unsloth/cogito-v2-preview-deepseek-671b-moe-gguf overview

{
  "metadata": {},
  "card_data": {
    "tags": [
      "unsloth"
    ],
    "license": "mit",
    "library_name": "transformers",
    "base_model": [
      "deepcogito/cogito-v2-preview-deepseek-671B-MoE"
    ],
    "frontmatter": {
      "tags": [
        "unsloth"
      ],
      "license": "mit",
      "library_name": "transformers",
      "base_model": [
        "deepcogito/cogito-v2-preview-deepseek-671B-MoE"
      ]
    },
    "hero_image_url": "https://github.com/unslothai/unsloth/raw/main/images/unsloth%20new%20logo.png",
    "summary": "Blog Post The Cogito v2 LLMs are instruction tuned generative models. All models are released under an open license for commercial use. # Evaluations For detailed evaluations, please refer to the Blog Post. # Usage Here is a snippet below for usage with Transformers: ``python import transformers import torch model_id = \"deepcogito/cogito-v2-preview-deepseek-671B-MoE\" pipeline = transformers.pipeline( \"text-generation\", model=model_id, model_kwargs={\"torch_dtype\": torch.bfloat16}, device_map=\"auto\", ) messages = [ {\"role\": \"system\", \"content\": \"You are a pirate chatbot who always responds in pirate speak!\"}, {\"role\": \"user\", \"content\": \"Give me a short introduction to LLMs.\"}, ] outputs = pipeline( messages, max_new_tokens=512, ) print(outputs[0][\"generated_text\"][-1]) ``",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "---\ntags:\n- unsloth\nlicense: mit\nlibrary_name: transformers\nbase_model:\n- deepcogito/cogito-v2-preview-deepseek-671B-MoE\n---\n> [!NOTE]\n>  Includes Unsloth **chat template fixes**! <br> For `llama.cpp`, use `--jinja`\n>\n\n<div>\n<p style=\"margin-top: 0;margin-bottom: 0;\">\n    <em><a href=\"https://docs.unsloth.ai/basics/unsloth-dynamic-v2.0-gguf\">Unsloth Dynamic 2.0</a> achieves superior accuracy & outperforms other leading quants.</em>\n  </p>\n  <div style=\"display: flex; gap: 5px; align-items: center; \">\n    <a href=\"https://github.com/unslothai/unsloth/\">\n      <img src=\"https://github.com/unslothai/unsloth/raw/main/images/unsloth%20new%20logo.png\" width=\"133\">\n    </a>\n    <a href=\"https://discord.gg/unsloth\">\n      <img src=\"https://github.com/unslothai/unsloth/raw/main/images/Discord%20button.png\" width=\"173\">\n    </a>\n    <a href=\"https://docs.unsloth.ai/\">\n      <img src=\"https://raw.githubusercontent.com/unslothai/unsloth/refs/heads/main/images/documentation%20green%20button.png\" width=\"143\">\n    </a>\n  </div>\n</div>\n\n\n<p align=\"center\">\n  <img src=\"images/deep-cogito-logo.png\" alt=\"Logo\" width=\"40%\">\n</p>\n\n\n# Cogito v2 preview - 671B MoE\n\n[Blog Post](https://www.deepcogito.com/research/cogito-v2-preview)\n\nThe Cogito v2 LLMs are instruction tuned generative models. All models are released under an open license for commercial use.\n\n- Cogito v2 models are hybrid reasoning models. Each model can answer directly (standard LLM), or self-reflect before answering (like reasoning models).\n- The LLMs are trained using **Iterated Distillation and Amplification (IDA)** - an scalable and efficient alignment strategy for superintelligence using iterative self-improvement.\n- The models have been optimized for coding, STEM, instruction following and general helpfulness, and have significantly higher multilingual, coding and tool calling capabilities than size equivalent counterparts.\n  - In both standard and reasoning modes, Cogito v2-preview models outperform their size equivalent counterparts on common industry benchmarks. \n- This model is trained in over 30 languages and supports a context length of 128k.\n\n# Evaluations\nFor detailed evaluations, please refer to the [Blog Post](https://www.deepcogito.com/research/cogito-v2-preview).\n\n\n# Usage\nHere is a snippet below for usage with Transformers:\n\n```python\nimport transformers\nimport torch\n\nmodel_id = \"deepcogito/cogito-v2-preview-deepseek-671B-MoE\"\n\npipeline = transformers.pipeline(\n    \"text-generation\",\n    model=model_id,\n    model_kwargs={\"torch_dtype\": torch.bfloat16},\n    device_map=\"auto\",\n)\n\nmessages = [\n    {\"role\": \"system\", \"content\": \"You are a pirate chatbot who always responds in pirate speak!\"},\n    {\"role\": \"user\", \"content\": \"Give me a short introduction to LLMs.\"},\n]\n\noutputs = pipeline(\n    messages,\n    max_new_tokens=512,\n)\n\nprint(outputs[0][\"generated_text\"][-1])\n```\n\n\n\n## Implementing extended thinking\n- By default, the model will answer in the standard mode. \n- To enable thinking, you can do any one of the two methods:\n  - Set `enable_thinking=True` while applying the chat template.\n  - Add a specific system prompt, along with prefilling the response with \"\\<think\\>\\n\". \n\n**NOTE: Unlike Cogito v1 models, we initiate the response with \"\\<think\\>\\n\" at the beginning of every output when reasoning is enabled. This is because hybrid models can be brittle at times, and adding a \"\\<think\\>\\n\" ensures that the model does indeed respect thinking.**\n\n### Method 1 - Set enable_thinking=True in the tokenizer\nIf you are using Huggingface tokenizers, then you can simply use add the argument `enable_thinking=True` to the tokenization (this option is added to the chat template).\n\nHere is an example - \n```python\nfrom transformers import AutoModelForCausalLM, AutoTokenizer\n\nmodel_name = \"deepcogito/cogito-v2-preview-deepseek-671B-MoE\"\n\nmodel = AutoModelForCausalLM.from_pretrained(\n    model_name,\n    torch_dtype=\"auto\",\n    device_map=\"auto\"\n)\ntokenizer = AutoTokenizer.from_pretrained(model_name)\n\nprompt = \"Give me a short introduction to LLMs.\"\nmessages = [\n    {\"role\": \"system\", \"content\": \"You are a pirate chatbot who always responds in pirate speak!\"},\n    {\"role\": \"user\", \"content\": prompt}\n]\n\ntext = tokenizer.apply_chat_template(\n    messages,\n    tokenize=False,\n    add_generation_prompt=True,\n    enable_thinking=True\n)\nmodel_inputs = tokenizer([text], return_tensors=\"pt\").to(model.device)\n\ngenerated_ids = model.generate(\n    **model_inputs,\n    max_new_tokens=512\n)\ngenerated_ids = [\n    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)\n]\n\nresponse = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]\nprint(response)\n```\n\n### Method 2 - Add a specific system prompt, along with prefilling the response with \"\\<think\\>\\n\". \nTo enable thinking using this method, you need to do two parts - \n\n\nStep 1 - Simply use this in the system prompt `system_instruction = 'Enable deep thinking subroutine.'`\n\nIf you already have a system_instruction, then use `system_instruction = 'Enable deep thinking subroutine.' + '\\n\\n' + system_instruction`.\n\nStep 2 - Prefil the response with the tokens `\"<think>\\n\"`.\n\nHere is an example - \n\n```python\nimport transformers\nimport torch\n\nmodel_name = \"deepcogito/cogito-v2-preview-deepseek-671B-MoE\"\n\nmodel = AutoModelForCausalLM.from_pretrained(\n    model_name,\n    torch_dtype=\"auto\",\n    device_map=\"auto\"\n)\ntokenizer = AutoTokenizer.from_pretrained(model_name)\n\n# Step 1 - Add deep thinking instruction.\nDEEP_THINKING_INSTRUCTION = \"Enable deep thinking subroutine.\"\n\nmessages = [\n    {\"role\": \"system\", \"content\": DEEP_THINKING_INSTRUCTION},\n    {\"role\": \"user\", \"content\": \"Write a bash script that takes a matrix represented as a string with format '[1,2],[3,4],[5,6]' and prints the transpose in the same format.\"},\n]\n\ntext = tokenizer.apply_chat_template(\n    messages,\n    tokenize=False,\n    add_generation_prompt=True\n)\n\n# Step 2 - Prefill response with \"<think>\\n\".\ntext += \"<think>\\n\"\n\n# Now, continue as usual.\nmodel_inputs = tokenizer([text], return_tensors=\"pt\").to(model.device)\n\ngenerated_ids = model.generate(\n    **model_inputs,\n    max_new_tokens=512\n)\ngenerated_ids = [\n    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)\n]\n\nresponse = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]\nprint(response)\n```\n\n\nSimilarly, if you have a system prompt, you can append the `DEEP_THINKING_INSTRUCTION` to the beginning in this way - \n\n```python\nDEEP_THINKING_INSTRUCTION = \"Enable deep thinking subroutine.\"\n\nsystem_prompt = \"Reply to each prompt with only the actual code - no explanations.\"\nprompt = \"Write a bash script that takes a matrix represented as a string with format '[1,2],[3,4],[5,6]' and prints the transpose in the same format.\"\n\nmessages = [\n    {\"role\": \"system\", \"content\": DEEP_THINKING_INSTRUCTION + '\\n\\n' + system_prompt},\n    {\"role\": \"user\", \"content\": prompt}\n]\n```\n\n\n# Tool Calling\nCogito models support tool calling (single, parallel, multiple and parallel_multiple) both in standard and extended thinking mode.\n\nHere is a snippet -\n\n```python\n# First, define a tool\ndef get_current_temperature(location: str) -> float:\n    \"\"\"\n    Get the current temperature at a location.\n    \n    Args:\n        location: The location to get the temperature for, in the format \"City, Country\"\n    Returns:\n        The current temperature at the specified location in the specified units, as a float.\n    \"\"\"\n    return 22.  # A real function should probably actually get the temperature!\n\n# Next, create a chat and apply the chat template\nmessages = [\n  {\"role\": \"user\", \"content\": \"Hey, what's the temperature in Paris right now?\"}\n]\n\nmodel_inputs = tokenizer.apply_chat_template(messages, tools=[get_current_temperature], add_generation_prompt=True)\n\ntext = tokenizer.apply_chat_template(messages, tools=[get_current_temperature], add_generation_prompt=True, tokenize=False)\ninputs = tokenizer(text, return_tensors=\"pt\", add_special_tokens=False).to(model.device)\noutputs = model.generate(**inputs, max_new_tokens=512)\noutput_text = tokenizer.batch_decode(outputs)[0][len(text):]\nprint(output_text)\n```\n\nThis will result in the output - \n```\n<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>function<｜tool▁sep｜>get_current_temperature\n```json\n{\"location\":\"Paris, France\"}\n```<｜tool▁call▁end｜><｜tool▁calls▁end｜><｜end▁of▁sentence｜>\n```\n\nYou can then generate text from this input as normal. If the model generates a tool call, you should add it to the chat like so:\n\n```python\ntool_call = {\"name\": \"get_current_temperature\", \"arguments\": {\"location\": \"Paris, France\"}}\nmessages.append({\"role\": \"assistant\", \"tool_calls\": [{\"type\": \"function\", \"function\": tool_call}]})\n```\n\nand then call the tool and append the result, with the `tool` role, like so:\n\n```python\nmessages.append({\"role\": \"tool\", \"name\": \"get_current_temperature\", \"content\": \"22.0\"})\n```\n\nAfter that, you can `generate()` again to let the model use the tool result in the chat:\n\n```python\ntext = tokenizer.apply_chat_template(messages, tools=[get_current_temperature], add_generation_prompt=True, tokenize=False)\ninputs = tokenizer(text, return_tensors=\"pt\", add_special_tokens=False).to(model.device)\noutputs = model.generate(**inputs, max_new_tokens=512)\noutput_text = tokenizer.batch_decode(outputs)[0][len(text):]\n```\n\nThis should result in the string -\n```\n'The current temperature in Paris is 22.0 degrees.<｜end▁of▁sentence｜>'\n```\n\n## License\nThis repository and the model weights are licensed under **MIT License**.\n\n## Contact\nIf you would like to reach out to our team, send an email to [contact@deepcogito.com](contact@deepcogito.com).",
    "related_quantizations": []
  },
  "tags": [
    "transformers",
    "gguf",
    "unsloth",
    "base_model:deepcogito/cogito-v2-preview-deepseek-671B-MoE",
    "base_model:quantized:deepcogito/cogito-v2-preview-deepseek-671B-MoE",
    "license:mit",
    "endpoints_compatible",
    "region:us",
    "imatrix",
    "conversational"
  ],
  "likes": 2,
  "downloads": 92,
  "gated": false,
  "private": false,
  "last_modified": "2025-08-01T01:26:20.000Z",
  "created_at": "2025-07-29T12:15:54.000Z",
  "pipeline_tag": "",
  "library_name": "transformers"
}

{
  "_id": "6888bb7a8131503471756b07",
  "id": "unsloth/cogito-v2-preview-deepseek-671B-MoE-GGUF",
  "modelId": "unsloth/cogito-v2-preview-deepseek-671B-MoE-GGUF",
  "sha": "af30a3b0d25104b4040b0ce12d95f51b4e51da7c",
  "createdAt": "2025-07-29T12:15:54.000Z",
  "lastModified": "2025-08-01T01:26:20.000Z",
  "author": "unsloth",
  "downloads": 92,
  "likes": 2,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "transformers",
  "siblings_count": 232
}