Fix bug 0917 (#74)

Author: Jintao-Huang

examples/pytorch/llm/scripts/baichuan2_7b_chat/lora_ddp/infer.sh

@@ -4,12 +4,12 @@ python src/llm_infer.py \
     --sft_type lora \
     --template_type baichuan \
     --dtype bf16 \
-    --ckpt_dir "runs/baichuan2-7b-chat/vx_xxx/checkpoint-xxx" \
+    --ckpt_dir "output/baichuan2-7b-chat/vx_xxx/checkpoint-xxx" \
     --eval_human false \
     --dataset damo-agent-mini-zh \
     --max_length 4096 \
     --max_new_tokens 2048 \
     --temperature 0.9 \
-    --top_k 50 \
+    --top_k 20 \
     --top_p 0.9 \
     --do_sample true \

examples/pytorch/llm/scripts/baichuan2_7b_chat/lora_ddp/sft.sh

@@ -10,7 +10,7 @@ torchrun \
     --sft_type lora \
     --template_type baichuan \
     --dtype bf16 \
-    --output_dir runs \
+    --output_dir output \
     --ddp_backend nccl \
     --dataset damo-agent-mini-zh \
     --train_dataset_sample -1 \

examples/pytorch/llm/scripts/baichuan2_7b_chat/qlora/infer.sh

@@ -4,14 +4,14 @@ python src/llm_infer.py \
     --sft_type lora \
     --template_type baichuan \
     --dtype bf16 \
-    --ckpt_dir "runs/baichuan2-7b-chat/vx_xxx/checkpoint-xxx" \
+    --ckpt_dir "output/baichuan2-7b-chat/vx_xxx/checkpoint-xxx" \
     --eval_human false \
     --dataset advertise-gen \
     --max_length 2048 \
     --quantization_bit 4 \
     --bnb_4bit_comp_dtype bf16 \
     --max_new_tokens 1024 \
     --temperature 0.9 \
-    --top_k 50 \
+    --top_k 20 \
     --top_p 0.9 \
     --do_sample true \

examples/pytorch/llm/scripts/baichuan2_7b_chat/qlora/sft.sh

@@ -6,7 +6,7 @@ python src/llm_sft.py \
     --sft_type lora \
     --template_type baichuan \
     --dtype bf16 \
-    --output_dir runs \
+    --output_dir output \
     --dataset advertise-gen \
     --train_dataset_sample -1 \
     --num_train_epochs 1 \

examples/pytorch/llm/scripts/chatglm2_6b/lora_ddp/infer.sh

@@ -4,12 +4,12 @@ python src/llm_infer.py \
     --sft_type lora \
     --template_type chatglm2 \
     --dtype bf16 \
-    --ckpt_dir "runs/chatglm2-6b/vx_xxx/checkpoint-xxx" \
+    --ckpt_dir "output/chatglm2-6b/vx_xxx/checkpoint-xxx" \
     --eval_human false \
     --dataset code-python-zh \
     --max_length 8192 \
     --max_new_tokens 1024 \
     --temperature 0.9 \
-    --top_k 50 \
+    --top_k 20 \
     --top_p 0.9 \
     --do_sample true \

examples/pytorch/llm/scripts/chatglm2_6b/lora_ddp/sft.sh

@@ -10,7 +10,7 @@ torchrun \
     --sft_type lora \
     --template_type chatglm2 \
     --dtype bf16 \
-    --output_dir runs \
+    --output_dir output \
     --ddp_backend nccl \
     --dataset code-python-zh \
     --train_dataset_sample -1 \

examples/pytorch/llm/scripts/internlm_7b_chat/lora_ddp/infer.sh renamed to examples/pytorch/llm/scripts/internlm_20b/lora_ddp/infer.sh

@@ -1,15 +1,15 @@
 CUDA_VISIBLE_DEVICES=0 \
 python src/llm_infer.py \
-    --model_type internlm-7b-chat \
+    --model_type internlm-20b \
     --sft_type lora \
-    --template_type internlm \
+    --template_type default-generation \
     --dtype bf16 \
-    --ckpt_dir "runs/internlm-7b-chat/vx_xxx/checkpoint-xxx" \
+    --ckpt_dir "output/internlm-20b/vx_xxx/checkpoint-xxx" \
     --eval_human false \
     --dataset jd-zh \
     --max_length 2048 \
     --max_new_tokens 1024 \
     --temperature 0.9 \
-    --top_k 50 \
+    --top_k 20 \
     --top_p 0.9 \
     --do_sample true \

examples/pytorch/llm/scripts/internlm_7b_chat/lora_ddp/sft.sh renamed to examples/pytorch/llm/scripts/internlm_20b/lora_ddp/sft.sh

@@ -4,11 +4,11 @@ torchrun \
     --nproc_per_node=$nproc_per_node \
     --master_port 29500 \
     src/llm_sft.py \
-    --model_type internlm-7b-chat \
+    --model_type internlm-20b \
     --sft_type lora \
-    --template_type internlm \
+    --template_type default-generation \
     --dtype bf16 \
-    --output_dir runs \
+    --output_dir output \
     --ddp_backend nccl \
     --dataset jd-zh \
     --train_dataset_sample -1 \
@@ -30,6 +30,6 @@ torchrun \
     --save_total_limit 2 \
     --logging_steps 10 \
     --push_to_hub false \
-    --hub_model_id internlm-7b-lora \
+    --hub_model_id internlm-20b-lora \
     --hub_private_repo true \
     --hub_token 'your-sdk-token' \

examples/pytorch/llm/scripts/internlm_20b/qlora/infer.sh

@@ -0,0 +1,17 @@
+CUDA_VISIBLE_DEVICES=0 \
+python src/llm_infer.py \
+    --model_type internlm-20b \
+    --sft_type lora \
+    --template_type default-generation \
+    --dtype bf16 \
+    --ckpt_dir "output/internlm-20b/vx_xxx/checkpoint-xxx" \
+    --eval_human false \
+    --dataset advertise-gen \
+    --max_length 2048 \
+    --quantization_bit 4 \
+    --bnb_4bit_comp_dtype bf16 \
+    --max_new_tokens 1024 \
+    --temperature 0.9 \
+    --top_k 20 \
+    --top_p 0.9 \
+    --do_sample true \

examples/pytorch/llm/scripts/internlm_20b/qlora/sft.sh

@@ -0,0 +1,34 @@
+# Experimental environment: A10
+# 20GB GPU memory
+CUDA_VISIBLE_DEVICES=0 \
+python src/llm_sft.py \
+    --model_type internlm-20b \
+    --sft_type lora \
+    --template_type default-generation \
+    --dtype bf16 \
+    --output_dir output \
+    --dataset advertise-gen \
+    --train_dataset_sample -1 \
+    --num_train_epochs 1 \
+    --max_length 2048 \
+    --quantization_bit 4 \
+    --bnb_4bit_comp_dtype bf16 \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --lora_dropout_p 0. \
+    --lora_target_modules ALL \
+    --gradient_checkpointing false \
+    --batch_size 1 \
+    --weight_decay 0. \
+    --learning_rate 1e-4 \
+    --gradient_accumulation_steps 16 \
+    --max_grad_norm 0.5 \
+    --warmup_ratio 0.03 \
+    --eval_steps 100 \
+    --save_steps 100 \
+    --save_total_limit 2 \
+    --logging_steps 10 \
+    --push_to_hub false \
+    --hub_model_id internlm-20b-qlora \
+    --hub_private_repo true \
+    --hub_token 'your-sdk-token' \