Skip to content

Commit fb15bd8

Browse files
author
walidsobhie-code
committed
feat: add ready-to-run notebooks for Kaggle and Colab
- Colab 128K context fine-tuning notebook - Kaggle 128K context fine-tuning notebook - Colab HumanEval benchmark notebook
1 parent 10182dc commit fb15bd8

3 files changed

Lines changed: 372 additions & 0 deletions

File tree

‎notebooks/colab_128k_training.ipynb‎

Lines changed: 138 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,138 @@
1+
{
2+
"cells": [
3+
{
4+
"cell_type": "markdown",
5+
"metadata": {},
6+
"source": [
7+
"# 🎯 Stack 2.9 — 128K Context Fine-tuning\n",
8+
"Fine-tune Qwen2.5-Coder-1.5B from 32K → 128K context\n",
9+
"\n",
10+
"**Runtime:** GPU (T4 16GB recommended) | **Time:** ~2-3 hours"
11+
]
12+
},
13+
{
14+
"cell_type": "markdown",
15+
"metadata": {},
16+
"source": [
17+
"## Step 1: Clone Stack 2.9 & Install Dependencies"
18+
]
19+
},
20+
{
21+
"cell_type": "code",
22+
"execution_count": null,
23+
"metadata": {},
24+
"outputs": [],
25+
"source": [
26+
"!git clone https://github.com/my-ai-stack/stack-2.9.git\n",
27+
"cd stack-2.9\n",
28+
"!pip install -q transformers peft datasets bitsandbytes accelerate huggingface_hub\n",
29+
"!pip install -q scipy torch --upgrade"
30+
]
31+
},
32+
{
33+
"cell_type": "markdown",
34+
"metadata": {},
35+
"source": [
36+
"## Step 2: Login to HuggingFace (push weights later)"
37+
]
38+
},
39+
{
40+
"cell_type": "code",
41+
"execution_count": null,
42+
"metadata": {},
43+
"outputs": [],
44+
"source": [
45+
"from huggingface_hub import login\n",
46+
"# Get your token at: https://huggingface.co/settings/tokens\n",
47+
"login(token=\"YOUR_HF_TOKEN\") # ← Replace with your token"
48+
]
49+
},
50+
{
51+
"cell_type": "markdown",
52+
"metadata": {},
53+
"source": [
54+
"## Step 3: Mount Google Drive (optional — for saving checkpoints)"
55+
]
56+
},
57+
{
58+
"cell_type": "code",
59+
"execution_count": null,
60+
"metadata": {},
61+
"outputs": [],
62+
"source": [
63+
"from google.colab import drive\n",
64+
"drive.mount('/content/drive')\n",
65+
"OUTPUT_DIR = \"/content/drive/MyDrive/stack-2.9-128k-output\""
66+
]
67+
},
68+
{
69+
"cell_type": "markdown",
70+
"metadata": {},
71+
"source": [
72+
"## Step 4: Run 128K Context Fine-tuning"
73+
]
74+
},
75+
{
76+
"cell_type": "code",
77+
"execution_count": null,
78+
"metadata": {},
79+
"outputs": [],
80+
"source": [
81+
"import subprocess\n",
82+
"result = subprocess.run([\n",
83+
" \"python3\", \"training/train_extended_context.py\",\n",
84+
" \"--model-path\", \"my-ai-stack/Stack-2-9-finetuned\",\n",
85+
" \"--data-path\", \"training/training-data/tool_examples_combined.jsonl\",\n",
86+
" \"--output-dir\", OUTPUT_DIR,\n",
87+
" \"--context-length\", \"131072\",\n",
88+
" \"--lora-rank\", \"64\",\n",
89+
" \"--epochs\", \"3\",\n",
90+
" \"--push-to-hub\",\n",
91+
" \"--hub-model-id\", \"YOUR_USERNAME/stack-2.9-128k\"\n",
92+
"], cwd=\"/content/stack-2.9\")\n",
93+
"print(result.stdout)\n",
94+
"print(result.stderr)"
95+
]
96+
},
97+
{
98+
"cell_type": "markdown",
99+
"metadata": {},
100+
"source": [
101+
"---\n",
102+
"\n",
103+
"## Alternative: Run on Base Qwen Model (if HF model not loaded)\n",
104+
"\n",
105+
"If the fine-tuned model isn't available, use the base model:"
106+
]
107+
},
108+
{
109+
"cell_type": "code",
110+
"execution_count": null,
111+
"metadata": {},
112+
"outputs": [],
113+
"source": [
114+
"# Change --model-path to:\n",
115+
"# \"Qwen/Qwen2.5-Coder-1.5B\"\n",
116+
"# And add --push-to-hub with your own model ID"
117+
]
118+
}
119+
],
120+
"metadata": {
121+
"accelerator": "GPU",
122+
"colab": {
123+
"provenance": [],
124+
"machine_shape": "hm"
125+
},
126+
"kernelspec": {
127+
"display_name": "Python 3",
128+
"language": "python",
129+
"name": "python3"
130+
},
131+
"language_info": {
132+
"name": "python",
133+
"version": "3.10.0"
134+
}
135+
},
136+
"nbformat": 4,
137+
"nbformat_minor": 4
138+
}

‎notebooks/colab_humaneval.ipynb‎

Lines changed: 104 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,104 @@
1+
{
2+
"cells": [
3+
{
4+
"cell_type": "markdown",
5+
"metadata": {},
6+
"source": [
7+
"# 📊 Stack 2.9 — HumanEval Benchmark\n",
8+
"Run full HumanEval (164 problems) pass@k evaluation\n",
9+
"\n",
10+
"**Runtime:** GPU (T4 16GB) | **Time:** ~30-60 min"
11+
]
12+
},
13+
{
14+
"cell_type": "markdown",
15+
"metadata": {},
16+
"source": [
17+
"## Step 1: Clone & Install"
18+
]
19+
},
20+
{
21+
"cell_type": "code",
22+
"execution_count": null,
23+
"metadata": {},
24+
"outputs": [],
25+
"source": [
26+
"!git clone https://github.com/my-ai-stack/stack-2.9.git\n",
27+
"cd stack-2.9\n",
28+
"!pip install -q transformers peft datasets human-eval accelerate"
29+
]
30+
},
31+
{
32+
"cell_type": "markdown",
33+
"metadata": {},
34+
"source": [
35+
"## Step 2: Run HumanEval Benchmark"
36+
]
37+
},
38+
{
39+
"cell_type": "code",
40+
"execution_count": null,
41+
"metadata": {},
42+
"outputs": [],
43+
"source": [
44+
"MODEL_PATH = \"my-ai-stack/Stack-2-9-finetuned\" # or your 128K fine-tuned model ID\n",
45+
"\n",
46+
"import subprocess\n",
47+
"result = subprocess.run([\n",
48+
" \"python3\", \"training/evaluate_model.py\",\n",
49+
" \"--model-path\", MODEL_PATH,\n",
50+
" \"--benchmark\", \"humaneval\",\n",
51+
" \"--num-samples\", \"10\",\n",
52+
" \"--max-new-tokens\", \"256\",\n",
53+
" \"--output\", \"/tmp/humaneval_results.json\"\n",
54+
"], cwd=\"/content/stack-2.9\", capture_output=True, text=True)\n",
55+
"\n",
56+
"print(result.stdout[-5000:] if result.stdout else \"No output\")\n",
57+
"print(\"STDERR:\", result.stderr[-1000:] if result.stderr else \"None\")"
58+
]
59+
},
60+
{
61+
"cell_type": "markdown",
62+
"metadata": {},
63+
"source": [
64+
"## Step 3: Parse Results"
65+
]
66+
},
67+
{
68+
"cell_type": "code",
69+
"execution_count": null,
70+
"metadata": {},
71+
"outputs": [],
72+
"source": [
73+
"import json\n",
74+
"with open(\"/tmp/humaneval_results.json\") as f:\n",
75+
" results = json.load(f)\n",
76+
"\n",
77+
"print(\"=== HumanEval Results ===\")\n",
78+
"for k, v in results.items():\n",
79+
" if isinstance(v, float):\n",
80+
" print(f\" pass@{k}: {v:.1%}\")\n",
81+
" else:\n",
82+
" print(f\" {k}: {v}\")"
83+
]
84+
}
85+
],
86+
"metadata": {
87+
"accelerator": "GPU",
88+
"colab": {
89+
"provenance": [],
90+
"machine_shape": "hm"
91+
},
92+
"kernelspec": {
93+
"display_name": "Python 3",
94+
"language": "python",
95+
"name": "python3"
96+
},
97+
"language_info": {
98+
"name": "python",
99+
"version": "3.10.0"
100+
}
101+
},
102+
"nbformat": 4,
103+
"nbformat_minor": 4
104+
}

‎notebooks/kaggle_128k_training.ipynb‎

Lines changed: 130 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,130 @@
1+
{
2+
"cells": [
3+
{
4+
"cell_type": "markdown",
5+
"metadata": {},
6+
"source": [
7+
"# 🎯 Stack 2.9 — 128K Context Fine-tuning\n",
8+
"Fine-tune Qwen2.5-Coder-1.5B from 32K → 128K context\n",
9+
"\n",
10+
"**Runtime:** GPU (P100 16GB) | **Time:** ~2-3 hours\n",
11+
"\n",
12+
"![Kaggle](https://img.shields.io/badge/Kaggle-P100-blue) ![Context](https://img.shields.io/badge/Context-128K-green)"
13+
]
14+
},
15+
{
16+
"cell_type": "markdown",
17+
"metadata": {},
18+
"source": [
19+
"## Step 1: Clone Repo & Install"
20+
]
21+
},
22+
{
23+
"cell_type": "code",
24+
"execution_count": null,
25+
"metadata": {},
26+
"outputs": [],
27+
"source": [
28+
"!git clone https://github.com/my-ai-stack/stack-2.9.git\n",
29+
"cd stack-2.9\n",
30+
"!pip install -q transformers peft datasets bitsandbytes accelerate huggingface_hub\n",
31+
"!pip install -q scipy torch --upgrade"
32+
]
33+
},
34+
{
35+
"cell_type": "markdown",
36+
"metadata": {},
37+
"source": [
38+
"## Step 2: HuggingFace Login"
39+
]
40+
},
41+
{
42+
"cell_type": "code",
43+
"execution_count": null,
44+
"metadata": {},
45+
"outputs": [],
46+
"source": [
47+
"from huggingface_hub import login\n",
48+
"import os\n",
49+
"# Add your HF token to Kaggle Secrets: https://www.kaggle.com/docs/secrets\n",
50+
"os.environ[\"HF_TOKEN\"] = \"YOUR_HF_TOKEN\"\n",
51+
"login(token=os.environ[\"HF_TOKEN\"])"
52+
]
53+
},
54+
{
55+
"cell_type": "markdown",
56+
"metadata": {},
57+
"source": [
58+
"## Step 3: Run Fine-tuning"
59+
]
60+
},
61+
{
62+
"cell_type": "code",
63+
"execution_count": null,
64+
"metadata": {},
65+
"outputs": [],
66+
"source": [
67+
"import subprocess\n",
68+
"\n",
69+
"result = subprocess.run([\n",
70+
" \"python3\", \"training/train_extended_context.py\",\n",
71+
" \"--model-path\", \"Qwen/Qwen2.5-Coder-1.5B\",\n",
72+
" \"--data-path\", \"training/training-data/tool_examples_combined.jsonl\",\n",
73+
" \"--output-dir\", \"/kaggle/working/stack-2.9-128k\",\n",
74+
" \"--context-length\", \"131072\",\n",
75+
" \"--lora-rank\", \"64\",\n",
76+
" \"--epochs\", \"3\",\n",
77+
" \"--push-to-hub\",\n",
78+
" \"--hub-model-id\", \"YOUR_USERNAME/stack-2.9-128k\"\n",
79+
"], cwd=\"/kaggle/working/stack-2.9\", env=dict(os.environ, HF_TOKEN=os.environ[\"HF_TOKEN\"]))\n",
80+
"\n",
81+
"print(result.stdout[-3000:] if result.stdout else \"No stdout\")\n",
82+
"print(result.stderr[-1000:] if result.stderr else \"No stderr\")"
83+
]
84+
},
85+
{
86+
"cell_type": "markdown",
87+
"metadata": {},
88+
"source": [
89+
"## Step 4: Download Checkpoints"
90+
]
91+
},
92+
{
93+
"cell_type": "code",
94+
"execution_count": null,
95+
"metadata": {},
96+
"outputs": [],
97+
"source": [
98+
"# After training, download the merged model\n",
99+
"!ls -la /kaggle/working/stack-2.9-128k/merged/\n",
100+
"# The merged/ folder contains the full 128K model ready to use"
101+
]
102+
}
103+
],
104+
"metadata": {
105+
"accelerator": "GPU",
106+
"kaggle": {
107+
"accelerator": "GPU",
108+
"dataSources": [],
109+
"dockerImageVersion": "gpu",
110+
"gpuRequirements": {
111+
"top": "p100"
112+
},
113+
"kernelImage": {
114+
"id": "docker",
115+
"name": "docker"
116+
}
117+
},
118+
"kernelspec": {
119+
"display_name": "Python 3",
120+
"language": "python",
121+
"name": "python3"
122+
},
123+
"language_info": {
124+
"name": "python",
125+
"version": "3.10.0"
126+
}
127+
},
128+
"nbformat": 4,
129+
"nbformat_minor": 4
130+
}

0 commit comments

Comments
 (0)