apunkt/doc-to-lora
1
0
Fork 0
mirror of https://github.com/SakanaAI/doc-to-lora.git synced 2026-07-23 17:01:04 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

install and run cmds

Browse source
This commit is contained in:
51616 2025-01-20 12:19:10 +00:00
parent 773094cb22
commit fd751f82e2
5 changed files with 8 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

8
batch_gemma_llama_vision.sh
View file

@ -20,11 +20,11 @@ conda activate /home/rujikorn_sakana_ai/.conda/envs/llm-modulator
accelerate launch --num_processes=8 --gradient_accumulation_steps=4 --gradient_clipping=1.0 \
--gpu_ids all --main_process_port 29554 intx_sft.py configs/pretrain_all.yaml \
--model_name_or_path=google/gemma-2-2b-it --num_train_epochs=5.1 --per_device_train_batch_size=4 \
--model_name_or_path=google/gemma-2-2b-it --num_train_epochs=5.1 --per_device_train_batch_size=8 \
--gradient_accumulation_steps=4 --per_device_eval_batch_size=4 --exp_setup=hyper_lora --aggregator_type=perceiver \
--target_modules=down_proj --num_blocks=1 --num_self_attends_per_block=16 \
--self_attention_widening_factor=1 --eval_steps=5000 --save_steps=5000 --learning_rate=2e-5 \
--neftune_noise_alpha=5 --use_light_weight_lora=True --light_weight_latent_size=512 \
--target_modules=down_proj,up_proj --num_blocks=1 --num_self_attends_per_block=32 \
--self_attention_widening_factor=4 --eval_steps=5000 --save_steps=5000 --learning_rate=2e-5 \
--neftune_noise_alpha=5 --use_light_weight_lora=True --light_weight_latent_size=1024 \
--load_best_model_at_end=True --metric_for_best_model=pwc_loss --add_negative_prompt=False \
--add_repeat_prompt=False --ctx_encoder_model_name_or_path=meta-llama/Llama-3.2-11B-Vision-Instruct

2
batch_llama_3_8b_1k_latent.sh
View file

@ -20,7 +20,7 @@ conda activate /home/rujikorn_sakana_ai/.conda/envs/llm-modulator
accelerate launch --num_processes=8 --gradient_accumulation_steps=4 --gradient_clipping=1.0 \
--gpu_ids all --main_process_port 29555 intx_sft.py configs/pretrain_all.yaml \
--model_name_or_path=meta-llama/Llama-3.1-8B-Instruct --num_train_epochs=5.1 --per_device_train_batch_size=4 \
--model_name_or_path=meta-llama/Llama-3.1-8B-Instruct --num_train_epochs=5.1 --per_device_train_batch_size=8 \
--gradient_accumulation_steps=4 --per_device_eval_batch_size=4 --exp_setup=hyper_lora --aggregator_type=perceiver \
--target_modules=down_proj,up_proj --num_blocks=1 --num_self_attends_per_block=32 \
--self_attention_widening_factor=4 --eval_steps=5000 --save_steps=5000 --learning_rate=2e-5 \

1
install.sh
View file

@ -1,2 +1,3 @@
pip install -e .
pip install flash-attn --no-build-isolation
# huggingface-cli login

1
pyproject.toml
View file

@ -28,6 +28,7 @@ dependencies = [
"flask",
"pandas",
"plotly",
"rouge-score",
]
[build-system]

1
requirements.txt
View file

@ -1,6 +1,7 @@
einops
jaxtyping
liger-kernel
rouge-score
transformers==4.46.3
deepspeed==0.15.4 # curretly 0.16.2
accelerate==1.2.1