ScaleDockHQ/llms-skills

cloud-llms

Run and consume LLMs in the cloud - pick a tier, provision, deploy, connect to one gateway endpoint, secure, set budgets, benchmark, scale, update and tear down. Covers frontier APIs (OpenAI, Anthropic Claude, Google Gemini), managed open-model APIs (Together…

لغة النص الأصلي: الإنجليزية

fine-tuning-llms

Fine-tune open-weight LLMs locally (NVIDIA DGX Spark unified memory, consumer RTX GPUs) or on rented cloud GPUs. Covers deciding whether to fine-tune vs prompting or RAG, dataset prep (chat templates, messages JSONL, ShareGPT/Alpaca, preference pairs, dedup,…

لغة النص الأصلي: الإنجليزية

llm-clients

Connect coding agents, IDE extensions, chat UIs and SDK apps to any OpenAI- or Anthropic-compatible LLM endpoint (local engine, self-hosted server, gateway such as LiteLLM, or cloud API). Use when configuring Claude Code (ANTHROPIC_BASE_URL,…

لغة النص الأصلي: الإنجليزية

llm-evals

Evaluate LLMs on your own tasks against any OpenAI-compatible endpoint (local llama.cpp/Ollama, self-hosted vLLM/SGLang behind LiteLLM or llama-swap, or a cloud API). Use when choosing between models or endpoints, building a golden set from real prompts, A/B…

لغة النص الأصلي: الإنجليزية

llm-gateway

Put one OpenAI- and Anthropic-compatible endpoint in front of many local and cloud LLM providers. Use when choosing an LLM gateway (LiteLLM proxy, OpenRouter, Portkey, Envoy AI Gateway / Agent Router, Kong AI Gateway, Cloudflare AI Gateway), installing the…

لغة النص الأصلي: الإنجليزية

llm-observability

Metrics, tracing, logging, cost tracking and alerting for LLM serving and LLM apps. Use when instrumenting or debugging vLLM, SGLang, llama.cpp server, TGI or llama-swap /metrics endpoints; scraping them with Prometheus and building Grafana dashboards (TTFT,…

لغة النص الأصلي: الإنجليزية

local-llms

Run LLMs locally on a laptop or desktop (macOS, Windows, Linux) with Ollama, LM Studio, llama.cpp (llama-server) or MLX (mlx-lm). Use when the user wants to install a local runtime, pick and size a model or GGUF/MLX quant for their RAM or VRAM (weights + KV…

لغة النص الأصلي: الإنجليزية

model-quantization

Produce quantized LLM checkpoints and publish them. Covers GGUF (llama.cpp convert_hf_to_gguf.py, llama-quantize, llama-imatrix; Q4_K_M, Q8_0, IQ4_XS), EXL3 (ExLlamaV3 convert.py, TabbyAPI), AWQ / GPTQ / FP8 / NVFP4 with LLM Compressor (oneshot,…

لغة النص الأصلي: الإنجليزية

self-hosted-genmedia

Self-host generative media servers (image, video, speech-to-text, text-to-speech) in the same stack as your LLMs, behind a LiteLLM gateway and llama-swap. Covers stable-diffusion.cpp sd-server and ComfyUI for images; ComfyUI workflows for open video models…

لغة النص الأصلي: الإنجليزية

self-hosted-llms

Plan, set up, operate, update, benchmark, optimize and clean up a self-hosted LLM and generative-AI serving stack on your own Linux servers with NVIDIA GPUs, with NVIDIA DGX Spark (GB10, single or dual node over ConnectX-7) as the first-class profile. Builds…

لغة النص الأصلي: الإنجليزية