~/ / Model files and chat templates

Model files and chat templates

Tools to look inside the files a model is made of, and the template that turns a conversation into its prompt. Read GGUF and safetensors headers, render a Hugging Face chat template the way transformers does, and write, check and test an Ollama Modelfile and its TEMPLATE.

Tools, in the order you use them

  1. Open a GGUF file: quantization, context, tokenizer, tensors, KV cacheGGUF Metadata ViewerOpen a GGUF file: quantization, context, layers, tokenizer, chat template, tensors and KV cache
  2. Open safetensors files: dtypes, parameters, weight size, KV cacheSafetensors Metadata ViewerOpen safetensors files: data types, parameters, weight size, split-model check and KV cache
  3. Render a Hugging Face chat template and see the exact promptHugging Face Chat Template TesterRender a chat template exactly like apply_chat_template: prompt text, variables and errors
  4. Build an Ollama Modelfile: FROM, SYSTEM, PARAMETER, MESSAGE and moreOllama Modelfile GeneratorBuild an Ollama Modelfile with a system message, parameters and stop sequences
  5. Check a Modelfile the way ollama create reads itOllama Modelfile CheckerCheck a Modelfile the way ollama create reads it: errors with lines, parameters and notes
  6. Run an Ollama TEMPLATE and see the exact promptOllama Template TesterRun an Ollama TEMPLATE on a conversation and see the exact prompt the model gets

Other topics