update all

README.md

@@ -7,4 +7,4 @@ sdk: docker
 app_port: 7860
 pinned: false
 ---
-
+key = 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

app.py

@@ -1,19 +1,31 @@
 import subprocess
 import torch
 import streamlit as st
+from streamlit import session_state as state
 import streamlit_ace
+import model
+import pipline
 
-if torch.cuda.is_available():
-    for i in range(torch.cuda.device_count()):
-        st.text(torch.cuda.get_device_properties(i).name)
-else:
-    st.text("Could not get device properties")
-input_user = ""
-with st.form("input code"):
-    code: str = streamlit_ace.st_ace(auto_update=True)
-    if st.form_submit_button("run code"):
-        prossess = subprocess.run(
-            code.rstrip().split(), shell=True, capture_output=True
-        )
-        st.text(prossess.stdout.decode("utf-8"))
-        st.error(prossess.stderr.decode("utf-8"))
+if "app" not in state:
+    state.app = "model"
+    # state.input_text = "This is the input text."
+    # state.word = "input"
+    state.out = ""
+
+st.title("Streamlit using Huggingface Transformers and langchain")
+
+def __run_pipline():
+    st.text(f"input_text: {state.input_text}\nword: {state.word}")
+    st.markdown(":green[Running pipline]")
+    st.text(pipline.pipeline(state.input_text, state.word))
+
+def __run_model():
+    st.text(f"input_text: {state.input_text}")
+    st.markdown(":green[Running model]")
+    st.text(model.model(state.input_text))
+
+
+st.text_area("input_text", key="input_text")
+st.text_input("word", key="word")
+st.button("pip line", on_click=__run_pipline)
+st.button("model", on_click=__run_model)

awesome_prompt.json

@@ -0,0 +1,12 @@
+{
+    "input_variables": [
+        "adjective",
+        "content"
+    ],
+    "output_parser": null,
+    "partial_variables": {},
+    "template": "Tell me a {adjective} joke about {content}.",
+    "template_format": "f-string",
+    "validate_template": true,
+    "_type": "prompt"
+}

model.py

@@ -1,55 +1,24 @@
-import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
-import warnings
+import torch
 
 
+torch_dtype = torch.bfloat16
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+model_name = "bigscience/bloomz-1b7"
 
 
 
-torch_dtype = torch.bfloat16
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-model_name = "mosaicml/mpt-7b"
-
-
-model = AutoModelForCausalLM.from_pretrained(
-    model_name,
-    torch_dtype=torch_dtype,
-    trust_remote_code=True,
-    use_auth_token=None,
-)
-tokenizer = AutoTokenizer.from_pretrained(
-    model_name,
-    trust_remote_code=True,
-    use_auth_token=None,
-)
-model.eval()
-model.to(device=device, dtype=torch_dtype)
-if tokenizer.pad_token_id is None:
-    warnings.warn(
-        "pad_token_id is not set for the tokenizer. Using eos_token_id as pad_token_id."
-    )
-    tokenizer.pad_token = tokenizer.eos_token
-    tokenizer.padding_side = "left"
-
-gkw = {
-    "temperature": 0.5,
-    "top_p": 0.92,
-    "top_k": 0,
-    "max_new_tokens": 512,
-    "use_cache": True,
-    "do_sample": True,
-    "eos_token_id": tokenizer.eos_token_id,
-    "pad_token_id": tokenizer.pad_token_id,
-    "repetition_penalty": 1.1,  # 1.0 means no penalty, > 1.0 means penalty, 1.2 from CTRL paper
-}
-
-
-def mpt_7b(s):
-    input_ids = tokenizer(s, return_tensors="pt").input_ids
-    input_ids = input_ids.to(model.device)
-    with torch.no_grad():
-        output_ids = model.generate(input_ids, **gkw)
-        # Slice the output_ids tensor to get only new tokens
-        new_tokens = output_ids[0, len(input_ids[0]) :]
-        output_text = tokenizer.decode(new_tokens, skip_special_tokens=True)
-    return output_text
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto", device_map="auto").to(device=device)
+
+
+def model(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 +1,36 @@
+import langchain as lc
+from langchain import PromptTemplate
+from langchain.prompts import load_prompt
+import wikipedia
+from model import model
+
+# save templates to a file
+
+
+# An example prompt with multiple input variables
+multiple_input_prompt = PromptTemplate(
+    input_variables=["adjective", "content"],
+    template="Tell me a {adjective} joke about {content}.",
+)
+multiple_input_prompt.save("awesome_prompt.json")  # Save to JSON file
+# multiple_input_prompt.format(adjective="funny", content="chickens")
+# -> "Tell me a funny joke about chickens."
+
+
+prompt = load_prompt("awesome_prompt.json")
+
+
+def pipeline(text, word):
+    model_output = ""
+    input_text = prompt.format(adjective="funny", content=text)
+    while word not in model_output:
+        model_output = model(input_text)
+        wikipedia_entry = wikipedia.search(word)[1]
+        wiki = wikipedia.summary(wikipedia_entry, auto_suggest=False, redirect=True)
+        input_text += model_output + wiki
+    return model_output
+
+
+if __name__ == "__main__":
+    print("pipline test")
+    pipeline("This is the input text.", "input")

requirements.txt

@@ -3,7 +3,7 @@ requests==2.27.*
 sentencepiece==0.1.*
 torch==2.*
 transformers==4.*
-uvicorn[standard]==0.17.*
 streamlit-ace
-einops
-langchain
+langchain
+openai
+accelerate

test.py

@@ -1,4 +1,4 @@
 import subprocess
 import model
 
-print(model.mpt_7b("Hello, world!, please generate some text for me."))
+print(model.model("Hello, world!, please generate some text for me."))

tests.sh

@@ -1,2 +1,3 @@
 echo "running tests"
-python test.py
+# python pipline.py
+# python model.py