Maple
Sub-2-BPW Moe Pruning to Run Gemma 4 26B at 45 TPS on a 6GB GPU
Get picks like this daily. Today's launch numbers, the top launches, AI and tech news, and a daily opportunity pick, straight to your inbox.
Want to reach these readers? Sponsor an issue.
HN · launched Nov 8, 2026 · 21 upvotes (top 27%) · 0 views
Niche: llm inference and deployment tools · 188 launches, 6 in the last 7 days · Explore this niche →
Category and tags
- Horizontal
- Model & infra
- Developer
- AI-native
- Hobby / open-source project
Visit website ↗ Source listing ↗
More like this: all 188 launches in llm inference and deployment tools →
Could you build this?
No The product is a research-level model compression technique (sub-2-bit-per-weight MoE pruning) that runs a large model on small hardware, which requires ML research expertise that vibe coding cannot supply.
What it would actually take: Building a comparable version means implementing and validating novel quantization and expert-pruning methods on the target model, plus heavy GPU experimentation to measure quality against throughput. It requires ML engineers who understand quantization kernels, MoE routing, and CUDA performance tuning. The website text is mostly scraped GitHub page boilerplate, so the exact method is not verifiable from what was provided.
Discussion
5 comments analyzed.
Concerns raised: performance impact unclear, only tested on trivial tasks, Gemma is MoE so expert balancing may not hold, skeptical of expectations for this model
Feature requests: Mac tokens-per-second benchmarks, known coding benchmark comparison against original model, apply the process to LFM2.5-24B-A2B
Competitors
Other products that read as similar to this one : 758 launches clear the similarity bar, closest 8 shown.
Attention rank: #203 of 759 (itself plus its competitors, highest first, normalized so YC and Product Hunt are compared fairly).
Launched 366 days after the earliest competitor.
- bonsai2-small-gpu · github · 2026-09-19 · 54 upvotes · similarity 0.55
- Maple-Preview · hn · 2026-08-04 · 173 upvotes · similarity 0.53
- A 6M-token movable window on a single 46GB GPU · hn · 2026-07-28 · 7 upvotes · similarity 0.52
- Capy-flash · hn · 2026-10-16 · 8 upvotes · similarity 0.51
- nemotron-30b-cpu-256k · github · 2026-10-29 · 71 upvotes · similarity 0.50
- GPUWorks · hn · 2026-11-04 · 30 upvotes · similarity 0.49
- Fine-tune an 8B model on a 4 GB laptop GPU · hn · 2026-08-04 · 139 upvotes · similarity 0.49
- shaderscan · github · 2026-11-11 · 12 upvotes · similarity 0.49
The niche, labels, vibe code rating and competitor comparison on this page are analysis by Nicheloom. If you use them, please credit nicheloom.com. How you may use our data.