Initial commit

.gitignore

@@ -0,0 +1 @@
+venv/

README.md

@@ -1,8 +1,8 @@
 ---
-title: Spanish Gpt2
+title: Spanish GPT-2
 emoji: 🔥
 colorFrom: yellow
-colorTo: yellow
+colorTo: red
 sdk: streamlit
 app_file: app.py
 pinned: false

app.py

@@ -0,0 +1,168 @@
+import json
+import random
+import requests
+
+from mtranslate import translate
+import streamlit as st
+
+from prompts import PROMPT_LIST
+
+
+headers = {}
+
+LOGO = "https://raw.githubusercontent.com/nlp-en-es/assets/main/logo.png"
+
+MODELS = {
+    "Model trained on OSCAR": {
+        "url": "https://api-inference.huggingface.co/models/flax-community/gpt-2-spanish"
+    },
+    "Model trained on the Large Spanish Corpus": {
+        "url": "https://api-inference.huggingface.co/models/mrm8488/spanish-gpt2"
+    },
+}
+
+
+def query(payload, model_name):
+    data = json.dumps(payload)
+    print("model url:", MODELS[model_name]["url"])
+    response = requests.request(
+        "POST", MODELS[model_name]["url"], headers=headers, data=data
+    )
+    return json.loads(response.content.decode("utf-8"))
+
+
+def process(
+    text: str, model_name: str, max_len: int, temp: float, top_k: int, top_p: float
+):
+    payload = {
+        "inputs": text,
+        "parameters": {
+            "max_new_tokens": max_len,
+            "top_k": top_k,
+            "top_p": top_p,
+            "temperature": temp,
+            "repetition_penalty": 2.0,
+        },
+        "options": {
+            "use_cache": True,
+        },
+    }
+    return query(payload, model_name)
+
+
+st.set_page_config(page_title="Spanish GPT-2 Demo", page_icon=LOGO)
+st.title("Spanish GPT-2")
+
+# Sidebar
+st.sidebar.image(LOGO)
+st.sidebar.subheader("Configurable parameters")
+
+max_len = st.sidebar.number_input(
+    "Maximum length",
+    value=100,
+    help="The maximum length of the sequence to be generated.",
+)
+
+temp = st.sidebar.slider(
+    "Temperature",
+    value=1.0,
+    min_value=0.1,
+    max_value=100.0,
+    help="The value used to module the next token probabilities.",
+)
+
+top_k = st.sidebar.number_input(
+    "Top k",
+    value=10,
+    help="The number of highest probability vocabulary tokens to keep for top-k-filtering.",
+)
+
+top_p = st.sidebar.number_input(
+    "Top p",
+    value=0.95,
+    help=" If set to float < 1, only the most probable tokens with probabilities that add up to top_p or higher are kept for generation.",
+)
+
+do_sample = st.sidebar.selectbox(
+    "Sampling?",
+    (True, False),
+    help="Whether or not to use sampling; use greedy decoding otherwise.",
+)
+
+st.markdown(
+    """
+    Spanish GPT-2 models trained from scratch on two different datasets. One
+    model is trained on the Spanish portion of
+    [OSCAR](https://huggingface.co/datasets/viewer/?dataset=oscar)
+    and the other on the
+    [large_spanish_corpus](https://huggingface.co/datasets/viewer/?dataset=large_spanish_corpus)
+    aka BETO's corpus.
+    
+    The models are trained with Flax and using TPUs sponsored by Google since this is part of the
+    [Flax/Jax Community Week](https://discuss.huggingface.co/t/open-to-the-community-community-week-using-jax-flax-for-nlp-cv/7104)
+    organised by HuggingFace.
+    """
+)
+
+model_name = st.selectbox(
+    "Model",
+    (["Model trained on OSCAR", "Model trained on the Large Spanish Corpus"]),
+)
+
+ALL_PROMPTS = list(PROMPT_LIST.keys()) + ["Custom"]
+prompt = st.selectbox("Prompt", ALL_PROMPTS, index=len(ALL_PROMPTS) - 1)
+if prompt == "Custom":
+    prompt_box = "Enter your text here"
+else:
+    prompt_box = random.choice(PROMPT_LIST[prompt])
+
+text = st.text_area("Enter text", prompt_box)
+
+if st.button("Run"):
+    with st.spinner(text="Getting results..."):
+        st.subheader("Result")
+        print(f"maxlen:{max_len}, temp:{temp}, top_k:{top_k}, top_p:{top_p}")
+        result = process(
+            text=text,
+            model_name=model_name,
+            max_len=int(max_len),
+            temp=temp,
+            top_k=int(top_k),
+            top_p=float(top_p),
+        )
+        print("result:", result)
+        if "error" in result:
+            if type(result["error"]) is str:
+                st.write(f'{result["error"]}.', end=" ")
+                if "estimated_time" in result:
+                    st.write(
+                        f'Please try again in about {result["estimated_time"]:.0f} seconds.'
+                    )
+            else:
+                if type(result["error"]) is list:
+                    for error in result["error"]:
+                        st.write(f"{error}")
+        else:
+            result = result[0]["generated_text"]
+            st.write(result.replace("\n", "  \n"))
+            st.text("English translation")
+            st.write(translate(result, "en", "es").replace("\n", "  \n"))
+
+st.markdown(
+    """
+    ### Team members
+    - Manuel Romero ([mrm8488](https://huggingface.co/mrm8488))
+    - María Grandury ([mariagrandury](https://huggingface.co/mariagrandury))
+    - Pablo González de Prado ([Pablogps](https://huggingface.co/Pablogps))
+    - Daniel Vera ([daveni](https://huggingface.co/daveni))
+    - Sri Lakshmi ([srisweet](https://huggingface.co/srisweet))
+    - José Posada ([jdposa](https://huggingface.co/jdposa))
+    - Santiago Hincapie ([shpotes](https://huggingface.co/shpotes))
+    - Jorge ([jorgealro](https://huggingface.co/jorgealro))
+    
+    ### More information
+    You can find more information about these models in their cards:
+    - [Model trained on OSCAR](https://huggingface.co/models/flax-community/gpt-2-spanish)
+    - [Model trained on the Large Spanish Corpus](https://huggingface.co/mrm8488/spanish-gpt2)
+    """
+)

prompts.py

@@ -0,0 +1,5 @@
+PROMPT_LIST = {
+    "Érase una vez...": ["Érase una vez "],
+    "¡Hola!": ["¡Hola! Me llamo "],
+    "¿Ser o no ser?": ["En mi opinión, 'ser' es "],
+}

requirements.txt

@@ -0,0 +1,4 @@
+streamlit
+mtranslate
+requests==2.24.0
+requests-toolbelt==0.9.1