[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-105389":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":9,"languages":9,"totalLinesOfCode":9,"stars":10,"forks":11,"watchers":12,"openIssues":13,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":15,"stars7d":15,"stars30d":15,"stars90d":14,"forks30d":14,"starsTrendScore":16,"compositeScore":17,"rankGlobal":9,"rankLanguage":9,"license":9,"archived":18,"fork":18,"defaultBranch":19,"hasWiki":20,"hasPages":18,"topics":21,"createdAt":9,"pushedAt":9,"updatedAt":22,"readmeContent":23,"aiSummary":9,"trendingCount":14,"starSnapshotCount":14,"syncStatus":24,"lastSyncTime":25,"discoverSource":26},105389,"Clover-35B-A3B-Preview","DiffuSpace\u002FClover-35B-A3B-Preview","DiffuSpace","A 35B diffusion language model.",null,119,5,3,4,0,11,33,58.93,false,"main",true,[],"2026-11-01 00:15:39","\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fdiffuspace.ai\u002F\">\n    \u003Cimg src=\"https:\u002F\u002Fassets.diffuspace.ai\u002Fbrand\u002Flogofull.svg\" alt=\"DiffuSpace\" width=\"360\" \u002F>\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Ch1 align=\"center\">Clover-35B-A3B-Preview\u003C\u002Fh1>\n\n\u003Cp align=\"center\">\n  \u003Cb>The first diffusion language model from DiffuSpace.\u003C\u002Fb>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FDiffuspace\u002FClover-35B-A3B-Preview\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F%F0%9F%A4%97_Hugging_Face-Model-yellow\" alt=\"Hugging Face\" \u002F>\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fmodelscope.cn\u002Forganization\u002FDiffuspace\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FModelScope-DiffuSpace-blue\" alt=\"ModelScope\" \u002F>\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fdiffuspace.ai\u002Fblog\u002Fclover-35b\u002F\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FBlog-Clover-green\" alt=\"Blog\" \u002F>\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FDiffuspace\u002FClover-35B-A3B-Preview\u002Fblob\u002Fmain\u002FLICENSE\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FModel_License-Apache_2.0-blue\" alt=\"Model License: Apache 2.0\" \u002F>\u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"#introduction\">Introduction\u003C\u002Fa> ·\n  \u003Ca href=\"#performance\">Performance\u003C\u002Fa> ·\n  \u003Ca href=\"#quick-start\">Quick Start\u003C\u002Fa> ·\n  \u003Ca href=\"#inference-backends\">Inference Backends\u003C\u002Fa> ·\n  \u003Ca href=\"#supervised-fine-tuning\">Fine-Tuning\u003C\u002Fa>\n\u003C\u002Fp>\n\n## Introduction\n\nWe introduce **Clover-35B-A3B-Preview**, the first model released by **DiffuSpace**. Clover delivers state-of-the-art performance among open-source diffusion language models at release on the benchmarks reported in our [launch blog](https:\u002F\u002Fdiffuspace.ai\u002Fblog\u002Fclover-35b\u002F).\n\nBuilt on [Qwen3.6-35B-A3B](https:\u002F\u002Fhuggingface.co\u002FQwen\u002FQwen3.6-35B-A3B), Clover combines a text-only mixture-of-experts backbone with block diffusion generation:\n\n- **35B total parameters, approximately 3B active per token.** A hybrid Gated DeltaNet (GDN) \u002F full-attention MoE architecture.\n- **Bidirectional attention within each block.** Additional learned GDN parameters support block diffusion while preserving a causal prefix.\n- **128K context.** Up to 131,072 tokens across the prompt and generated response.\n\n## Performance\n\n[![Clover agentic benchmark results](.\u002Fbenchmark.png)](https:\u002F\u002Fdiffuspace.ai\u002Fblog\u002Fclover-35b\u002F#agentic-benchmarks)\nSee the [launch blog](https:\u002F\u002Fdiffuspace.ai\u002Fblog\u002Fclover-35b\u002F#agentic-benchmarks) for benchmark results, evaluation settings, and comparisons.\n\n## Quick Start\n\n### Hugging Face Transformers\n\nInstall PyTorch with CUDA support, then the Hugging Face dependencies. This example uses Transformers 5.19.0 and PyTorch 2.10 or later.\n\n```bash\npip install \"transformers==5.19.0\" accelerate safetensors\n```\n\nThe example runs on one GPU with enough memory for the BF16 weights (approximately 69 GB) plus inference state.\n\n```python\nimport torch\nfrom transformers import AutoModelForCausalLM, AutoTokenizer\n\nmodel_id = \"Diffuspace\u002FClover-35B-A3B-Preview\"\ntokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)\nmodel = AutoModelForCausalLM.from_pretrained(\n    model_id,\n    trust_remote_code=True,\n    dtype=torch.bfloat16,\n    device_map=\"cuda:0\",\n).eval()\n\nmessages = [{\"role\": \"user\", \"content\": \"What is diffusion?\"}]\nprompt = tokenizer.apply_chat_template(\n    messages,\n    tokenize=False,\n    add_generation_prompt=True,\n    enable_thinking=True,\n)\ninputs = tokenizer(\n    prompt, add_special_tokens=False, return_tensors=\"pt\"\n).to(\"cuda:0\")\n\noutput = model.generate(**inputs, max_new_tokens=24576)\ncompletion = output[0, inputs[\"input_ids\"].shape[1]:]\nprint(tokenizer.decode(completion, skip_special_tokens=True))\n```\n\nGeneration defaults are included in the checkpoint. For decoding options and configuration details, see the [Hugging Face model card](https:\u002F\u002Fhuggingface.co\u002FDiffuspace\u002FClover-35B-A3B-Preview#generation-controls).\n\n## Inference Backends\n\n| Backend | Use case | Decoding |\n| --- | --- | --- |\n| [Hugging Face Transformers](#hugging-face-transformers) | Standalone Python inference and experimentation | Block diffusion denoising |\n| [SGLang](https:\u002F\u002Fgithub.com\u002Fsgl-project\u002Fsglang\u002Fpull\u002F46927) | Serving and evaluation | Block diffusion denoising and Self-speculative decoding |\n| [llama.cpp](https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp\u002Fpull\u002F31306) | Local inference | Self-speculative decoding |\n\n### SGLang and llama.cpp\n\nOur SGLang implementation supports both **Block diffusion denoising** and **Self-speculative decoding** for Clover. The llama.cpp implementation supports **Self-speculative decoding**. Backend code and setup guides will be added to this repository.\n\n> [!NOTE]\n> We recommend our official **SGLang implementation for serving and evaluation**. Small numerical differences between the Hugging Face implementation and SGLang's kernels can lead to different generated text.\n\n## Supervised Fine-Tuning\n\n**Coming soon.** SFT code and training instructions will be released in [`sft\u002F`](.\u002Fsft\u002F).\n\n## License and Acknowledgments\n\nThe model is distributed under **Apache-2.0**. See the [model license](https:\u002F\u002Fhuggingface.co\u002FDiffuspace\u002FClover-35B-A3B-Preview\u002Fblob\u002Fmain\u002FLICENSE) and [attributions](https:\u002F\u002Fhuggingface.co\u002FDiffuspace\u002FClover-35B-A3B-Preview\u002Fblob\u002Fmain\u002FATTRIBUTIONS.md) for details.\n\nClover builds on Qwen3.6-35B-A3B. We thank the [Qwen](https:\u002F\u002Fgithub.com\u002FQwenLM), [Hugging Face Transformers](https:\u002F\u002Fgithub.com\u002Fhuggingface\u002Ftransformers), [SGLang](https:\u002F\u002Fgithub.com\u002Fsgl-project\u002Fsglang), and [llama.cpp](https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp) communities for their open-source work.\n",2,"2026-10-31 04:26:37","CREATED_QUERY"]