Sync all files - fix TrainingArguments

.gitignore

@@ -0,0 +1,40 @@
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+.Python
+env/
+venv/
+.venv/
+
+# Training outputs
+d1337-cipher-output/
+*.bin
+*.safetensors
+*.pt
+*.pth
+
+# Logs
+*.log
+wandb/
+runs/
+
+# Cache
+.cache/
+*.cache
+
+# IDE
+.idea/
+.vscode/
+*.swp
+*.swo
+
+# OS
+.DS_Store
+Thumbs.db
+
+# Secrets
+.env
+*.env
+secrets.yaml

Dockerfile

@@ -0,0 +1,49 @@
+# D1337 CIPHER - Custom Training Environment
+# Optimized for 4x L40S (192GB VRAM) with QLoRA
+
+FROM nvidia/cuda:12.1.0-cudnn8-devel-ubuntu22.04
+
+# Set environment
+ENV DEBIAN_FRONTEND=noninteractive
+ENV PYTHONUNBUFFERED=1
+ENV PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:512
+ENV TOKENIZERS_PARALLELISM=false
+ENV HF_HOME=/app/.cache/huggingface
+
+# Install system dependencies
+RUN apt-get update && apt-get install -y \
+    python3.10 \
+    python3-pip \
+    python3.10-venv \
+    git \
+    git-lfs \
+    curl \
+    wget \
+    && rm -rf /var/lib/apt/lists/* \
+    && git lfs install
+
+# Create user for HuggingFace Spaces
+RUN useradd -m -u 1000 user
+USER user
+ENV PATH="/home/user/.local/bin:$PATH"
+
+WORKDIR /app
+
+# Copy requirements first (for caching)
+COPY --chown=user requirements.txt .
+
+# Install Python dependencies
+RUN pip3 install --no-cache-dir --upgrade pip && \
+    pip3 install --no-cache-dir -r requirements.txt
+
+# Copy application files
+COPY --chown=user . .
+
+# Expose port for Gradio
+EXPOSE 7860
+
+# Set HuggingFace cache directory
+RUN mkdir -p /app/.cache/huggingface
+
+# Run training script
+CMD ["python3", "train.py"]

README.md

@@ -0,0 +1,61 @@
+---
+title: D1337 CIPHER Training
+emoji: 🔥
+colorFrom: red
+colorTo: purple
+sdk: docker
+pinned: true
+license: mit
+app_port: 7860
+---
+
+# D1337 CIPHER - Custom Training Environment
+
+**D1337 SOVEREIGN LABS**
+
+Custom QLoRA training environment for fine-tuning GLM-4.7-Flash-abliterated (31B) on cybersecurity datasets.
+
+## Features
+
+- 🔥 **QLoRA Training** - Memory efficient training for 31B models
+- 🎯 **4x L40S Optimized** - Configured for 192GB VRAM
+- 📊 **Gradio UI** - Real-time monitoring and control
+- 🚀 **Auto Push to Hub** - Automatically saves to HuggingFace
+
+## Configuration
+
+| Parameter | Value |
+|-----------|-------|
+| Base Model | `huihui-ai/Huihui-GLM-4.7-Flash-abliterated` |
+| Dataset | `Desorden1337/d1337-cipher-dataset` |
+| LoRA Rank | 64 |
+| LoRA Alpha | 128 |
+| Epochs | 5 |
+| Learning Rate | 2e-4 |
+| Max Seq Length | 4096 |
+
+## Training Topics
+
+- SentinelOne EDR
+- CrowdStrike Falcon
+- Palo Alto Networks
+- Zero-day Research
+- AI/ML Security
+- Adversarial Attacks
+- Cloud Security
+
+## Usage
+
+1. Open this Space
+2. Click "Start Training"
+3. Monitor progress in logs
+4. Model will be saved to `Desorden1337/d1337-cipher-v1`
+
+## Hardware Requirements
+
+- **Minimum**: 4x L40S (192GB VRAM)
+- **Recommended**: 8x L40S or 4x A100
+
+---
+
+*D1337 SOVEREIGN LABS - Building the future of AI security*

config.yaml

@@ -0,0 +1,53 @@
+# D1337 CIPHER Training Configuration
+# =====================================
+# Edit this file to customize training
+
+model:
+  base_model: "huihui-ai/Huihui-GLM-4.7-Flash-abliterated"
+  output_model: "Desorden1337/d1337-cipher-v1"
+  trust_remote_code: true
+
+dataset:
+  name: "Desorden1337/d1337-cipher-dataset"
+  split: "train"
+  text_field: "messages"
+
+lora:
+  r: 64
+  alpha: 128
+  dropout: 0.05
+  target_modules:
+    - "q_proj"
+    - "k_proj"
+    - "v_proj"
+    - "o_proj"
+    - "gate_proj"
+    - "up_proj"
+    - "down_proj"
+
+training:
+  epochs: 5
+  batch_size: 1
+  gradient_accumulation: 8
+  learning_rate: 0.0002
+  weight_decay: 0.01
+  warmup_ratio: 0.1
+  max_seq_length: 4096
+  scheduler: "cosine"
+
+quantization:
+  use_4bit: true
+  quant_type: "nf4"
+  compute_dtype: "bfloat16"
+  double_quant: true
+
+hardware:
+  bf16: true
+  gradient_checkpointing: true
+  dataloader_workers: 4
+
+hub:
+  push_to_hub: true
+  private_repo: true
+  save_steps: 100
+  save_total_limit: 2

requirements.txt

@@ -0,0 +1,31 @@
+# D1337 CIPHER Training Dependencies
+# Optimized for QLoRA on 4x L40S
+
+# Core ML
+torch>=2.1.0
+git+https://github.com/huggingface/transformers.git
+datasets>=2.15.0
+accelerate>=0.27.0
+
+# PEFT/LoRA
+peft>=0.10.0
+bitsandbytes>=0.43.0
+
+# HuggingFace
+huggingface-hub>=0.20.0
+safetensors>=0.4.0
+
+# Training utilities
+trl>=0.8.0
+einops>=0.7.0
+scipy>=1.11.0
+
+# Monitoring
+wandb>=0.16.0
+tqdm>=4.66.0
+
+# Gradio for UI
+gradio>=4.0.0
+
+# Config
+pyyaml>=6.0