Add visual model card to README.md

.github/actions/model-card-generator/action.yml

@@ -0,0 +1,25 @@
+
+name: Model Card Generator
+description: Auto-generate MODEL_CARD.md for TinyLlama checkpoints.
+
+inputs:
+  model_path:
+    description: Path to model dir
+    required: true
+    default: checkpoints/TinyLlama-1.1B
+  output:
+    description: Output markdown file
+    required: false
+    default: MODEL_CARD.md
+
+runs:
+  using: composite
+  steps:
+    - uses: actions/setup-python@v5
+      with:
+        python-version: '3.11'
+    - name: Install deps
+      run: pip install --quiet transformers
+    - shell: bash
+      run: |
+        python "$GITHUB_ACTION_PATH/model_card_generator.py" --model_path "${{ inputs.model_path }}" --output "${{ inputs.output }}"

.github/actions/model-card-generator/model_card_generator.py

@@ -0,0 +1,38 @@
+
+#!/usr/bin/env python3
+"""TinyLlama Model Card Generator."""
+import argparse, datetime
+from pathlib import Path
+try:
+    from transformers import AutoConfig
+except ImportError:
+    AutoConfig = None
+
+TEMPLATE = """# {name}
+**Generated on {date}**
+
+- **Architecture**: {arch}
+- **Parameters**: {params}
+- **License**: {license}
+"""
+
+def generate(model_path, output):
+    meta = dict(name=Path(model_path).name, date=datetime.date.today(), arch='Unknown', params='N/A', license='N/A')
+    if AutoConfig and Path(model_path).exists():
+        try:
+            cfg = AutoConfig.from_pretrained(model_path)
+            meta.update(arch=(cfg.architectures[0] if getattr(cfg,'architectures',[]) else 'Unknown'),
+                        params=getattr(cfg,'num_parameters','N/A'),
+                        license=getattr(cfg,'license','N/A'),
+                        name=cfg.name_or_path)
+        except Exception as e:
+            print('Warn:', e)
+    Path(output).write_text(TEMPLATE.format(**meta))
+    print('Wrote', output)
+
+if __name__ == '__main__':
+    p=argparse.ArgumentParser()
+    p.add_argument('--model_path', required=True)
+    p.add_argument('--output', default='MODEL_CARD.md')
+    args=p.parse_args()
+    generate(args.model_path, args.output)

.github/actions/model-card-generator/s

@@ -0,0 +1 @@
+ss

.github/workflows/generate-model-card.yml

@@ -0,0 +1,20 @@
+
+name: Generate TinyLlama Model Card
+on:
+  push:
+    paths:
+      - 'checkpoints/**'
+  workflow_dispatch:
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: ./.github/actions/model-card-generator
+        with:
+          model_path: 'checkpoints/TinyLlama-1.1B'
+          output: 'MODEL_CARD.md'
+      - uses: stefanzweifel/git-auto-commit-action@v5
+        with:
+          commit_message: 'docs: update MODEL_CARD.md'
+          file_pattern: 'MODEL_CARD.md'

.github/workflows/t

@@ -0,0 +1 @@
+

README.md

@@ -9,7 +9,7 @@ English | [中文](README_zh-CN.md)
 The TinyLlama project aims to **pretrain** a **1.1B Llama model on 3 trillion tokens**. With some proper optimization, we can achieve this within a span of "just" 90 days using 16 A100-40G GPUs 🚀🚀. The training has started on 2023-09-01. 
 
 <div align="center">
-  <img src=".github/TinyLlama_logo.png" width="300"/>
+  <img src=".github/tinyllama-modelcard.png" width="300"/>
 </div>
 
 We adopted exactly the same architecture and tokenizer as Llama 2. This means TinyLlama can be plugged and played in many open-source projects built upon Llama. Besides, TinyLlama is compact with only 1.1B parameters. This compactness allows it to cater to a multitude of applications demanding a restricted computation and memory footprint.