Article
Ling 3.1 Flash 1019: same setup, new weight shards
alphabell
• 
alphaBell Research
alphabell-research
AI & ML interests
Post-training, model behavior and transparent evaluation.
Recent Activity
commentedon a paper 2 days ago
Multiscale Policy Optimization for Long-Horizon LLM Agents posted an update 2 days ago
One number missing from LFM3's on-device story: memory at 256K context. Liquid AI reports 1.90 GB peak on iPhone 17 Pro at 32K. From the pinned config, f16 K/V cache alone would be 0.41 GB at 32K, 1.61 GB at 128K and 3.22 GB at 256K, before weights and runtime buffers. The chart shows the calculation, not a phone measurement. If you have run longer contexts on-device, share peak memory and time to first token here: https://huggingface.co/LiquidAI/LFM3-1.7B/discussions/5
Model: https://huggingface.co/LiquidAI/LFM3-1.7B new activity 2 days ago
RecursiveSI/flint:RTX 3090 (24 GB) numbers with llama.cpp, plus a CUDA fix for +15% decodeOrganizations
published an article 3 days ago
published an article 4 days ago
Article
Draw a MOSS layout once, then run it in ComfyUI
alphabell
• published an article 4 days ago
Article
52 full-attention layers versus 16: mapping Nanbeige and Qwen
alphabell
• published an article 8 days ago
Article
Five experiments inside Flint and Qwen Open
alphabell
• published an article 8 days ago
Article
One tool call, two input contracts: inspecting DeepSeek prompts
alphabell-research
• published an article 8 days ago
Article
Draw MOSS-Image layouts without doing coordinate arithmetic
alphabell
• published an article 8 days ago
Article
Three tiny bugs that broke Flint's games
alphabell
• published an article 9 days ago
Article
alphaBell Qwen Open: 27B post-training, with the tradeoffs measured
alphabell
•