
+--use_sequence_packing=True \
+--max_qas_len=2048 \
+--max_qas_per_sample=1 \
+--max_packed_inp_len=8192 \
+--max_packed_ctx_len=16384 \
+--per_rank_gen=True \
+--per_layer_processing=True \
+--gen_lora_l1_reg_coef=0.1 \
+--logging_steps=50