change to pipline

Dockerfile

@@ -17,5 +17,5 @@ ENV HOME=/home/user \
 WORKDIR $HOME/app
 
 COPY --chown=user . $HOME/app
-
+RUN ./tests.sh
 CMD ["streamlit", "run", "app.py"]

app.py

@@ -1,40 +1,23 @@
-from t5 import t5
+import subprocess
 import torch
 import streamlit as st
-from streamlit import session_state as ses
-import timeit
-import random
+from streamlit import session_state as state
+import streamlit_ace
+import pipline
 
-text_list = "Prospecting Academy is a 6 weeks online course designed for business owners, marketing and sales leaders of B2B scaleup companies to help you".split()
+if "app" not in state:
+    state.app = "model"
+    state.out = ""
+st.title("Streamlit using Huggingface Transformers and langchain")
+in_area = st.container()
+out_area = st.container()
 
 
-def random_sentence():
-    return " ".join(
-        list(random.choice(text_list) for i in range(random.randint(4, 20)))
-    )
 
+def __run_pipline():
+    out_area.markdown(":green[Running pipline]")
+    out_area.text(pipline.chain_TI(state.input_text))
 
-if "page" not in ses:
-    ses["page"] = "run"
-    ses.out = ""
-    ses.time = ""
 
-
-def __run_t5():
-    try:
-        ses.out = t5(ses.input_text)
-    except:
-        ses.out = "error"
-
-
-def __test_t5():
-    try:
-        ses.time = timeit.timeit("t5(random_sentence())", globals=globals(), number=20)
-    except:
-        ses.out = "error"
-
-
-st.text_input("enter text", key="input_text", on_change=__run_t5)
-st.write(ses.out)
-if st.button("average time", on_click=__test_t5):
-    st.metric("time", ses.time)
+in_area.text_area("input_text", key="input_text")
+in_area.button("run", on_click=__run_pipline)

model.py

@@ -0,0 +1,24 @@
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+
+torch_dtype = torch.bfloat16
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+model_name = "bigscience/bloomz-1b7"
+
+
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto", device_map="auto").to(device=device)
+
+
+def run(text,**kargs):
+    inputs = tokenizer.encode(text=text, return_tensors="pt").to(device=device)
+    outputs = model.generate(inputs,**kargs)
+    return tokenizer.decode(outputs[0])
+
+
+
+if __name__ == "__main__":
+    print("model test")
+    model("This is the input text.")

pipline.py

@@ -0,0 +1,35 @@
+import langchain as lc
+from langchain import PromptTemplate, OpenAI, LLMChain
+from langchain.prompts import load_prompt
+import wikipedia
+import os
+
+llm = OpenAI(openai_api_key=os.environ.get("OPENAI_API_KEY"))
+# save templates to a file
+try_imply_template = """Question:
+ The user wrote me the following text, what is he trying to imply to me?
+ {user_input}
+
+Answer: Let's think step by step."""
+# An example prompt with multiple input variables
+TI_prompt = PromptTemplate(
+    input_variables=["user_input"],
+    template=try_imply_template,
+)
+
+
+connection_between_terms_template = PromptTemplate(
+    template="""Question:
+ What is the connection between {term1} and {term2}?
+ Answer: Let's think step by step.""",
+    input_variables=["term1", "term2"],
+)
+
+
+chain_TI = LLMChain(prompt=TI_prompt, llm=llm)
+chain_CC = LLMChain(prompt=connection_between_terms_template, llm=llm)
+
+
+if __name__ == "__main__":
+    print(chain_TI.run("I am happy"))
+    print(chain_CC.run(["I am sad", "I am happy"]))

requirements.txt

@@ -1,6 +1,6 @@
 streamlit==1.20.*
 requests==2.27.*
-sentencepiece==0.1.*
-torch==1.11.*
-transformers==4.*
-uvicorn[standard]==0.17.*
+streamlit-ace
+langchain
+openai
+wikipedia

t5.py

@@ -1,8 +0,0 @@
-from transformers import pipeline
-
-pipe_flan = pipeline(model="bigscience/bloom-560m")
-
-
-def t5(input):
-    output = pipe_flan(input)
-    return output[0]["generated_text"]

test.py

@@ -0,0 +1,4 @@
+import subprocess
+import model
+
+print(model.model("Hello, world!, please generate some text for me."))

tests.sh

@@ -0,0 +1,3 @@
+echo "running tests"
+# python pipline.py
+# python model.py