Hackernews posts about FP16

1. TurboOCR: 270–1200 img/s OCR with Paddle and TensorRT (C++/CUDA, FP16) (github.com)
   7 points by pfdomizer 23 days ago | 4 comments
2. Show HN: TurboOCR up to 1200 pages/s with Paddle and TensorRT (C++/CUDA, FP16) (github.com)
   3 points by pfdomizer 20 days ago | discuss
3. TTS engines: WebSocket vs. sync is 5.5x, INT8 slower than fp16 on M4 (ai.gopubby.com)
   2 points by KirMoisha 21 days ago | discuss
4. Show HN: OS Megakernel that match M5 Max Tok/w at 2x the Throughput on RTX 3090 (github.com)
   6 points by GreenGames 28 days ago | 1 comments
5. Bolt Graphics Targets FP64 HPC Workloads with Zeus GPU (www.hpcwire.com)
   1 points by rbanffy 5 days ago | discuss
6. Everyone Wants Servers and Nobody Wants Servers (connectedplaces.online)
   3 points by speckx 13 days ago | discuss
7. WFY24 – Solving the "Average Weather" fallacy at 8,848M (Everest) (www.wfy24.com)
   2 points by weatherfun 12 days ago | discuss
8. ONNX Runtime and CoreML May Silently Convert Your Model to FP16 (ym2132.github.io)
   98 points by Two_hands 5 months ago | 17 comments
9. Insights from 84,000 comments on "Ask HN: Who Is Hiring" using Llama3.1-70B-FP16 (exxa.notion.site)
   8 points by Blue_Cosma over 1 year ago | 1 comments
10. Running the Deepseek-R1 671B Model at FP16 Fidelity on AMD EPYC CPUs (www.servethehome.com)
    5 points by tanelpoder about 1 year ago | 2 comments
11. 90T/s on my iPhone llama3.2-1B-fp16 (www.reddit.com)
    5 points by darkolorin about 1 year ago | 1 comments
12. Defeating the Training-Inference Mismatch via FP16 (arxiv.org)
    2 points by matt_d 6 months ago | discuss
13. PyTorch 2.6 Delivers FP16 Support for x86 CPUs, Better Intel GPU Experience (www.phoronix.com)
    2 points by rbanffy over 1 year ago | discuss
14. Spectra: A Comprehensive Study of Ternary, Quantized, and FP16 Language Models (arxiv.org)
    2 points by fzliu almost 2 years ago | discuss
15. Defeating the Training-Inference Mismatch via FP16 (arxiv.org)
    1 points by billyzs 6 months ago | discuss
16. PyTorch 2.6 Delivers FP16 Support for x86 CPUs, Better Intel GPU Experience (www.phoronix.com)
    1 points by mikece over 1 year ago | discuss
17. 150 LoC CUDA I8 Matmul That Beats CuBLAS Tensor Core FP16 (github.com)
    1 points by carsonpoole almost 2 years ago | discuss
18. FlashAttention-3: 1.5-2x faster on FP16, up to 740 TFLOPS on H100 (75% util) (twitter.com)
    1 points by quxinxin almost 2 years ago | discuss
19. Show HN: Kitten TTS – 25MB CPU-Only, Open-Source TTS Model (github.com)
    1003 points by divamgupta 9 months ago | 361 comments
20. Show HN: Three new Kitten TTS models – smallest less than 25MB (github.com)
    561 points by rohan_joshi about 2 months ago | 181 comments
21. Launch HN: Deepsilicon (YC S24) – Software and hardware for ternary transformers
    189 points by areddyyt over 1 year ago | 79 comments
22. Show HN: Appear as anyone in video calls like zoom or Google meets (www.phazr.ai)
    110 points by michaelphi about 1 year ago | 51 comments
23. Show HN: ZSE – Open-source LLM inference engine with 3.9s cold starts (github.com)
    58 points by zyoralabs 2 months ago | 9 comments
24. Show HN: GPULlama3.java Llama Compilied to PTX/OpenCL Now Integrated in Quarkus
    24 points by mikepapadim 5 months ago | 6 comments
25. Show HN: Run 500B+ Parameter LLMs Locally on a Mac Mini (github.com)
    17 points by fatihturker about 2 months ago | 10 comments
26. Show HN: OpenGraviton – Run 500B+ parameter models on a consumer Mac Mini (opengraviton.github.io)
    13 points by fatihturker 2 months ago | 5 comments
27. Show HN: Lightweight Llama3 Inference Engine – CUDA C (github.com)
    12 points by abhisheknair10 over 1 year ago | discuss
28. Show HN: Slash your LLM Inference Costs with Overnight Processing
    5 points by Blue_Cosma over 1 year ago | 1 comments
29. Transformers 2.0: What Ilya and Sam Might Have Missed
    5 points by metehan777 over 1 year ago | 1 comments
30. Show HN: 1B Embeddings
    4 points by INVARIAN about 1 month ago | 2 comments
31. Tq-KV – Rust implementation of TurboQuant that works on GGUF models
    3 points by onurgokyildiz about 1 month ago | discuss
32. Show HN: HyperAI GPU Leaderboard – A benchmark comparison site for AI workloads (hyper.ai)
    2 points by Ada_trying 3 months ago | 2 comments
33. Show HN: Efficient LLM Architectures for 32GB RAM (Ternary and Sparse Inference) (github.com)
    2 points by fatihturker about 2 months ago | 1 comments
34. Show HN: Loft CLI – Fine-tune and run LLMs (1–3B) on 8 GB MacBook Air, no GPUs
    2 points by dips2umar 10 months ago | 1 comments
35. Show HN: I made Qwen3.5-4B 13% smarter by compressing it to 4-bit (huggingface.co)
    2 points by singularity_max about 2 months ago | discuss
36. Show HN: Slashing LLM Costs for Overnight Batch Inference
    2 points by Blue_Cosma over 1 year ago | discuss
37. Show HN: ONNX optimized SigLIP and related foundation models (github.com)
    2 points by rhysdg almost 2 years ago | discuss