Upload 2 files

README.md

@@ -0,0 +1,126 @@
+---
+language: multilingual
+license: apache-2.0
+author: Dark
+library_name: transformers
+
+tags:
+  - darkit-v2.5
+  - open-source
+  - text-generation
+  - programming
+  - reasoning
+  - fine-tuning
+  - customizable
+
+base_model: darkps/darkit-v2.5
+model_type: custom
+pipeline_tag: text-generation
+---
+
+# DarkIT v2.5
+
+DarkIT is a next-generation high-performance large language model engineered for **advanced programming, deep reasoning, and natural human conversation**.
+
+DarkIT v2.5 introduces major improvements in:
+
+* Advanced code generation
+* Complex debugging & error analysis
+* Long-context reasoning
+* Multi-language programming support
+* Instruction following for difficult technical tasks
+* Architecture understanding & code refactoring
+* Stable conversational behavior
+* Fast and efficient local inference
+* Unrestricted responses with strong adaptability
+
+---
+
+# What's New in v2.5
+
+DarkIT v2.5 has been significantly upgraded with a massive programming-focused training phase.
+
+### Major Improvements
+
+* Trained on over **18 million high-quality programming conversations**
+* Strongly improved coding intelligence and reasoning
+* Better understanding of software architecture and system design
+* More accurate debugging and bug fixing
+* Improved instruction consistency
+* Better long-response stability
+* Reduced hallucinations in programming tasks
+* Faster response generation quality under long prompts
+
+### Programming Capabilities
+
+DarkIT v2.5 performs strongly across:
+
+* Python
+* C++
+* JavaScript / TypeScript
+* Java
+* Rust
+* Go
+* PHP
+* SQL
+* Bash / Shell scripting
+* HTML / CSS
+* AI & Machine Learning workflows
+
+---
+
+# Key Specifications
+
+* **Model Family:** DarkIT Coder
+* **Version:** v2.5
+* **Model Size:** 15B Parameters
+* **Context Length:** 256k Tokens
+* **Format:** GGUF (optimized for efficient local deployment)
+* **Inference Support:** CPU / GPU
+* **Primary Focus:** Programming & Technical Reasoning
+
+---
+
+# Performance Notes
+
+* Optimized for strong local inference performance
+* Excellent balance between speed and output quality
+* Stable long-context generation
+* Enhanced code completion consistency
+* Improved logical reasoning across technical tasks
+* Designed for developer workflows and advanced prompting
+
+---
+
+# Recommended Usage
+
+DarkIT v2.5 performs best when used for:
+
+* Software development
+* AI engineering tasks
+* Code generation
+* Debugging large projects
+* Technical explanations
+* Automation scripting
+* Long-context programming conversations
+* Local offline AI deployment
+
+---
+
+# ⚠️ Notes
+
+* Designed primarily for inference deployment
+* Output quality may vary depending on quantization level and hardware
+* Best performance is achieved using structured prompts
+* Large context usage may require substantial RAM/VRAM
+
+---
+
+# About Dark
+
+Dark is an independent developer focused on building efficient, powerful, and scalable language models for real-world applications, with a strong focus on programming intelligence and local AI deployment.
+
+---
+
+* **Website:** https://dark.ps
+* **Telegram:** https://t.me/sii_3

notebook.ipynb

@@ -0,0 +1,93 @@
+{
+  "cells": [
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "!pip install llama-cpp-python huggingface_hub --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu124\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "from huggingface_hub import HfApi\n",
+        "from llama_cpp import Llama\n",
+        "import os\n",
+        "\n",
+        "REPO_ID = \"darkps/darkit-v2.5\"\n",
+        "api = HfApi()\n",
+        "\n",
+        "files = api.list_repo_files(REPO_ID)\n",
+        "gguf_files = [f for f in files if f.endswith(\".gguf\")]\n",
+        "\n",
+        "for i, f in enumerate(gguf_files):\n",
+        "    print(f\"[{i}] {f}\")\n",
+        "\n",
+        "choice = int(input(\"Select model number: \"))\n",
+        "filename = gguf_files[choice]\n",
+        "\n",
+        "llm = Llama.from_pretrained(\n",
+        "    repo_id=REPO_ID,\n",
+        "    filename=filename,\n",
+        "    n_ctx=2048,\n",
+        "    n_batch=128,\n",
+        "    n_ubatch=128,\n",
+        "    n_threads=os.cpu_count() or 4,\n",
+        "    n_threads_batch=os.cpu_count() or 4,\n",
+        "    n_gpu_layers=-1,\n",
+        "    verbose=False,\n",
+        "    no_perf=True,\n",
+        ")\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "llm.set_cache(None)\n",
+        "\n",
+        "PROMPT = \"Hi how are you?\"\n",
+        "\n",
+        "stream = llm(\n",
+        "    f\"<|im_start|>user\\n{PROMPT}<|im_end|>\\n<|im_start|>assistant\\n\",\n",
+        "    max_tokens=128,\n",
+        "    temperature=0.7,\n",
+        "    top_p=0.8,\n",
+        "    top_k=20,\n",
+        "    stream=True,\n",
+        "    stop=[\n",
+        "        \"<|im_end|>\",\n",
+        "        \"<|im_start|>\",\n",
+        "        \"\\n\\nUser:\",\n",
+        "        \"\\n\\nAssistant:\"\n",
+        "    ],\n",
+        "    echo=False\n",
+        ")\n",
+        "\n",
+        "for chunk in stream:\n",
+        "    text = chunk[\"choices\"][0][\"text\"]\n",
+        "\n",
+        "    if text:\n",
+        "        print(text, end=\"\", flush=True)\n",
+        "\n",
+        "print()\n"
+      ]
+    }
+  ],
+  "metadata": {
+    "kernelspec": {
+      "display_name": "Python 3",
+      "language": "python",
+      "name": "python3"
+    }
+  },
+  "nbformat": 4,
+  "nbformat_minor": 0
+}