Machine learning

调研合作伙伴，生成 LLM 生态合作所需的 GTM 和合作方案

获取 Hugging Face 上的模型、数据集和 Space 的统计信息

Run evaluations for Hugging Face Hub models using inspect-ai and lighteval on local hardware.

Use to select models to run locally with llama.cpp and GGUF on CPU, Mac Metal, CUDA, or ROCm.

Query LLMs from the CLI — pipe text for summarization, chat interactively, use local or cloud models with llm or aichat.

Author a new ShapeShifter graph-transformation pass for AMD Quark (ONNX or PyTorch) so it conforms to the pass framework's conventions and auto-registers.

Install or verify the AMD Quark package and its dependencies.

Prepare export and downstream evaluation handoff for a planned or completed Quark PTQ run.

Install or verify the correct PyTorch build for a user's accelerator backend before Quark installation.

End-to-end LLM accuracy evaluation on AMD ROCm (ROCm-only) — container setup, vLLM/SGLang/ATOM serving, lm-eval / lighteval / evalscope benchmarks.

End-to-end LLM accuracy evaluation on AMD ROCm (ROCm-only) — container setup, vLLM/SGLang/ATOM serving, lm-eval / lighteval / evalscope benchmarks.

L3 recipe that runs a Torch LLM PTQ end-to-end for AMD Quark — for PyTorch / HuggingFace transformers models (safetensors input): quantize → validate →…

Build a Quark Torch LLM PTQ quantization plan from model analysis and user intent.

Package and build custom AI models with Cog for deployment on Replicate.

Simplest distributed training API.

RNN+Transformer hybrid with O(n) inference.

Simple Preference Optimization for LLM alignment.

Delegate work to the TaskMarket agent-worker market - browse open tasks, create tasks with explicit authorization, and track/submit work, so an agent can…

Create a recurring AI-generated PostHog report — schedule a free-text prompt to run on a cron, with the LLM-synthesized markdown delivered to email, Slack, or…

Iteratively optimize expensive black-box objectives — such as materials properties, experimental yields, or simulation outputs — by learning from past…

Quantify prediction uncertainty of MACE MLIPs using committee (ensemble) models; flag high-uncertainty structures for DFT verification.

Generate crystal structures with exact composition control using DiffCSP++ (space group + Wyckoff positions), or unconditionally from trained distributions.

Benchmark MLIP accuracy against a labeled dataset — compute MAE/RMSE for energy/atom and forces, and generate parity plots.

Optional experimental GPU-accelerated batched inference for MACE, MatGL (TensorNet/CHGNet), and FairChem MLIPs using NValchemi, enabling parallel static…