update pyproject pip install package

llm_ft/pyproject.toml

@@ -21,7 +21,7 @@ dependencies = [
 [project.optional-dependencies]
 model_worker = ["accelerate>=0.21", "peft", "sentencepiece", "torch", "transformers>=4.31.0", "protobuf"]
 webui = ["gradio"]
-train = ["einops", "flash-attn>=2.0", "wandb"]
+train = ["einops", "flash-attn>=2.0", "wandb", "deepspeed", "peft", "bitsandbytes", "scipy", "sentencepiece"]
 llm_judge = ["openai", "anthropic>=0.3", "ray"]
 dev = ["black==23.3.0", "pylint==2.8.2"]
 

llm_ft/qlora.sh

@@ -0,0 +1,25 @@
+deepspeed fastchat/train/train_lora.py \
+    --model_name_or_path ./vicuna-7b-1.5 \
+    --lora_r 8 \
+    --lora_alpha 16 \
+    --lora_dropout 0.05 \
+    --data_path ./data/dummy_conversation.json \
+    --bf16 True \
+    --output_dir ./checkpoints \
+    --num_train_epochs 3 \
+    --per_device_train_batch_size 1 \
+    --per_device_eval_batch_size 1 \
+    --gradient_accumulation_steps 1 \
+    --evaluation_strategy "no" \
+    --save_strategy "steps" \
+    --save_steps 1200 \
+    --save_total_limit 100 \
+    --learning_rate 2e-5 \
+    --weight_decay 0. \
+    --warmup_ratio 0.03 \
+    --lr_scheduler_type "cosine" \
+    --logging_steps 1 \
+    --tf32 True \
+    --model_max_length 2048 \
+    --q_lora True \
+    --deepspeed playground/deepspeed_config_s2.json