llm-bg/tucan-9b-v1.0-gguf overview

{
  "metadata": {},
  "card_data": {
    "license": "gemma",
    "language": [
      "bg"
    ],
    "base_model": [
      "s-emanuilov/Tucan-9B-v1.0"
    ],
    "tags": [
      "function_calling",
      "MCP",
      "tool_use"
    ],
    "frontmatter": {
      "license": "gemma",
      "language": [
        "bg"
      ],
      "base_model": [
        "s-emanuilov/Tucan-9B-v1.0"
      ],
      "tags": [
        "function_calling",
        "MCP",
        "tool_use"
      ]
    },
    "hero_image_url": "",
    "summary": "",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "---\nlicense: gemma\nlanguage:\n- bg\nbase_model:\n- s-emanuilov/Tucan-9B-v1.0\ntags:\n- function_calling\n- MCP\n- tool_use\n---\n\n# Tucan-9B-v1.0-GGUF\n\n## Bulgarian Language Models for Function Calling 🇧🇬\n\n**Paper: https://arxiv.org/abs/2506.23394**\n\n## Overview 🚀\n\nTUCAN (Tool-Using Capable Assistant Navigator) is a series of open-source Bulgarian language models fine-tuned specifically for function calling and tool use. \n\nThese models can interact with external tools, APIs, and databases, making them appropriate for building AI agents and [Model Context Protocol (MCP)](https://arxiv.org/abs/2503.23278) applications.\n\nBuilt on top of [BgGPT models](https://huggingface.co/collections/INSAIT-Institute/bggpt-gemma-2-673b972fe9902749ac90f6fe) from [INSAIT Institute](https://insait.ai/), these models have been enhanced with function-calling capabilities.\n\n## Motivation 🎯\n\nAlthough BgGPT models demonstrate [strong Bulgarian language comprehension](https://arxiv.org/pdf/2412.10893), they face challenges in maintaining the precise formatting necessary for consistent function calling. Despite implementing detailed system prompts, their performance in this specific task remains suboptimal.\n\nThis project addresses that gap by fine-tuning BgGPT, providing the Bulgarian AI community with proper tool-use capabilities in their native language.\n\n## Models and variants 📦\nAvailable in three sizes with full models, LoRA adapters, and quantized GGUF variants:\n\n| Model Size | Full Model | LoRA Adapter | GGUF (Quantized) |\n|------------|------------|--------------|------------------|\n| **2.6B** | [Tucan-2.6B-v1.0](https://huggingface.co/llm-bg/Tucan-2.6B-v1.0) | [LoRA](https://huggingface.co/llm-bg/Tucan-2.6B-v1.0-LoRA) | [GGUF](https://huggingface.co/llm-bg/Tucan-2.6B-v1.0-GGUF) |\n| **9B** | [Tucan-9B-v1.0](https://huggingface.co/llm-bg/Tucan-9B-v1.0) | [LoRA](https://huggingface.co/llm-bg/Tucan-9B-v1.0-LoRA) | [GGUF](https://huggingface.co/llm-bg/Tucan-9B-v1.0-GGUF) 📍|\n| **27B** | [Tucan-27B-v1.0](https://huggingface.co/llm-bg/Tucan-27B-v1.0) | [LoRA](https://huggingface.co/llm-bg/Tucan-27B-v1.0-LoRA) | [GGUF](https://huggingface.co/llm-bg/Tucan-27B-v1.0-GGUF) |\n\n*GGUF variants include: q4_k_m, q5_k_m, q6_k, q8_0, q4_0 quantizations*  \n\n## Usage 🛠️\n\n### Quick start ⚡\n```bash\npip install -U \"transformers[torch]\" accelerate bitsandbytes\n```\n\n### Prompt format ⚙️\n**Critical:** Use this format for function calling for the best results.\n\n<details>\n<summary><strong>📋 Required System Prompt Template</strong></summary>\n\n```\n<bos><start_of_turn>user\nТи си полезен AI асистент, който предоставя полезни и точни отговори.\n\nИмаш достъп и можеш да извикаш една или повече функции, за да помогнеш с потребителското запитване. Използвай ги, само ако е необходимо и подходящо.\n\nКогато използваш функция, форматирай извикването ѝ в блок ```tool_call``` на отделен ред, a след това ще получиш резултат от изпълнението в блок ```toll_response```.\n\n## Шаблон за извикване: \n```tool_call\n{\"name\": <function-name>, \"arguments\": <args-json-object>}```\n\n## Налични функции:\n[your function definitions here]\n\n## Потребителска заявка : \n[your query in Bulgarian]<end_of_turn>\n<start_of_turn>model\n```\n\n</details>\n\n### Note 📝\n**The model only generates the `tool_call` blocks with function names and parameters - it doesn't actually execute the functions.** Your client application must parse these generated calls, execute the actual functions (API calls, database queries, etc.), and provide the results back to the model in `tool_response` blocks for the conversation to continue the interperation of the results. A full demo is comming soon.\n\n### Python example 🐍\n\n<details>\n<summary><strong>💻 Complete Working Example</strong></summary>\n\n```python\nimport torch\nimport json\nfrom transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig\n\n# Load model\nmodel_name = \"s-emanuilov/Tucan-2.6B-v1.0\"\ntokenizer = AutoTokenizer.from_pretrained(model_name)\nmodel = AutoModelForCausalLM.from_pretrained(\n    model_name,\n    torch_dtype=torch.bfloat16,\n    device_map=\"auto\",\n    attn_implementation=\"eager\"  # Required for Gemma models\n)\n\n# Create prompt with system template\ndef create_prompt(functions, user_query):\n    system_prompt = \"\"\"Ти си полезен AI асистент, който предоставя полезни и точни отговори.\n\nИмаш достъп и можеш да извикаш една или повече функции, за да помогнеш с потребителското запитване. Използвай ги, само ако е необходимо и подходящо.\n\nКогато използваш функция, форматирай извикването ѝ в блок ```tool_call``` на отделен ред, a след това ще получиш резултат от изпълнението в блок ```toll_response```.\n\n## Шаблон за извикване: \n```tool_call\n{{\"name\": <function-name>, \"arguments\": <args-json-object>}}```\n\"\"\"\n    \n    functions_text = json.dumps(functions, ensure_ascii=False, indent=2)\n    full_prompt = f\"{system_prompt}\\n## Налични функции:\\n{functions_text}\\n\\n## Потребителска заявка:\\n{user_query}\"\n    \n    chat = [{\"role\": \"user\", \"content\": full_prompt}]\n    return tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)\n\n# Example usage\nfunctions = [{\n    \"name\": \"create_calendar_event\",\n    \"description\": \"Creates a new event in Google Calendar.\",\n    \"parameters\": {\n        \"type\": \"object\",\n        \"properties\": {\n            \"title\": {\"type\": \"string\"},\n            \"date\": {\"type\": \"string\"},\n            \"start_time\": {\"type\": \"string\"},\n            \"end_time\": {\"type\": \"string\"}\n        },\n        \"required\": [\"title\", \"date\", \"start_time\", \"end_time\"]\n    }\n}]\n\nquery = \"Създай събитие 'Годишен преглед' за 8-ми юни 2025 от 14:00 до 14:30.\"\n\n# Generate response\nprompt = create_prompt(functions, query)\ninputs = tokenizer(prompt, return_tensors=\"pt\").to(model.device)\n\noutputs = model.generate(\n    **inputs,\n    max_new_tokens=1024,\n    temperature=0.1,\n    top_k=25,\n    top_p=1.0,\n    repetition_penalty=1.1,\n    do_sample=True,\n    eos_token_id=[tokenizer.eos_token_id, tokenizer.convert_tokens_to_ids(\"<end_of_turn>\")],\n    pad_token_id=tokenizer.eos_token_id\n)\n\nresult = tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True)\nprint(result)\n```\n\n</details>\n\n## Performance & Dataset 📊\n\n> 📄 **Full methodology, dataset details, and comprehensive evaluation results coming in the upcoming paper**\n\n**Dataset:** 8,000+ bilingual (Bulgarian/English) function-calling examples across 1,000+ topics, including tool calls with single/multiple arguments, optional parameters, follow-up queries, multi-tool selection, ambiguous queries requiring clarification, and conversational interactions without tool use. Data sourced from manual curation and synthetic generation (Gemini Pro 2.5/GPT-4.1/Sonnet 4).\n\n**Results:** ~40% improvement in tool-use capabilities over base BgGPT models in internal benchmarks.\n\n## Questions & Contact 💬\nFor questions, collaboration, or feedback: **[Connect on LinkedIn](https://www.linkedin.com/in/simeon-emanuilov/)**\n\n## Acknowledgments 🙏\nBuilt on top of [BgGPT series](https://huggingface.co/collections/INSAIT-Institute/bggpt-gemma-2-673b972fe9902749ac90f6fe).\n\n## License 📄\nThis work is licensed under [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/).",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "function_calling",
    "MCP",
    "tool_use",
    "bg",
    "arxiv:2506.23394",
    "arxiv:2503.23278",
    "arxiv:2412.10893",
    "base_model:llm-bg/Tucan-9B-v1.0",
    "base_model:quantized:llm-bg/Tucan-9B-v1.0",
    "license:gemma",
    "endpoints_compatible",
    "region:us"
  ],
  "likes": 1,
  "downloads": 88,
  "gated": false,
  "private": false,
  "last_modified": "2025-07-01T15:30:17.000Z",
  "created_at": "2025-06-08T07:39:02.000Z",
  "pipeline_tag": "",
  "library_name": ""
}

{
  "_id": "68453e1694189465c043dce3",
  "id": "llm-bg/Tucan-9B-v1.0-GGUF",
  "modelId": "llm-bg/Tucan-9B-v1.0-GGUF",
  "sha": "722f63d6c83648ee681ff0e077dfa2b2cb6629fd",
  "createdAt": "2025-06-08T07:39:02.000Z",
  "lastModified": "2025-07-01T15:30:17.000Z",
  "author": "llm-bg",
  "downloads": 88,
  "likes": 1,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "",
  "siblings_count": 10
}