S. Ota

@ota.bsky.social

484 Followers 211 Following

Interests: Reinforcement Learning, Natural Language Processing and Artificial General Intelligence. arXiv papers bot: @paper.bsky.social

A rich-style status bar to Claude Code. It shows context usage and the next reset times for the 5-hour and 7-day limits.
Nape Pro 設定 レイヤー 0 # マウス基本操作＋ブラウザ＋日本語入力 - 左クリック - 右クリック - Chrome の `戻る` - Chrome の `進む` - 英数 (長押しでレイヤー2) - かな (長押しでレイヤー1)
GMK Handarbeit+ & Cherry G81-3000HBD Handarbeit https://www.instagram.com/p/B9YikEtH8tZ
SWG Midnight Rainbow keycaps
Nepe Pro with Neo65 Core Plus
2本指ジェスチャー
Topre RealForce 89U
SWG Skeletoon
Measured on English Golden Rule Set (GRS) using mean F1 score across 60 test cases.
Added a rich-style status bar to Claude Code. It shows context usage and the next reset times for the 5-hour and 7-day limits.
# Qwen3.6 models for a 48 GB GPU # up to 182K context, ~50 tokens/s, higher quality MODEL="Qwen/Qwen3.6-27B-FP8" # full 256K context, ~3x faster, but slightly lower quality # MODEL="Qwen/Qwen3.6-35B-A3B-FP8" vllm serve ${MODEL} \ --trust-remote-code \ --tensor-parallel-size 1 \ --max-model-len auto \ --max-num-seqs 2 \ --max-num-batched-tokens 8192 \ --gpu-memory-utilization 0.95 \ --enable-auto-tool-choice \ --tool-call-parser qwen3_coder \ --reasoning-parser qwen3 \ --speculative-config.method mtp \ --speculative-config.num_speculative_tokens 2
{ "json.schemaDownload.trustedDomains": { "https://raw.githubusercontent.com/": true } }
# bash shopt -s histappend HISTCONTROL=ignoredups PROMPT_COMMAND="${PROMPT_COMMAND:+$PROMPT_COMMAND; }history -a; history -n" # zsh setopt SHARE_HISTORY setopt HIST_IGNORE_DUPS setopt HIST_FIND_NO_DUPS setopt HIST_REDUCE_BLANKS
disable the algif_aead module.
あなたが使えるツールを教えてください。
/Volumes/SDCard/network-config for Ubuntu 26.04
a WIP image of arXiv trends page
An image of 60% GRIN layout
Claude Code settings for Opus 4.6
Skyfeed settings for the Claude Code | Gemini CLI feed.
RegEx patterns for Custom Keyboard feed.
Screenshot of layers and keymap.
Results of SWE-bench Verified at 2025-04-07.
image alt text can be 2000 characters long.
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build --config Release -j 4 aria2c --header="Authorization: Bearer $(cat ~/.cache/huggingface/token)" -x 4 https://huggingface.co/google/gemma-3-4b-it-qat-q4_0-gguf/resolve/main/gemma-3-4b-it-q4_0.gguf -o gemma-3-4b-it-q4_0.gguf -d models ./build/bin/llama-server -t 4 -ngl 0 -c 8192 -n -1 --host 0.0.0.0 --port 8000 --no-webui -m models/gemma-3-4b-it-q4_0.gguf # on another terminal uv tool install -p 3.11 open-webui@latest open-webui serve --port 8080 # open http://localhost:8080/ with web browser # add "http://localhost:8000/v1" as an OpenAI API
This is a Marp theme based on the Selenized color palette. It is based on the Marp default theme, but the colors have been tweaked using the Selenized color palette.
Setup gemma-3-27b-it with vllm. At the moment you need to install both vllm and transformers from source.
Nagasaki airport
htop on the feed-generator server
Claude 3.7 Sonnet Thinking responses.
Hide `What’s happening`, `Who to follow`, `Subscribe to Premium` and more on the Twitter sidebar.
git diff --cached | sgpt "Generate a commit message following the Conventional Commits specification. e.g. feat, fix, chore, docs, etc."
A screenshot of htop on e2-micro.
Bluesky meetup!
Benchmark results of Gemma compare with Llama 2 and Mistral.
train/loss data.
DeepL translated "oracular" to "オラオラ系".
🤗 Open LLM Leaderboard 13B models
How to generate the probability of the next token using the compress function.
Code to make GPT rate the given tweet on a scale of 1 to 10.
content.js
you need high performance, high volume indexes
The timeline of open-source LLMs.
NVIDIA L4 costs $0.22 / hour.
過去30日間に Reddit や Hacker News で人気のあった学術論文(arXiv)のトップ30を集計し、Bluesky に投稿するボット
Summarize the top 30 most popular arXiv papers on Reddit and Hacker News in the last 30 days and post them on Bluesky.