upload_to_hf.py

# ============================================================
#  Upload Emotion Classifier to HuggingFace Hub
# ============================================================
# Run this AFTER kaggle_train.py finishes.
# You can run it in the same Kaggle notebook or locally.
#
# Prerequisites:
#   pip install huggingface_hub transformers
#
# Steps:
#   1. Create a free account at https://huggingface.co
#   2. Go to Settings → Access Tokens → New token (write permission)
#   3. Replace YOUR_HF_TOKEN and YOUR_HF_USERNAME below
#   4. Run this script
# ============================================================

import os
from pathlib import Path
from transformers import DistilBertForSequenceClassification, DistilBertTokenizerFast
from huggingface_hub import HfApi, login, create_repo

# ── Config ───────────────────────────────────────────────────────────────────
HF_TOKEN    = "hf_xxxxxxxxxxxxxxxxxxxx"          # ← paste your HF write token
HF_USERNAME = "your-hf-username"                  # ← your HF username
REPO_NAME   = "emotion-classifier-distilbert"     # ← repo name (will be created)
MODEL_DIR   = "/kaggle/working/emotion_model/final"   # path from training script

FULL_REPO_ID = f"{HF_USERNAME}/{REPO_NAME}"

# ── Login ─────────────────────────────────────────────────────────────────────
login(token=HF_TOKEN)
api = HfApi()

# ── Create repo (public) ──────────────────────────────────────────────────────
print(f"Creating repo: {FULL_REPO_ID}")
create_repo(FULL_REPO_ID, repo_type="model", exist_ok=True, private=False)

# ── Load model & tokenizer from local dir ────────────────────────────────────
model     = DistilBertForSequenceClassification.from_pretrained(MODEL_DIR)
tokenizer = DistilBertTokenizerFast.from_pretrained(MODEL_DIR)

# ── Push model & tokenizer ────────────────────────────────────────────────────
print("Pushing model to HuggingFace Hub …")
model.push_to_hub(FULL_REPO_ID, token=HF_TOKEN)
tokenizer.push_to_hub(FULL_REPO_ID, token=HF_TOKEN)

# ── Write a model card (README.md) ────────────────────────────────────────────
MODEL_CARD = f"""---
language: en
tags:
  - text-classification
  - emotion-detection
  - distilbert
license: apache-2.0
datasets:
  - custom-emotion-dataset
metrics:
  - accuracy
  - f1
pipeline_tag: text-classification
---

# Emotion Classifier — DistilBERT

A fine-tuned [DistilBERT](https://huggingface.co/distilbert-base-uncased) model
for **6-class emotion detection** from English text.

## Labels

| ID | Emotion  |
|----|----------|
| 0  | sadness  |
| 1  | anger    |
| 2  | love     |
| 3  | surprise |
| 4  | fear     |
| 5  | joy      |

## Training details

- Base model: `distilbert-base-uncased`
- Epochs: 10
- Batch size: 32
- Max sequence length: 128
- Optimizer: AdamW (lr=2e-5, weight_decay=0.01)
- Scheduler: Linear with warmup (10 %)
- Train / Val / Test split: 70 / 15 / 15

## Quick usage

```python
from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="{FULL_REPO_ID}",
    return_all_scores=True,
)

result = classifier("I feel so happy today!")
print(result)
```

## Citation
If you use this model, please credit the training dataset and this repo.
"""

api.upload_file(
    path_or_fileobj=MODEL_CARD.encode(),
    path_in_repo="README.md",
    repo_id=FULL_REPO_ID,
    repo_type="model",
    token=HF_TOKEN,
)

print(f"\n✅ Done! Model live at: https://huggingface.co/{FULL_REPO_ID}")