Language Models

Because standard Whisper sanitizes speech, and verbatim use cases actually need every filler, stutter, and false start.

A privacy-first Android fork that runs LLMs, image generation, and speech AI entirely offline, then locks itself behind your fingerprint.

A curated reading list that maps how the post-training world is moving from static SFT to student-rollout distillation with live teacher feedback.

An open-source AI assistant built in Kotlin that remembers context across chats, draws interactive screens, and runs a sandboxed Linux distro on your phone.

Agnes AI is a hosted multimodal API that speaks OpenAI's protocol, letting you reroute existing clients to its text, image, video, and agent models by changing a base URL.

A Chinese-language LLM curriculum that rebuilds Stanford CS336 into six hands-on assignments, from writing a tokenizer to distributed training and GRPO.

Datus is an open-source agent that keeps LLMs from hallucinating SQL by building a living, learning knowledge base around your data stack.

It breaks complex tasks across a team of specialized LLM agents that refine their own skills as they work.

HRM-Text claims to cut pretraining costs by 130–600× compute and 150–900× data, shipping a full 1B-parameter framework with FSDP2, FlashAttention 3, and a hierarchical recurrent architecture.

It turns ChatGPT’s browser-only image generation into a poolable, OpenAI-compatible API so you can self-host programmatic access to GPT-Image-2 and friends.

LinguaGacha uses LLMs to batch-translate novels, subtitles, and game scripts while auto-generating glossaries so character names stay consistent across an entire work.

TinyEngram open-sources experiments showing that DeepSeek's Engram architecture can inject domain knowledge into Qwen and Stable Diffusion more efficiently than LoRA, with less catastrophic forgetting.

Dynamic Fine-Tuning reweights each token's contribution to the SFT loss using its own predicted probability, aiming to correct the skewed reward signal that stifles generalization.

OpenMythos is an independent attempt to reconstruct Anthropic’s rumored Claude Mythos architecture as a trainable Recurrent-Depth Transformer with switchable attention and sparse MoE layers.

VideoRAG indexes absurdly long videos into a multi-modal knowledge graph so you can query them in natural language instead of scrubbing timelines.

It exists to replace 'formula first, API later' with broken experiments that teach you why PPO actually works.

GLiNER squeezes zero-shot named entity recognition into a bidirectional transformer small enough to run on a laptop, then adds a production Ray Serve stack on top.

A practical Chinese-language curriculum that drags learners from API panic to local model deployment, with Kaggle and Colab notebooks for every step.

Over a hundred original vector diagrams that deconstruct LLM architectures, training pipelines, and RL algorithms into readable maps.

An open-source project that fine-tunes large language models to decompile Linux x86_64 binaries into C code, validating the results by checking whether they re-execute and pass their original tests.