mirror of
https://github.com/ollama/ollama.git
synced 2026-07-07 00:08:13 +00:00
964ea42c09
This is a rewrite of the create functionality for the MLX engine. The core idea behind the create functionality is to break the import/convert into a pipeline of distinct phases: * Read (scan the safetensors directory for the various bits of metadata) * Classify (determine what the import type) * Plan (determine any transforms that need to be done) * Write (transform any data as necessary and write out the blobs) * Create the manifest Each architecture has a "policy" which determines how to convert the model correctly. A number of different formats for safetensors are supported including: * nvfp4 (two formats: model optimized, torch) * fp8 datatypes (convert to mxfp8) * standard bf16 based weights A number of cleanups/simplifications have been done including: * using the baked in names for the tensors instead of munging them into something else * unified 3d expert tensors (instead of separate per expert tensors) * fewer unnecessary transforms to the various tensors in a model (keep a model as close to the source as possible) * unified capability checking * draft model handling (for MTP) is done on the same path Image generation has been intentionally removed. |
||
|---|---|---|
| .. | ||
| client | ||
| blockfp8.go | ||
| blockfp8_test.go | ||
| classify.go | ||
| classify_test.go | ||
| cohere2moe.go | ||
| create.go | ||
| create_test.go | ||
| draft.go | ||
| draft_test.go | ||
| dtype.go | ||
| gemma4.go | ||
| gemma4_test.go | ||
| inventory.go | ||
| inventory_test.go | ||
| laguna.go | ||
| mlxthread.go | ||
| pipeline.go | ||
| pipeline_test.go | ||
| plan.go | ||
| plan_test.go | ||
| prequant.go | ||
| prequant_test.go | ||
| quantize.go | ||
| quantpolicy.go | ||
| qwen35.go | ||
| transform.go | ||
| writer.go | ||
| writer_test.go | ||