index.html

<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<title>Corelyn MLMConstruct</title>
<style>
@import url('https://fonts.googleapis.com/css2?family=IBM+Plex+Mono:wght@400;500&family=IBM+Plex+Sans:wght@300;400;500&display=swap');
*{box-sizing:border-box;margin:0;padding:0}
#app{background:#fff;min-height:600px;font-family:'IBM Plex Sans',sans-serif;color:#111;max-width:1100px;margin:auto}
.header{padding:2rem 2rem 1.5rem;border-bottom:1px solid #111;display:flex;align-items:center;justify-content:space-between}
.header span{font-family:'IBM Plex Mono',monospace;font-size:10px;color:#999;letter-spacing:1px}
.container{padding:2rem}
.row{display:grid;grid-template-columns:1fr 1fr;gap:1.5rem}
.field{display:flex;flex-direction:column;gap:6px;margin-bottom:1.25rem}
.field label{font-family:'IBM Plex Mono',monospace;font-size:10px;color:#999;letter-spacing:1.5px;text-transform:uppercase}
.field input,.field textarea,.field select{background:#fff;border:none;border-bottom:1px solid #ddd;font-family:'IBM Plex Sans',sans-serif;font-size:14px;padding:8px 0;outline:none}
.field textarea{resize:vertical;min-height:80px}
.btn-generate{width:100%;padding:14px;background:#111;color:#fff;font-family:'IBM Plex Mono',monospace;font-size:11px;letter-spacing:2px;text-transform:uppercase;border:none;cursor:pointer;margin-top:1rem}
.btn-generate:hover{background:#333}
.divider{height:1px;background:#eee;margin:2rem 0}
.status{display:none;padding:1rem 0}
.status.visible{display:block}
.status-line{font-family:'IBM Plex Mono',monospace;font-size:11px;color:#bbb;padding:3px 0}
.status-line.active{color:#111}.status-line.done{color:#999}.status-line.err{color:#c00}
.output{display:none;margin-top:2rem}.output.visible{display:block}
.result-card{border:1px solid #eee;background:#fafafa;padding:1.5rem;margin-bottom:2rem}
.meta-row{display:flex;gap:2rem;flex-wrap:wrap;font-family:'IBM Plex Mono',monospace;font-size:11px}
.meta-row span{color:#000;font-weight:500}
.dataset-list{margin-top:1rem}
.sample-item{padding:10px 0;border-bottom:1px solid #eee}
.sample-text{font-size:13px;line-height:1.6}
.code-wrap{background:#f8f8f8;border:1px solid #eee;padding:1.2rem;overflow:auto;max-height:400px}
.code-wrap pre{font-family:'IBM Plex Mono',monospace;font-size:11px}
.result-actions{display:flex;gap:10px;margin-top:1rem}
.btn-download,.btn-copy{border:1px solid #ddd;background:#fff;padding:6px 14px;font-family:'IBM Plex Mono',monospace;font-size:10px;cursor:pointer}
.btn-download:hover,.btn-copy:hover{border-color:#111;color:#111}
</style>
</head>
<body>

<div id="app">

  <div class="header">
    <img src="images/logo.png" width="150">
    <span>dataset forge + script generator</span>
  </div>

  <div class="container">

    <div class="row">
      <div class="field">
        <label>API Key</label>
        <input type="password" id="api-key" placeholder="Enter API key">
      </div>
      <div class="field">
        <label>Model</label>
        <select id="model-select">
          <optgroup label="NVIDIA">
            <option value="nvidia/moonshotai/kimi-k2.5">KimiK2</option>
            <option value="nvidia/qwen/qwen3.5-397b-a17b">Qwen3.5</option>
            <option value="nvidia/microsoft/phi-3.5-mini-instruct">Phi-3.5 Mini</option>
            <option value="nvidia/meta/llama-3.2-1b-instruct">LLaMA-3.2-1B</option>
            <option value="nvidia/deepseek-ai/deepseek-v3.1">DeepSeek-3.1</option>
            <option value="nvidia/z-ai/glm5">GLM5</option>
            <option value="nvidia/mistralai/mistral-7b-instruct-v0.2">Mistral-7B</option>
            <option value="nvidia/google/gemma-7b">Gemma-7B</option>
            <option value="nvidia/tiiuae/falcon3-7b-instruct">Falcon3-7B</option>
          </optgroup>
          <optgroup label="Cerebras">
            <option value="cerebras/llama3.1-8b">LLaMA-3.1-8B</option>
            <option value="cerebras/gpt-oss-120b">GPT-OSS-120B</option>
          </optgroup>
        </select>
      </div>
    </div>

    <div class="row">
      <div class="field">
        <label>Model Name</label>
        <input id="model-name" placeholder="MyLLM-130M">
      </div>
      <div class="field">
        <label>Output Directory</label>
        <input id="out-dir" placeholder="./my_model">
      </div>
    </div>

    <div class="field">
      <label>Dataset Prompt</label>
      <textarea id="dataset-prompt" placeholder="Describe what dataset you want"></textarea>
    </div>

    <div class="row">
      <div class="field">
        <label>Sample Count</label>
        <select id="sample-count">
          <option>5</option>
          <option selected>10</option>
          <option>20</option>
          <option>50</option>
          <option>100</option>
          <option>200</option>
          <option>500</option>
        </select>
      </div>
      <div class="field">
        <label>Format</label>
        <select id="output-format">
          <option value="text">Plain text</option>
          <option value="qa">Q&A pairs</option>
          <option value="instruction">Instruction/Response</option>
        </select>
      </div>
    </div>

    <button class="btn-generate" id="btn-gen" onclick="generate()">Generate Dataset</button>

    <div class="status" id="status">
      <div class="status-line" id="s1">initializing...</div>
      <div class="status-line" id="s2">generating dataset...</div>
      <div class="status-line" id="s3">building training script...</div>
      <div class="status-line" id="s4">complete</div>
    </div>

    <div class="output" id="output">
      <div class="result-card">
        <div class="meta-row" id="meta-row"></div>
        <div class="result-actions">
          <button class="btn-download" onclick="downloadPy()">Download train.py</button>
          <button class="btn-download" onclick="downloadDataset()">Download Dataset</button>
          <button class="btn-copy" onclick="copyCode()">Copy Code</button>
        </div>
      </div>

      <div class="divider"></div>
      <div id="dataset-list" class="dataset-list"></div>
      <div class="divider"></div>
      <div class="code-wrap"><pre id="code-out"></pre></div>
    </div>

  </div>
</div>

<script>
let generatedCode=''
let dataset=[]

document.getElementById("api-key").value =
  localStorage.getItem("dataset_api_key") || ""

document.getElementById("api-key").addEventListener("change", e => {
  localStorage.setItem("dataset_api_key", e.target.value)
})

function setS(n,state){
  document.getElementById('s'+n).className='status-line '+state
}

async function generate(){
  const modelName=document.getElementById('model-name').value.trim()
  const prompt=document.getElementById('dataset-prompt').value.trim()
  const count=document.getElementById('sample-count').value
  const format=document.getElementById('output-format').value
  const apiKey=document.getElementById('api-key').value
  const model=document.getElementById('model-select').value

  if(!modelName||!prompt||!apiKey){
    alert("Fill required fields")
    return
  }

  document.getElementById('status').classList.add("visible")
  setS(1,'done'); setS(2,'active');

  const sys=`Generate ${count} dataset samples about ${prompt}. Return JSON array of strings only.`

  const res=await fetch(
    "https://api.corelyn.ro/chat/completions",
    {
      method:"POST",
      headers:{"Content-Type":"application/json"},
      body:JSON.stringify({apiKey,model,prompt:sys})
    }
  )

  const data=await res.json()
  const raw=data.choices?.[0]?.message?.content||data.text||""

  try{
    const clean = raw.replace(/```json|```/g,"").trim()
    const parsed = JSON.parse(clean)
    dataset = Array.isArray(parsed) ? parsed : [clean]
  } catch{
    dataset = raw.split("\n").map(x=>x.trim()).filter(x=>x.length>10)
  }

  setS(2,'done'); setS(3,'active');

  generatedCode = buildTrainPy(modelName, dataset)

  setS(3,'done'); setS(4,'done');

  renderResults(modelName,count,format)
}

function renderResults(modelName,count,format){
  document.getElementById("meta-row").innerHTML=`
    <div>model <span>${modelName}</span></div>
    <div>samples <span>${count}</span></div>
    <div>format <span>${format}</span></div>
    <div>
      First run <span>pip install -q transformers datasets accelerate sentencepiece torch</span>.<br>
      If you have a GPU and want CUDA support, use <span>pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121</span> instead.
    </div>

  `

  document.getElementById("dataset-list").innerHTML=
    dataset.slice(0,20).map(x=>`
      <div class="sample-item"><div class="sample-text">${x}</div></div>
    `).join("")

  document.getElementById("code-out").textContent=generatedCode
  document.getElementById("output").classList.add("visible")
}

function downloadPy(){
  const blob=new Blob([generatedCode],{type:"text/x-python"})
  const a=document.createElement("a")
  a.href=URL.createObjectURL(blob)
  a.download="train.py"
  a.click()
}

function downloadDataset(){
  const blob=new Blob([JSON.stringify(dataset,null,2)],{type:"application/json"})
  const a=document.createElement("a")
  a.href=URL.createObjectURL(blob)
  a.download="dataset.json"
  a.click()
}

function copyCode(){navigator.clipboard.writeText(generatedCode)}

function buildTrainPy(modelName,dataset){
  const pyDataset = dataset.map((item,i)=>{try{return JSON.parse(item)}catch{return {id:i+1,text:item}}})
  return `# ============================================
# Auto Generated LLaMA Training Script
# Model: ${modelName}
# ============================================

import torch
from datasets import Dataset, load_dataset
from transformers import (
    AutoTokenizer,
    LlamaConfig,
    LlamaForCausalLM,
    TrainingArguments,
    Trainer,
    DataCollatorForLanguageModeling
)

device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using device:", device)

tokenizer = AutoTokenizer.from_pretrained("TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T")
tokenizer.pad_token = tokenizer.eos_token

data = ${JSON.stringify(pyDataset,null,2)}
dataset = Dataset.from_list(data)
print("Loaded samples:", len(dataset))

def tokenize(example):
    text = example.get("text") or example.get("model_response") or ""
    tokens = tokenizer(text,truncation=True,padding="max_length",max_length=512)
    tokens["labels"] = tokens["input_ids"].copy()
    return tokens

tokenized_dataset = dataset.map(tokenize,batched=False)

config = LlamaConfig(
    vocab_size=tokenizer.vocab_size,
    hidden_size=768,
    intermediate_size=3072,
    num_hidden_layers=16,
    num_attention_heads=12,
    max_position_embeddings=1024,
    rms_norm_eps=1e-6,
    initializer_range=0.02
)

model = LlamaForCausalLM(config)
model.gradient_checkpointing_enable()
model = model.to(device)

total_params = sum(p.numel() for p in model.parameters())
print("Total parameters:", total_params)

data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False)

training_args = TrainingArguments(
    output_dir="./${modelName}",
    per_device_train_batch_size=2,
    gradient_accumulation_steps=8,
    learning_rate=3e-4,
    weight_decay=0.01,
    warmup_steps=1000,
    num_train_epochs=3,
    logging_steps=50,
    save_steps=1000,
    fp16=torch.cuda.is_available(),
    report_to="none"
)

trainer = Trainer(model=model,args=training_args,train_dataset=tokenized_dataset,data_collator=data_collator)

trainer.train()
trainer.save_model("./${modelName}")
tokenizer.save_pretrained("./${modelName}")

print("Training complete. Model saved to ./${modelName}")
`
}
</script>

</body>
</html>