Mabry PRO

artificial-citizen

10 1 60

https://www.joshmabry.dev

Mabry1985

AI & ML interests

None yet

Recent Activity

posted an update about 8 hours ago

New 27b fine tune that reduces the overthinking seen in the Qwen series. Matching or improving on the base models performance within noise. More evals and quants on the way. https://huggingface.co/protoLabsAI/ThinkingCap-Qwen3.6-27B-MTP-GGUF

liked a model about 8 hours ago

protoLabsAI/ThinkingCap-Qwen3.6-27B-MTP-GGUF

updated a model about 8 hours ago

protoLabsAI/ThinkingCap-Qwen3.6-27B-MTP-GGUF

View all activity

Organizations

Posts 3

Post

New 27b fine tune that reduces the overthinking seen in the Qwen series. Matching or improving on the base models performance within noise.

More evals and quants on the way.

protoLabsAI/ThinkingCap-Qwen3.6-27B-MTP-GGUF

Post

A 9B running 306 tok/s on a single GPU.

We shipped calibrated NVFP4 + MTP builds of Ornith-1.0-9B, and found something along the way: NVFP4 and speculative decoding are multiplicative on Blackwell. MTP's verify step batches compute straight into the FP4 tensor cores — +52% lift vs +17% on Q4_K_M, acceptance-controlled so it's the kernels, not the draft head.

- GGUF (6.6 GB, MTP baked in): 306 tok/s on RTX PRO 6000 — faster than our 4B record
- vLLM (10.4 GB, W4A4 + MTP sidecar): ~1.5× bf16+MTP at 55% the VRAM
- Full release gate vs bf16 on the card: FC 96% (beats base), claw −0.028, coherence verified to 60K
- On Ampere? Honestly: use Q4_K_M. The FP4 win is Blackwell's tensor cores.

Every number traces to a row in protoLabsAI/lab-benchmarks (CC-BY-4.0).

🔗 huggingface.co/protoLabsAI/Ornith-1.0-9B-NVFP4
🔗 huggingface.co/protoLabsAI/Ornith-1.0-9B-MTP-GGUF

Want a different size/format? Open a discussion — we usually ship within 48h.

View all Posts

models 0

None public yet

datasets 6