Store › model › Embeddings
Qwen3.8-27B-DSpark-GGUF
by Anbeeld · source Hugging Face · updated 2026-09-06
other1.0 GB~2 GB RAMsource aliveunlabeled
GGUF quantizations of RadixArk DSpark draft model for Qwen3.8 27B.
Add to LogiShell Open in the web IDE
The button opens LogiShell with this card; nothing installs from a link by itself. Inside the app the install goes through lsh models install hf:Anbeeld/Qwen3.8-27B-DSpark-GGUF and its progress lives in the Resource Center.
Source and license
- Source: https://huggingface.co/Anbeeld/Qwen3.8-27B-DSpark-GGUF
- License: other (custom license: read it at the source before installing)
- Requirements: about 2 GB of RAM, 1.0 GB on disk (estimate: suggested file size × 1.15 + 0.5 GB; a real measurement comes with lsh models). Runs with llama.cpp.
- Tags:
transformersggufsafetensorsqwen3feature-extractionspeculative-decodingdsparkspecforgesglangqwen3.8text-generationcustom_codetext-generation-inferenceendpoints_compatibleconversational
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
Numbers as of 2026-10-02 20:59 UTC, from the source API.
Summary
Summary not ready yet: the numbers are here, the text is not. It is written by the collector through the LogiShell model facade when a provider key is present.
Reviews
No reviews yet. Reviews are written inside LogiShell: open this card in the app.
Files
| file | quant | size |
|---|---|---|
Qwen3.8-27B-DSpark-Q2_K.gguf | Q2_K | 651 MB |
Qwen3.8-27B-DSpark-Q3_K_M.gguf | Q3_K_M | 846 MB |
Qwen3.8-27B-DSpark-Q4_K_M.gguf | Q4_K_M | 1.0 GB |
Qwen3.8-27B-DSpark-Q5_K_M.gguf | Q5_K_M | 1.2 GB |
Qwen3.8-27B-DSpark-Q6_K.gguf | Q6_K | 1.4 GB |
Qwen3.8-27B-DSpark-Q8_0.gguf | Q8_0 | 1.8 GB |
Qwen3.8-27B-DSpark-bf16.gguf | BF16 | 3.5 GB |
From the source README
Qwen3.8 27B DSpark GGUF
GGUF quantizations of RadixArk DSpark draft model for Qwen3.8 27B.
Use with BeeLlama.cpp, a llama.cpp fork with advanced quantization features.
Qwen3.8-27B-DSpark
A DSpark speculative-decoding draft model for Qwen3.8-27B target models, trained with SpecForge and served with SGLang.
The checkpoint has been evaluated with both RadixArk/Qwen3.8-27B-NVFP4 and Qwen/Qwen3.8-27B-FP8 targets. The acceptance-length evaluation below uses the NVFP4 target. The throughput evaluation uses the FP8 target.
Checkpoint
- Draft parameters: 1,857,358,337 (1.86B)
- Draft weight dtype: BF16
- Hidden size: 5,120
- Transformer layers: five full-attention layers
- Attention: GQA with 32 query heads and eight key/value heads
- Target auxiliary feature layers: 5, 19, 33, 47, 61
- Markov head: VanillaMarkov, rank 256
- Training target width: 16 future positions
- Serving gamma: seven draft proposals
- Target verification width: eight tokens, including the target bonus token
- Maximum position embeddings: 262,144
The serving configuration uses `block_size=7`. The separate `training_block_size=16` records the supervision width used during training.
Acceptance length
Results cover 64,675 completed requests across 17 workloads.
| Category | Workload | Prompts | DSpark v1 | DSpark v2 |
|---|---|---:|---:|---:|
| Code | HumanEval | 164 | 3.0437 | 3.8468 |
| Code | MBPP | 257 | 3.2299 | 4.0603 |
| Code | LiveCodeBench | 1,055 | 2.5915 | 3.3462 |
| Code | BigCodeBench | 1,140 | 2.7752 | 3.4678 |
| Math | GSM8K | 1,319 | 3.6030 | 4.5162 |
| M…
Card id model:hf:Anbeeld/Qwen3.8-27B-DSpark-GGUF · collected 2026-10-02 20:59 UTC · JSON