minor change

app.py

@@ -1,13 +1,9 @@
-# A mirror to gradio launch stream
+# Copyright: DAMO Academy, Alibaba Group
 # By Xuan Phi Nguyen at DAMO Academy, Alibaba Group
 
+# Description:
 """
-Load FasterLlama with original VLLM codebase,
-
-require changing config names to LlamaForCausalLM
-
-tensor_parallel must == 1
-
+VLLM-based demo script to launch Language chat model for South East Asian Languages
 """
 
 
@@ -75,6 +71,11 @@ HF_MODEL_NAME=DAMO-NLP-SG/seal-13b-chat-a
 
 # ==============================
 print(f'DEBUG mode: {DEBUG}')
+print(f'Torch version: {torch.__version__}')
+try:
+    print(f'Torch CUDA version: {torch.version.cuda}')
+except Exception as e:
+    print(f'Failed to print cuda version: {e}')
 
 
 
@@ -976,9 +977,12 @@ export CUDA_VISIBLE_DEVICES=0
 export MODEL_PATH=seal_13b_a
 export MODEL_PATH=${dataroot}/hf_train/pretrain_lm/swpn/merlion13s108Hi8kPretFlCW12k.LMFromHf.a.gc.t5k0.vizhthid.mean_std.TrainTask.NLNL.Multi.Vi.SeaV2Cq13M.SeaV2Cq13M.m4k.b8.lr1e5.linear.wa0k.ms858k.grac1.se1.8g.v4c.zfsdp/step_6000
 
-export MODEL_PATH=${dataroot}/hf_train/pretrain_lm/swpn/mer13s108Hi16kPretFlCWNLP12k_SFT2.LMFromHf.a.gc.t5k0.vizhthid.mean_std.TrainTask.NLNL.Multi.Vi.Sft2Censor.Sft2Censor.m4k.b8.lr1e5.linear.wa0k.ms1144k.grac1.se1.6g.v4c.zfsdp/step_2000
+export MODEL_PATH=${dataroot}/hf_train/pretrain_lm/swpn/mer13s108Hi16kPretFlCWNLP12k_SFT2.LMFromHf.a.gc.t5k0.vizhthid.mean_std.TrainTask.NLNL.Multi.Vi.Sft2Censor.Sft2Censor.m4k.b8.lr1e5.linear.wa0k.ms1144k.grac1.se1.6g.v4c.zfsdp/step_4000
+# 70-30 model
+export MODEL_PATH=${dataroot}/hf_train/pretrain_lm/swpn/mer13s108Hi16kPretFlCWNLP12k_SFT2.LMFromHf.a.gc.t5k0.vizhthid.mean_std.TrainTask.NLNL.Multi.BgSft2aCensor0a.BgSft2Cens.BgSft2Cens.m4k.b2.lr1e5.linear.wa0k.ms4577k.grac1.se1.6g.v4c73.zfsdp/step_500
 export PORT=8799
 export BLOCK_ZH=1
+export DEBUG=0
 python app.py