Add application file

Dockerfile

@@ -0,0 +1,21 @@
+FROM python:3.10
+
+WORKDIR /code
+
+COPY ./requirements.txt /code/requirements.txt
+
+RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
+# RUN apt-get update && apt-get install -y libbitsandbytes-dev
+
+RUN useradd -m -u 1000 user
+
+USER user
+
+ENV HOME=/home/user \
+	PATH=/home/user/.local/bin:$PATH
+
+WORKDIR $HOME/app
+
+COPY --chown=user . $HOME/app
+
+CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "7860"]

README.md

@@ -1,11 +1,75 @@
 ---
-title: Braingpt Implement
-emoji: 🔥
-colorFrom: green
-colorTo: pink
+title: Brainbench
+emoji: 🦀
+colorFrom: gray
+colorTo: purple
 sdk: docker
 pinned: false
-short_description: This space build of brainbench. Implementing BrainGPT
+short_description: Implementing braingpt using BrainGPT-7B-v0.1
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+
+This app currently uses **BrainGPT** as the model engine for generating and evaluating abstracts.
+
+---
+
+## 🚀 Features
+
+- 🧪 Presents users with neuroscience abstracts (either original or altered).
+- ✅ Users decide if an abstract is AI-modified and rate their confidence.
+- 🤖 BrainGPT model evaluates the same abstract.
+- 📊 Results compare user guesses vs model output in a clear, styled interface.
+- 📂 Backed by a curated dataset hosted on Hugging Face Datasets.
+- 🛠️ Fully Dockerized FastAPI application deployed via Hugging Face Spaces.
+
+---
+
+## 📦 Tech Stack
+
+- **Frontend**: Jinja2 + Tailwind-style CSS
+- **Backend**: FastAPI
+- **Model**: `BrainGPT-7B-v0.1` via Hugging Face Transformers
+- **Hosting**: Hugging Face Spaces (Docker SDK)
+- **Dataset**: Custom neuroscience benchmark (Parquet format)
+
+---
+
+## 🔧 Endpoints
+
+- `/`: Home page
+- `/start`: Begin a new session (random abstract trials)
+- `/trial`: Active abstract assessment
+- `/submit-trial`: Submit a response, compare with model
+- `/results`: View summary of performance
+- `/predict`: POST endpoint used internally to get BrainGPT-7B-v0.1 output
+
+---
+
+## 🧠 Example Use Case
+
+1. A neuroscience researcher lands on the app.
+2. They read an abstract and guess whether it’s been altered.
+3. They rate their confidence.
+4. BrainGPT-7B-v0.1 also analyzes the abstract.
+5. After 3 rounds, the app displays a comparison: who got what right?
+
+---
+
+## 📌 Current Model
+
+- Model: [`BrainGPT-7B-v0.1`](https://huggingface.co/BrainGPT/BrainGPT-7B-v0.1)
+- Reason: Fast, lightweight, deployable in CPU-only Spaces
+
+---
+
+## 🗺️ Roadmap
+
+- [ ] Add leaderboard tracking
+- [ ] Expand dataset with more domains
+
+---
+
+## 🙌 Credits
+
+Built by **Dr. David Andai** — a general practitioner & data scientist passionate about mental health and AI.

app/main.py

@@ -0,0 +1,41 @@
+
+from fastapi import FastAPI, Request, Form
+from fastapi.responses import JSONResponse
+from app.model_loader import load_model
+import torch
+
+app = FastAPI()
+model, tokenizer = load_model()
+
+# @app.post("/predict")
+# async def predict(request: Request):
+#     data = await request.json()
+#     input_text = data.get("input", "")
+#     inputs = tokenizer(input_text, return_tensors="pt")
+#     with torch.no_grad():
+#         output = model.generate(
+#             **inputs,
+#             max_new_tokens=60,
+#             do_sample=False,
+#             temperature=0.3
+#         )
+#     response = tokenizer.decode(output[0], skip_special_tokens=True)
+#     return JSONResponse(content={"output": response})
+
+@app.post("/predict")
+async def predict(request: Request):
+    data = await request.json()
+    input_text = data.get("input", "")
+
+    inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
+
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=60,
+            do_sample=False,
+            temperature=0.3
+        )
+
+    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return JSONResponse(content={"output": response})

app/model_loader.py

@@ -0,0 +1,62 @@
+
+import os
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel
+
+def load_model():
+    hf_token = os.getenv("HF_TOKEN")
+    if not hf_token:
+        raise RuntimeError("HF_TOKEN not set.")
+    
+    # Use a user-writable cache directory (important for Docker non-root)
+    HF_CACHE = os.path.expanduser("~/.cache/huggingface")
+    os.makedirs(HF_CACHE, exist_ok=True)
+
+    os.environ["TRANSFORMERS_CACHE"] = HF_CACHE
+    os.environ["HF_HOME"] = HF_CACHE
+
+    base_model = AutoModelForCausalLM.from_pretrained(
+        "meta-llama/Llama-2-7b-chat-hf",
+        use_auth_token=hf_token,
+        cache_dir="/tmp/hf_cache",
+        torch_dtype="auto",
+        device_map="auto"
+    )
+    model = PeftModel.from_pretrained(
+        base_model,
+        "BrainGPT/BrainGPT-7B-v0.1",
+        use_auth_token=hf_token,
+        cache_dir="/tmp/hf_cache"
+    )
+    tokenizer = AutoTokenizer.from_pretrained(
+        "meta-llama/Llama-2-7b-chat-hf",
+        use_auth_token=hf_token,
+        cache_dir="/tmp/hf_cache"
+    )
+    return model, tokenizer
+
+## GPT 2 Model
+# import os
+# from transformers import AutoModelForCausalLM, AutoTokenizer
+
+# def load_model():
+#     # Use a user-writable cache directory (important for Docker non-root)
+#     HF_CACHE = os.path.expanduser("~/.cache/huggingface")
+#     os.makedirs(HF_CACHE, exist_ok=True)
+
+#     os.environ["TRANSFORMERS_CACHE"] = HF_CACHE
+#     os.environ["HF_HOME"] = HF_CACHE
+
+#     model_name = "gpt2"
+
+#     tokenizer = AutoTokenizer.from_pretrained(
+#         model_name,
+#         cache_dir=HF_CACHE
+#     )
+
+#     model = AutoModelForCausalLM.from_pretrained(
+#         model_name,
+#         cache_dir=HF_CACHE
+#     )
+
+#     return model, tokenizer

requirements.txt

@@ -0,0 +1,8 @@
+
+transformers
+peft
+torch
+accelerate
+fastapi
+uvicorn
+bitsandbytes