model_specs: declare every file the safetensors sources require - #5
Closed
christopherthompson81 wants to merge 1 commit into
Closed
christopherthompson81 wants to merge 1 commit into
christopherthompson81 wants to merge 1 commit into
Conversation
Six safetensors packages list fewer files than their own source needs, so they
install cleanly and then fail at load with
missing model package file '<id>': <models>/...
`files` is an explicit list rather than a filter -- the package manager iterates
it for snapshot downloads -- so a downloader has no way to discover the rest.
`optional_files` and `optional_tensors` exist for genuinely optional resources
and load through add_optional_resource_map; everything in `files` and `tensors`
is required and throws at package.cpp:283 when absent.
chatterbox_safetensors requires 9, declared 4
index_tts2_safetensors requires 20, declared 3
omnivoice_safetensors requires 7, declared 3
qwen3_tts_1_7b_base_safetensors requires 8, declared 6
seed_vc_mlx_safetensors requires 28, declared 3
supertonic_3_safetensors requires 13, declared 3
Every file added here is already published, at exactly the path the spec
declares, in the repository the package already downloads from -- so this is a
declaration fix and needs no new hosting. Existing entries keep their order and
the new ones follow in the order the source lists them.
voxcpm2_safetensors has the same shape but is deliberately left alone: it also
requires `audiovae.safetensors`, which OpenBMB/VoxCPM2 does not publish, so
completing its `files` list would not make it loadable and the gap there is
hosting rather than declaration.
Found by checking, for every package, that each `model:`-rooted path in its
format's source appears in the package's `files`. Worth having as a build-time
check -- with the caveat that it has to be format-aware, since a GGUF source
names the same config files and the GGUF embeds them, so a format-blind version
reports three quarters of the catalogue.
Owner
Author
|
Reviewed and sent upstream as 0xShug0#550. Closing — this copy has done its job. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Review copy in the fork — not for upstream yet.
Six safetensors packages list fewer files than their own source requires, so they install cleanly and then fail at load with
missing model package file '<id>'.filesis an explicit list rather than a filter, so a downloader has no way to discover the rest.seed_vc_mlx_safetensorsindex_tts2_safetensorssupertonic_3_safetensorschatterbox_safetensorsomnivoice_safetensorsqwen3_tts_1_7b_base_safetensorsEvery file added is already published at exactly the path the spec declares, in the repo the package already downloads from — 36 of 37 checked against the HuggingFace API — so this needs no new hosting.
Two judgement calls to review
voxcpm2_safetensorsis deliberately untouched. It has the same shape, but one of its two missing files (audiovae.safetensors) is not published byOpenBMB/VoxCPM2. Completing only the hosted half would leave it still failing with one fewer clue as to why.index_tts2goes 3 → 21, not → 20. It already declaredbpe.model, which the safetensors source does not name infilesortensors. Kept rather than pruned — it ships today, and removing a file the package already fetches is a different decision from adding ones it needs.Not verified end to end
The evidence is static: the spec says the file is required, and the repo publishes it at that path. Nothing here has been installed and loaded against the patched lists.
supertonicis the cheapest real test at ~414 MB.Tracked as christopherthompson81/AudioCpp-Bindings#79.