Lucebox/Laguna-XS.2-DFlash-GGUF overview
Laguna XS.2 DFlash DFlash speculative decoding drafter for poolside/Laguna XS.2. Full continued training from v23 step18000 on 60k clean regenerated rows 10k @…
Runs locally from ~882.1 MB disk (4 GB VRAM class GPUs with llama.cpp / guIDE).
Repository Files & Downloads
| File | Type | Quantization | Size | Link |
|---|---|---|---|---|
| laguna-dflash.gguf | GGUF | GGUF | 882.1 MB | Download |
Model Details
Model README
Laguna-XS.2 DFlash
DFlash speculative-decoding drafter for poolside/Laguna-XS.2. Full continued
training from v23-step18000 on 60k clean regenerated rows (10k @16k ctx +
50k Open-PerfectBlend @4k ctx).
Real serving gate
| metric | prev. | now | delta |
|---|---:|---:|---|
| accept_rate | 29.5% | 41.1% | +39.5% rel |
| avg_commit | 3.36 | 4.29 | +27.7% |
| mixed tok/s | 50.47 | 65.25 | +29.3% |
| MATH tok/s | 45.25 | 69.32 | +53.2% |
| GSM8K tok/s | 50.72 | 63.84 | +25.9% |
| HumanEval tok/s | 56.36 | 63.04 | +11.9% |
| agent tok/s | 49.53 | 60.95 | +23.1% |
GGUF includes DSpark Markov/confidence aux heads
Run Lucebox/Laguna-XS.2-DFlash-GGUF with guIDE
Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.
Source: Hugging Face · Compare models