{"v":1,"id":"model:hf:Anbeeld/Qwen3.8-27B-DSpark-GGUF","slug":"model-anbeeld-qwen3-8-27b-dspark-gguf","kind":"model","category":"embedding","title":"Qwen3.8-27B-DSpark-GGUF","summary":"GGUF quantizations of RadixArk DSpark draft model for Qwen3.8 27B.","source":{"provider":"hf","ref":"Anbeeld/Qwen3.8-27B-DSpark-GGUF","url":"https://huggingface.co/Anbeeld/Qwen3.8-27B-DSpark-GGUF","rev":"77843de41a67edb1c6294d641ea38fa78f9c8124","fetchedAt":"2026-10-02T20:59:46.537Z","etag":"W/\"3117-tQXC2kGV9wvz/sSLrJCVwPp9EZc\""},"author":{"name":"Anbeeld","url":"https://huggingface.co/Anbeeld"},"license":{"spdx":null,"raw":"other","open":null,"note":"custom license: read it at the source before installing"},"metrics":{"downloads":3416,"downloadsWeek":327838,"likes":2,"takenAt":"2026-10-02T20:59:46.537Z"},"tags":["transformers","gguf","safetensors","qwen3","feature-extraction","speculative-decoding","dspark","specforge","sglang","qwen3.8","text-generation","custom_code","text-generation-inference","endpoints_compatible","conversational"],"pipeline":"feature-extraction","links":{"github":"QwenLM/Qwen3","npm":"node-llama-cpp"},"updatedAt":"2026-09-06T17:53:12.000Z","collectedAt":"2026-10-02T20:59:46.537Z","review":{"numbers":["3,416 downloads on Hugging Face","2 likes","license other","1.0 GB for Qwen3.8-27B-DSpark-Q4_K_M.gguf","327,838 npm downloads a week for node-llama-cpp","latest node-llama-cpp@3.22.1"],"log":null},"trust":"unlabeled","health":{"status":"alive","checkedAt":"2026-10-02T20:59:46.537Z","http":200},"description":"# Qwen3.8 27B DSpark GGUF\n\nGGUF quantizations of [**RadixArk DSpark draft model**](https://huggingface.co/RadixArk/Qwen3.8-27B-DSpark) for [**Qwen3.8 27B**](https://huggingface.co/Qwen/Qwen3.8-27B).\n\nUse with [BeeLlama.cpp](https://github.com/Anbeeld/beellama.cpp), a llama.cpp fork with advanced quantization features.\n\n---\n\n# Qwen3.8-27B-DSpark\n\nA DSpark speculative-decoding draft model for Qwen3.8-27B target models, trained with [SpecForge](https://github.com/sgl-project/SpecForge) and served with [SGLang](https://github.com/sgl-project/sglang).\n\nThe checkpoint has been evaluated with both [RadixArk/Qwen3.8-27B-NVFP4](https://huggingface.co/RadixArk/Qwen3.8-27B-NVFP4) and [Qwen/Qwen3.8-27B-FP8](https://huggingface.co/Qwen/Qwen3.8-27B-FP8) targets. The acceptance-length evaluation below uses the NVFP4 target. The throughput evaluation uses the FP8 target.\n\n## Checkpoint\n\n- Draft parameters: 1,857,358,337 (1.86B)\n- Draft weight dtype: BF16\n- Hidden size: 5,120\n- Transformer layers: five full-attention layers\n- Attention: GQA with 32 query heads and eight key/value heads\n- Target auxiliary feature layers: 5, 19, 33, 47, 61\n- Markov head: VanillaMarkov, rank 256\n- Training target width: 16 future positions\n- Serving gamma: seven draft proposals\n- Target verification width: eight tokens, including the target bonus token\n- Maximum position embeddings: 262,144\n\nThe serving configuration uses `block_size=7`. The separate `training_block_size=16` records the supervision width used during training.\n\n## Acceptance length\n\nResults cover 64,675 completed requests across 17 workloads.\n\n| Category | Workload | Prompts | DSpark v1 | DSpark v2 |\n|---|---|---:|---:|---:|\n| Code | HumanEval | 164 | 3.0437 | **3.8468** |\n| Code | MBPP | 257 | 3.2299 | **4.0603** |\n| Code | LiveCodeBench | 1,055 | 2.5915 | **3.3462** |\n| Code | BigCodeBench | 1,140 | 2.7752 | **3.4678** |\n| Math | GSM8K | 1,319 | 3.6030 | **4.5162** |\n| M…\n\nSource: https://huggingface.co/Anbeeld/Qwen3.8-27B-DSpark-GGUF","install":{"kind":"model","hfId":"Anbeeld/Qwen3.8-27B-DSpark-GGUF","gated":false,"format":"gguf","files":[{"name":"Qwen3.8-27B-DSpark-Q2_K.gguf","size":682687168,"quant":"Q2_K","sha256":"5e3baf04207a673367e15bb49bc59d5b313d9979d5e909f4ef88704ba4b84b2e"},{"name":"Qwen3.8-27B-DSpark-Q3_K_M.gguf","size":887169728,"quant":"Q3_K_M","sha256":"51bcbcd5c6254109d77ec6673866e515700322d77e527a6bf47c8c71058e90ce"},{"name":"Qwen3.8-27B-DSpark-Q4_K_M.gguf","size":1104595648,"quant":"Q4_K_M","sha256":"02307a130d58ad478996c640753f2fe5f813f548c48e1881b7df56506ca7d545"},{"name":"Qwen3.8-27B-DSpark-Q5_K_M.gguf","size":1313163968,"quant":"Q5_K_M","sha256":"1d911e9e1cd395e553f9994cd28685410f0d1b2a57a22a56a02764b0f6645304"},{"name":"Qwen3.8-27B-DSpark-Q6_K.gguf","size":1534767808,"quant":"Q6_K","sha256":"580bfabb4a46fa434925969cdd601e0f4bd0b86199d834f6fe91be0b428b0fdd"},{"name":"Qwen3.8-27B-DSpark-Q8_0.gguf","size":1984580288,"quant":"Q8_0","sha256":"7087a8fd05aeb7099ef658b202b45d78629abfe99459c52fce392f022570584a"},{"name":"Qwen3.8-27B-DSpark-bf16.gguf","size":3725789920,"quant":"BF16","sha256":"dc0d5567dc2e9b956a509d9f2dfefb694f3a7e5628911c01babe9c7270fda136"}],"totalBytes":11232754528,"suggestedFile":"Qwen3.8-27B-DSpark-Q4_K_M.gguf","requirements":{"ramGb":2,"diskBytes":1104595648,"note":"estimate: suggested file size × 1.15 + 0.5 GB; a real measurement comes with lsh models"},"runWith":["llama.cpp"],"command":"lsh models install hf:Anbeeld/Qwen3.8-27B-DSpark-GGUF"}}