TelperionAI/Qwen3.8-27B-MLA
archived · low-signalmodel
Qwen3.8-27B is a hybrid model: 48 Gated DeltaNet (linear attention) layers and 16 full-attention layers with grouped-query attention (24 query heads, 4 KV heads, head dim 256). Dense, 27B params. Context: 256K tokens (config). License: apache-2.0. Vendor-reported: "Fidelity to the bf16 base model, teacher-forced on 562k held-out positions: top-1 disagreement where the base model is confident (top-1 vs top-2 logprob margin 2-5; lower is closer)."
- Vendor-reported: Fidelity to the bf16 base model, teacher-forced on 562k held-out positions: top-1 disagreement where the base model is confident (top-1 vs top-2 logprob margin 2-5; lower is closer).
- Org
- TelperionAI
- Released
- 2026-09-30
- Status
- archived · low-signal
- License
- apache-2.0 (permissive)
- Parameters
- 26.9B
- Architecture
- dense · Qwen3_5MLAForConditionalGeneration
- Context
- 262,144 tokens
- Task
- text-generation · text
- Signal
- 12 HF likes · 1288 downloads
- Archived
- 2026-10-22 (no longer listed by the collector; stats as last seen)
Runs on
mixed: real files + estimate; need = weights + FP16 KV cache for 8K tokens + 1GB. A sizing calculation, not tested.
- Q4 ~19.3 GB (estimated size): 24GB GPU (3090 / 4090), 32GB GPU (5090), 24GB GPU (7900 XTX), Mac 36GB, CPU only, 32GB RAM (slow)
- Q8 ~31.7 GB (estimated size): 48GB GPU (RTX 6000 Ada / A6000), Mac 48GB, Mac 64GB (M1 Max), CPU only, 64GB RAM (slow)
- FP16 ~58.8 GB: 96GB GPU (RTX Pro 6000 Blackwell), 1× DGX Spark (128GB), 2× DGX Spark (256GB), Strix Halo 128GB unified (~96GB usable), Mac 96GB, Mac 128GB, Mac 192GB, Mac 256GB, Mac 512GB, CPU only, 128GB RAM (slow), CPU only, 256GB RAM (slow)
Links
First published here 2026-09-30 · updated 2026-10-22 · markdown · JSON