Upload 4 files

Dockerfile

@@ -0,0 +1,10 @@
+FROM python:3.10-slim
+
+WORKDIR /app
+
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]

khmer_spell_lstm.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bdb4948f7bf7b078fd6195db4d8745aa7f65d96eac7edd164da06b55d803a22d
+size 10207041

main.py

@@ -0,0 +1,95 @@
+import torch
+import torch.nn as nn
+import re
+from fastapi import FastAPI
+from pydantic import BaseModel
+
+# =====================================================
+# App
+# =====================================================
+app = FastAPI(title="Khmer Spell Correction API")
+
+# =====================================================
+# Utils
+# =====================================================
+def preprocess_khmer_text(text: str) -> str:
+    text = re.sub(r"\s+", " ", text)
+    text = re.sub(r"[^\u1780-\u17FF\u200B\u0020-\u007E]", "", text)
+    return text.strip()
+
+# =====================================================
+# Model
+# =====================================================
+class KhmerSpellLSTM(nn.Module):
+    def __init__(self, vocab_size, embedding_dim, hidden_dim):
+        super().__init__()
+        self.embedding = nn.Embedding(vocab_size, embedding_dim, padding_idx=0)
+        self.lstm = nn.LSTM(
+            embedding_dim,
+            hidden_dim,
+            batch_first=True,
+            bidirectional=True
+        )
+        self.fc = nn.Linear(hidden_dim * 2, vocab_size)
+
+    def forward(self, x):
+        x = self.embedding(x)
+        x, _ = self.lstm(x)
+        return self.fc(x)
+
+# =====================================================
+# Load Model ONCE
+# =====================================================
+device = torch.device("cpu")
+
+checkpoint = torch.load("khmer_spell_lstm.pth", map_location=device)
+
+char_to_idx = checkpoint["char_to_idx"]
+idx_to_char = {i: c for c, i in char_to_idx.items()}
+max_length = checkpoint["max_length"]
+
+model = KhmerSpellLSTM(
+    vocab_size=len(char_to_idx),
+    embedding_dim=128,
+    hidden_dim=256
+)
+
+model.load_state_dict(checkpoint["model_state_dict"])
+model.eval()
+
+# =====================================================
+# Inference
+# =====================================================
+def predict(text: str) -> str:
+    text = preprocess_khmer_text(text)
+    input_len = len(text)
+
+    seq = [char_to_idx.get(c, char_to_idx["<UNK>"]) for c in text]
+    seq += [char_to_idx["<PAD>"]] * (max_length - len(seq))
+    seq = torch.tensor(seq[:max_length]).unsqueeze(0)
+
+    with torch.no_grad():
+        out = model(seq)
+        pred = torch.argmax(out, dim=-1)[0][:input_len]
+
+    return "".join(idx_to_char[i.item()] for i in pred)
+
+# =====================================================
+# Schema
+# =====================================================
+class TextInput(BaseModel):
+    text: str
+
+# =====================================================
+# Routes
+# =====================================================
+@app.get("/")
+def health():
+    return {"status": "Khmer Spell API running"}
+
+@app.post("/predict")
+def spell_correct(data: TextInput):
+    return {
+        "input": data.text,
+        "output": predict(data.text)
+    }

requirements.txt

@@ -0,0 +1,4 @@
+fastapi
+uvicorn
+torch
+pydantic