Feature/llama factory llama2 pipeline (#89)

* added llama-factory under llm_rl

* added sft training bash

* added datasets from llama-factory; will delete later

* finished llama-2-13b train and inference

* fixed minor errors

* changed config

* added deepspeed config

* added more training config to train bash

* adding fix for wandb tags and distributed ranks

* added fastchat data to replicate training for 2k

.gitignore

@@ -20,6 +20,7 @@ llm_ft/checkpoints/*
 llm_ft/*_checkpoints/*
 !**/dummy_conversation.json
 !llm_ft/deepspeed_config_s2.json
+!llm_rl/data/*.json
 
 # Editor
 .idea
@@ -193,4 +194,8 @@ cython_debug/
 #  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
 #  and can be added to the global gitignore or merged into this file.  For a more nuclear
 #  option (not recommended) you can uncomment the following to ignore the entire idea folder.
-#.idea/
+#.idea/
+
+./llm_rl/preprocess/GPT4-4_Redis_Easy_No_Slide
+
+llm_rl/*cache/

llm_rl/cli_inference-llama-2-13b.sh

@@ -0,0 +1,6 @@
+python src/cli_demo.py \
+    --model_name_or_path meta-llama/Llama-2-13b-hf \
+    --cache_dir ./model_cache \
+    --template llama2-sotopia \
+    --finetuning_type lora \
+    --checkpoint_dir /workspace/sotopia-llm/llm_rl/llama2-13b-sft_cache/checkpoint-35