File size: 10,170 Bytes
99e14ba
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
391265d
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>KoarAI LFM2.5 Training Studio</title>
    <link href="https://cdn.jsdelivr.net/npm/tailwindcss@2.2.19/dist/tailwind.min.css" rel="stylesheet">
    <style>
        body { background-color: #0f172a; color: #f8fafc; font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, sans-serif; }
        .card { background-color: #1e293b; border: 1px solid #334155; border-radius: 12px; }
        .slider-accent { accent-color: #6366f1; }
        pre { background-color: #090d16; border: 1px solid #1e293b; }
    </style>
</head>
<body class="p-6 md:p-10">
    <div class="max-w-6xl mx-auto">
        <header class="mb-8 border-b border-gray-700 pb-6 flex items-center justify-between">
            <div>
                <h1 class="text-3xl font-extrabold text-indigo-400 flex items-center gap-2">
                    🧠 KoarAI LFM2.5-350M Training Studio
                </h1>
                <p class="text-gray-400 mt-2">
                    Interactive Dataset Mixer, Reasoning (CoT) Balancer & Colab Pipeline Generator for <b>LiquidAI/LFM2.5-350M</b>.
                </p>
            </div>
            <a href="https://huggingface.co/KoarAI/LFM2.5-350M-Thinking-0003" target="_blank" class="bg-indigo-600 hover:bg-indigo-500 text-white px-4 py-2 rounded-lg font-medium shadow-md transition">
                View Model on HF πŸš€
            </a>
        </header>

        <div class="grid grid-cols-1 lg:grid-cols-2 gap-8">
            <!-- Left Column: Controls -->
            <div class="card p-6 shadow-xl">
                <h2 class="text-xl font-bold mb-4 text-indigo-300">πŸŽ›οΈ 1. Dataset Mix Configuration</h2>
                
                <div class="space-y-4">
                    <div>
                        <div class="flex justify-between text-sm mb-1">
                            <span class="font-medium text-gray-300">πŸ‡·πŸ‡Ί Russian Deep CoT (ZeroAgency/ru-thinking-reasoning-r1)</span>
                            <span id="val_ru_cot" class="font-bold text-indigo-400">700</span>
                        </div>
                        <input type="range" id="ru_cot" min="0" max="2000" step="50" value="700" class="w-full slider-accent" oninput="update()">
                    </div>

                    <div>
                        <div class="flex justify-between text-sm mb-1">
                            <span class="font-medium text-gray-300">πŸ’¬ Russian Natural Chat (Vikhrmodels/GrandMaster-PRO-MAX)</span>
                            <span id="val_ru_chat" class="font-bold text-indigo-400">700</span>
                        </div>
                        <input type="range" id="ru_chat" min="0" max="2000" step="50" value="700" class="w-full slider-accent" oninput="update()">
                    </div>

                    <div>
                        <div class="flex justify-between text-sm mb-1">
                            <span class="font-medium text-gray-300">🌐 English General Chat (LDJnr/Capybara)</span>
                            <span id="val_en_chat" class="font-bold text-indigo-400">800</span>
                        </div>
                        <input type="range" id="en_chat" min="0" max="2000" step="50" value="800" class="w-full slider-accent" oninput="update()">
                    </div>

                    <div>
                        <div class="flex justify-between text-sm mb-1">
                            <span class="font-medium text-gray-300">πŸ“ Math & Logic CoT (r0b0tlab/qwen3.8-max-distillation-50k)</span>
                            <span id="val_math_cot" class="font-bold text-indigo-400">600</span>
                        </div>
                        <input type="range" id="math_cot" min="0" max="2000" step="50" value="600" class="w-full slider-accent" oninput="update()">
                    </div>

                    <div>
                        <div class="flex justify-between text-sm mb-1">
                            <span class="font-medium text-gray-300">πŸ’» Clean Python & Code (iamtarun/python_code_instructions)</span>
                            <span id="val_code" class="font-bold text-indigo-400">350</span>
                        </div>
                        <input type="range" id="code" min="0" max="2000" step="50" value="350" class="w-full slider-accent" oninput="update()">
                    </div>
                </div>

                <h2 class="text-xl font-bold mt-8 mb-4 text-indigo-300">βš™οΈ 2. Hyperparameters</h2>
                <div class="grid grid-cols-2 gap-4">
                    <div>
                        <label class="block text-xs text-gray-400 mb-1">Learning Rate</label>
                        <input type="text" id="lr" value="2.5e-5" class="w-full bg-gray-900 border border-gray-700 rounded p-2 text-sm text-white" oninput="update()">
                    </div>
                    <div>
                        <label class="block text-xs text-gray-400 mb-1">Epochs</label>
                        <input type="number" id="epochs" value="2.3" step="0.1" class="w-full bg-gray-900 border border-gray-700 rounded p-2 text-sm text-white" oninput="update()">
                    </div>
                </div>
            </div>

            <!-- Right Column: Output Preview & Script -->
            <div class="space-y-6">
                <!-- Summary Card -->
                <div class="card p-6 shadow-xl border-l-4 border-indigo-500">
                    <h3 class="text-lg font-bold text-white mb-3">πŸ“Š Balance Summary</h3>
                    <div class="grid grid-cols-2 gap-2 text-sm text-gray-300 mb-3" id="summary_stats">
                        <!-- Populated by JS -->
                    </div>
                    <div class="w-full bg-gray-700 h-3 rounded-full overflow-hidden flex" id="balance_bar">
                        <!-- Populated by JS -->
                    </div>
                </div>

                <!-- Code Container -->
                <div class="card p-6 shadow-xl">
                    <div class="flex justify-between items-center mb-3">
                        <h3 class="text-lg font-bold text-indigo-300">πŸ“œ Generated PyTorch / Colab Script</h3>
                        <button onclick="copyCode()" class="bg-gray-800 hover:bg-gray-700 text-xs px-3 py-1.5 rounded border border-gray-600 transition">
                            πŸ“‹ Copy Code
                        </button>
                    </div>
                    <pre class="p-4 rounded-lg text-xs overflow-x-auto text-green-400 max-h-96" id="code_box"></pre>
                </div>
            </div>
        </div>
    </div>

    <script>
        function update() {
            const ru_cot = parseInt(document.getElementById('ru_cot').value);
            const ru_chat = parseInt(document.getElementById('ru_chat').value);
            const en_chat = parseInt(document.getElementById('en_chat').value);
            const math_cot = parseInt(document.getElementById('math_cot').value);
            const code = parseInt(document.getElementById('code').value);
            const lr = document.getElementById('lr').value;
            const epochs = parseFloat(document.getElementById('epochs').value);

            document.getElementById('val_ru_cot').innerText = ru_cot;
            document.getElementById('val_ru_chat').innerText = ru_chat;
            document.getElementById('val_en_chat').innerText = en_chat;
            document.getElementById('val_math_cot').innerText = math_cot;
            document.getElementById('val_code').innerText = code;

            const total = ru_cot + ru_chat + en_chat + math_cot + code;
            
            document.getElementById('summary_stats').innerHTML = `
                <div>Total Samples: <b class="text-white">${total}</b></div>
                <div>Est. T4 Time: <b class="text-white">~${Math.max(5, Math.round(total * epochs / 150))} min</b></div>
                <div>Russian CoT: <b class="text-indigo-400">${Math.round(ru_cot/total*100)}%</b></div>
                <div>Natural Chat: <b class="text-green-400">${Math.round((ru_chat+en_chat)/total*100)}%</b></div>
            `;

            const code_template = `# ==============================================================================
# Auto-Generated Training Script for LiquidAI/LFM2.5-350M
# Total Samples: ${total} | Epochs: ${epochs} | Learning Rate: ${lr}
# ==============================================================================
!pip install -q --upgrade "transformers>=4.48.0" "datasets>=3.0.0" "accelerate>=1.0.0"

import os, re, json, torch
from datasets import load_dataset, Dataset
from transformers import AutoModelForCausalLM, AutoTokenizer, Trainer, TrainingArguments, DataCollatorForSeq2Seq
from huggingface_hub import login

HF_TOKEN = "YOUR_HF_TOKEN"
BASE_MODEL = "LiquidAI/LFM2.5-350M"
TARGET_REPO = "KoarAI/LFM2.5-350M-Thinking-0003"
MAX_LENGTH = 512

login(token=HF_TOKEN)

tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL, token=HF_TOKEN, trust_remote_code=True)
tokenizer.add_special_tokens({"additional_special_tokens": ["<think>", "</think>"]})

all_samples = []

# Dataset Mix Loader (${total} total samples)
# 1. Russian Deep CoT: ${ru_cot}
# 2. Russian Natural Dialogues: ${ru_chat}
# 3. English General Chat: ${en_chat}
# 4. Math & Logic CoT: ${math_cot}
# 5. Clean Code: ${code}

training_args = TrainingArguments(
    output_dir="./lfm_output",
    num_train_epochs=${epochs},
    learning_rate=${lr},
    lr_scheduler_type="cosine",
    warmup_steps=40,
    per_device_train_batch_size=2,
    gradient_accumulation_steps=4,
    weight_decay=0.01,
    logging_steps=10,
    bf16=torch.cuda.is_available()
)
`;
            document.getElementById('code_box').innerText = code_template;
        }

        function copyCode() {
            const code = document.getElementById('code_box').innerText;
            navigator.clipboard.writeText(code);
            alert("Training script copied to clipboard!");
        }

        update();
    </script>
</body>
</html>