PFN open-sources PLaMo 3 Translate: 0.8B model beats rivals 30x its size
pmttyji · reddit · 2026-11-07
- Preferred Networks (PFN) released weights of PLaMo 3 Translate, a translation-specialized LLM, on Hugging Face in four sizes: 0.8B, 2B, 8B, and 31B.
- The company claims even the 0.8B model outperforms existing translation LLMs—including ones with 30x more parameters—on Japanese-English and Japanese-other language pairs.
- Key techniques: tournament-based data synthesis, back-translation, and iterative self-distillation, distilled into smaller models. Supports 53 languages, accepts context and glossaries, and offers plain vs. polite Japanese output.
- Ships in BF16 plus multiple quantizations (FP8, W4A16, W4AFP8, NVFP4, GGUF, MLX); benchmarks show near-identical accuracy to BF16.
- PLaMo Community License: free for personal and non-commercial use; commercial use is free for entities with revenue ≤ ¥1 billion (registration required), beyond that a separate license is needed.
More from Models
- ChatGPT reportedly mass-banning accounts, long-time paying users included — zyyyyl9144 · 2026-11-07
- Dev runs 284B-parameter DeepSeek-V4-Flash on CPU only, squeezed into 64GB DDR5 RAM — solyarisoftware · 2026-11-07
- Inkling 2 Small jumps +240 pts in Code Arena: WebDev, would be #1 US open model — SonglinYang4 · 2026-11-07
- MIT's Vector Policy Optimization replaces scalar RL rewards to boost LLM test-time search, accepted at NeurIPS 2026 — kastnerkyle · 2026-11-07
- Uncensored DeepSeek-distilled Qwen3.5-9B merge hits HF trending — xtools-at · 2026-11-07
- Anthropic paper: 3M tokens of moral dilemmas cut model malicious behavior to 0%, 28x efficiency — garrytan · 2026-11-07