emozilla c6ed362663 fix(local-models): store ggml-org's MTP head under its own name ...

ggml-org's head and unsloth's share a filename, and a download is skipped when its destination
already exists. A machine holding unsloth's head, which aborts mainline llama.cpp at load, would
launch with it instead of fetching ggml-org's. The head now lands as
mtp-Qwen3.8-Flash-Next-ggml-org-Q8_0.gguf.

Co-authored-by: Ryan Hill <38674843+RyanUnderhill@users.noreply.github.com>

2026-10-06 16:08:23 -04:00

..

__init__.py

…

binaries.py

feat(pm): llama.cpp CUDA on Linux arm64; bundle libgomp for Linux engines

2026-10-03 22:32:13 -04:00

bootstrap.py

fix(local-runtime): leave a live owner's server to its existing path; keep legacy owner check

2026-10-03 23:06:58 -04:00

capabilities.py

Old pre-decomposition import paths are gone: plugin compat layer removed on schedule (#126164)

2026-09-28 10:21:41 -07:00

catalog.json

fix(local-models): store ggml-org's MTP head under its own name

2026-10-06 16:08:23 -04:00

catalog.py

feat(local-models): pin Flash Next's downloads to the revisions the catalog describes

2026-10-06 16:08:23 -04:00

context_policy.py

fix(local-runtime): price QSA compute buffers at the launch posture

2026-10-06 16:07:56 -04:00

detect.py

…

devices.py

fix(local-runtime): size AMD/Intel GPUs by the engine's own device type

2026-10-03 02:25:26 -07:00

endpoint.py

fix(local-runtime): budget the engine the shared server runs, in the caller's profile

2026-10-03 02:25:26 -07:00

estimator.py

fix(local-runtime): price QSA compute buffers at the launch posture

2026-10-06 16:07:56 -04:00

gguf.py

feat(local-runtime): price tensors the engine reads from disk as not loaded

2026-10-06 16:07:55 -04:00

growth.py

fix(local-runtime): read lazy tensors on demand on integrated GPUs too

2026-10-06 16:07:56 -04:00

hardware.py

fix(local-runtime): read lazy tensors on demand on integrated GPUs too

2026-10-06 16:07:56 -04:00

hf_browse.py

feat(local-models): run Qwen3.8 Flash Next as UD-IQ4_XS with its MTP head

2026-10-06 16:07:55 -04:00

load_progress.py

…

presets.py

fix(local-runtime): price QSA compute buffers at the launch posture

2026-10-06 16:07:56 -04:00

processes.py

fix: scrub credentials at the llama-server spawn only, not the generic probe spawner (review follow-up)

2026-09-20 11:50:48 -07:00

recovery.py

fix(local-runtime): leave a live owner's server to its existing path; keep legacy owner check

2026-10-03 23:06:58 -04:00

supervisor.py

fix(local-runtime): identify the managed server by boot-relative start time

2026-10-02 14:31:00 -04:00