CLI reference
Generated from forge (voxprint_forge.cli.build_parser).
Do not edit this file by hand. Run python -m voxprint_forge.cli_reference.
forge
usage: forge [-h] [--version] [--json] [--root ROOT] [--plan PLAN] [--forge-out FORGE_OUT] {plan-summary,verify-logins,preflight,status,accounts,sources,run-all,fetch,abliterate,quantize,eval,card,publish-prep,upload,fit,canvas,diffusion-int4,compare,kl,lora,models,index} ...
positional arguments:
{plan-summary,verify-logins,preflight,status,accounts,sources,run-all,fetch,abliterate,quantize,eval,card,publish-prep,upload,fit,canvas,diffusion-int4,compare,kl,lora,models,index}
plan-summary Print the plan's time estimate.
verify-logins Check Hugging Face and Kaggle logins.
preflight Run the seconds-long checks and print the report.
status Read status.json and report stalled or failed work.
accounts Store Hugging Face, Kaggle, and Drive credentials in the OS
keyring.
sources Record that a consumer has finished with a shared download.
run-all Run every enabled model through upload.
fetch Download pinned files for one model.
abliterate Run the refusal-direction ablation for one model.
quantize Build the importance matrix and GGUF quants.
eval Record perplexity, refusal rate, and prose samples.
card Render the model card.
publish-prep Write the registry draft.
upload Upload files and verify sha256 and size. Hub repos are public.
Kaggle stays private.
fit Measure quants and record the VRAM/RAM split for each tier.
canvas Prepare canvas image tiers.
diffusion-int4 Wan 2.2 GGUF and FP8.
compare Score original and compressed arms for one synth4 job.
kl Perplexity and KL of GGUF quants against an F16 base.
lora Train a LoRA from your photos: prepare, train, extend, export,
review, continue, finish, status.
models List or fetch model-forge model ids.
index Build or push the model index, or backfill voxprint-model.json.
options:
-h, --help show this help message and exit
--version show program's version number and exit
--json Print one JSON object on stdout.
--root ROOT Repo root. Default: search from the working directory.
--plan PLAN Plan TOML. Default: plans/plan.toml.
--forge-out FORGE_OUT
Output root. Default: the plan output directory.
forge plan-summary
usage: forge plan-summary [-h]
options:
-h, --help show this help message and exit
forge verify-logins
usage: forge verify-logins [-h]
options:
-h, --help show this help message and exit
forge preflight
usage: forge preflight [-h] [--vm] [--for {all,lora}]
[--min-free-gb MIN_FREE_GB] [--refset]
options:
-h, --help show this help message and exit
--vm Check a fresh VM image before a paid GPU job: Python,
torch/CUDA, [compare] versions, build tools,
llama.cpp, the convert_hf_to_gguf.py imports, the
voxprint_common wheel, HF_TOKEN and disk. Exit 6 on
any FAIL.
--for {all,lora} With --vm: what the VM is for. 'lora' skips cmake,
ninja, nvcc, c++, llama.cpp and the converter imports.
Default: all.
--min-free-gb MIN_FREE_GB
Disk needed for --vm. Default: 150.
--refset Check this computer for Build a set from references:
diffusers img2img, Z-Image-Turbo, the photo
preparation models and env, DINOv2, CUDA 16 GB+, 32 GB
RAM and disk. Exit 6 on a missing requirement.
forge status
usage: forge status [-h]
options:
-h, --help show this help message and exit
forge accounts
usage: forge accounts [-h] {add,list,test,remove} ...
positional arguments:
{add,list,test,remove}
add Save a token for one target.
list Show connected account names.
test Check one stored connection.
remove Delete one stored credential.
options:
-h, --help show this help message and exit
forge accounts add
usage: forge accounts add [-h] --target {hf,kaggle,drive} [--token TOKEN]
[--login {paste,browser}]
options:
-h, --help show this help message and exit
--target {hf,kaggle,drive}
--token TOKEN
--login {paste,browser}
forge accounts list
usage: forge accounts list [-h]
options:
-h, --help show this help message and exit
forge accounts test
usage: forge accounts test [-h] --target {hf,kaggle,drive}
options:
-h, --help show this help message and exit
--target {hf,kaggle,drive}
forge accounts remove
usage: forge accounts remove [-h] --target {hf,kaggle,drive}
options:
-h, --help show this help message and exit
--target {hf,kaggle,drive}
forge sources
usage: forge sources [-h] {release} ...
positional arguments:
{release}
release Release one source for one consumer.
options:
-h, --help show this help message and exit
forge sources release
usage: forge sources release [-h] --source SOURCE --consumer CONSUMER
options:
-h, --help show this help message and exit
--source SOURCE
--consumer CONSUMER
forge run-all
usage: forge run-all [-h] [--jobs JOBS]
options:
-h, --help show this help message and exit
--jobs JOBS How many models run at once. Default: FORGE_JOBS or 1.
forge fetch
usage: forge fetch [-h] --model MODEL
options:
-h, --help show this help message and exit
--model MODEL Model id from the plan.
forge abliterate
usage: forge abliterate [-h] --model MODEL
options:
-h, --help show this help message and exit
--model MODEL Model id from the plan.
forge quantize
usage: forge quantize [-h] --model MODEL
options:
-h, --help show this help message and exit
--model MODEL Model id from the plan.
forge eval
usage: forge eval [-h] --model MODEL
options:
-h, --help show this help message and exit
--model MODEL Model id from the plan.
forge card
usage: forge card [-h] --model MODEL
options:
-h, --help show this help message and exit
--model MODEL Model id from the plan.
forge publish-prep
usage: forge publish-prep [-h] --model MODEL
options:
-h, --help show this help message and exit
--model MODEL Model id from the plan.
forge upload
usage: forge upload [-h] --model MODEL
options:
-h, --help show this help message and exit
--model MODEL Model id from the plan.
forge fit
usage: forge fit [-h] --model MODEL [--budget-gb BUDGET_GB]
options:
-h, --help show this help message and exit
--model MODEL Model id from the plan.
--budget-gb BUDGET_GB
Keep tiers of this size. Repeatable.
forge canvas
usage: forge canvas [-h] {run,upload} ...
positional arguments:
{run,upload}
run Run a canvas plan.
upload Upload parts that were deferred.
options:
-h, --help show this help message and exit
forge canvas run
usage: forge canvas run [-h] [--force] [--defer-upload] [--deadline DEADLINE]
PLAN
positional arguments:
PLAN
options:
-h, --help show this help message and exit
--force Start even if an LLM job is running.
--defer-upload Prepare files and skip the upload.
--deadline DEADLINE ISO time. Skip parts that would miss it.
forge canvas upload
usage: forge canvas upload [-h] [--parts PARTS] [--force] PLAN
positional arguments:
PLAN
options:
-h, --help show this help message and exit
--parts PARTS Comma-separated part ids. Default: the whole plan.
--force Start even if an LLM job is running.
forge diffusion-int4
usage: forge diffusion-int4 [-h] {run,estimates,disk-floor,fetch} ...
positional arguments:
{run,estimates,disk-floor,fetch}
run Convert Wan 2.2 to GGUF and FP8, smoke, card, and
upload.
estimates Print the planning minutes for each step.
disk-floor Exit 2 when free space stays under 15 percent after
reclaim.
fetch Download the Wan experts.
options:
-h, --help show this help message and exit
forge diffusion-int4 run
usage: forge diffusion-int4 run [-h] [--dry-run] [--svd-rank SVD_RANK]
[--done-dir DONE_DIR] [--watch WATCH]
[--only ONLY]
options:
-h, --help show this help message and exit
--dry-run CPU smoke with tiny files. No GPU and no downloads.
--svd-rank SVD_RANK SVD rank. Default 256.
--done-dir DONE_DIR Marker directory. Default: ~/.done/diffusion-int4.
--watch WATCH Watch file. Default: ~/watch.json.
--only ONLY Run one step. Earlier steps must already have markers.
forge diffusion-int4 estimates
usage: forge diffusion-int4 estimates [-h]
options:
-h, --help show this help message and exit
forge diffusion-int4 disk-floor
usage: forge diffusion-int4 disk-floor [-h] [--path PATH]
[--done-dir DONE_DIR]
options:
-h, --help show this help message and exit
--path PATH Filesystem to check. Default: forge-out.
--done-dir DONE_DIR Marker directory whose verified sources can be deleted.
forge diffusion-int4 fetch
usage: forge diffusion-int4 fetch [-h] --dest DEST
options:
-h, --help show this help message and exit
--dest DEST Directory that receives the snapshots.
forge compare
usage: forge compare [-h] [--job JOB] [--plan PLAN] [--make-fp8] [--out OUT]
[--reclaim-cache]
options:
-h, --help show this help message and exit
--job JOB Job id from the compare plan, for example D1.
--plan PLAN Compare plan. Default: plans/compare-synth4.toml.
--make-fp8 Write local FP8 files for D1 or D4 from the reference
snapshot and exit.
--out OUT Directory for --make-fp8. Files stay local and are not
uploaded.
--reclaim-cache Delete every snapshot under $FORGE_OUT/compare-cache and
exit.
forge kl
usage: forge kl [-h] [--job JOB] [--plan PLAN] [--bf16 BF16] [--quant QUANT]
[--chunks CHUNKS] [--ctx CTX] [--render RENDER]
options:
-h, --help show this help message and exit
--job JOB L-job id from the KL plan, for example L1.
--plan PLAN KL plan. Default: plans/kl-synth4.toml.
--bf16 BF16 Hugging Face repo@revision of the BF16 safetensors.
--quant QUANT Repeatable. A local .gguf, or repo@revision:*.QUANT*.gguf.
--chunks CHUNKS Perplexity chunks. Default: the plan, or 100.
--ctx CTX Perplexity context. Default: the plan, or 512.
--render RENDER Rewrite summary.json/summary.md in this results folder from
its summary.json. No GPU, no download.
forge lora
usage: forge lora [-h]
{prepare,export,train,extend,refset,review,continue,finish,status} ...
positional arguments:
{prepare,export,train,extend,refset,review,continue,finish,status}
prepare Check, crop and describe photos into a project.
export Save the LoRA as one .safetensors file with voxprint-
model.json.
train Train a LoRA on the prepared photos of a project.
extend Continue an existing LoRA with the newest photos of a
project.
refset Build a photo set from 1-10 references, round by
round.
review List or mark the samples of a paused review.
continue Train the next round on the liked pictures, then pause
for another review.
finish Save the current LoRA, draw a final preview, and
record the steps it got.
status Read live training progress from train.json.
options:
-h, --help show this help message and exit
forge lora prepare
usage: forge lora prepare [-h] --project PROJECT [--photos PHOTOS]
[--photo PHOTO] [--exclude EXCLUDE]
[--kind {person,object,style}] [--name NAME]
[--stage-name STAGE_NAME] [--trigger TRIGGER]
[--consent-obtained] [--subject-adult]
[--subject-age SUBJECT_AGE] [--rights-confirmed]
[--release RELEASE] [--consent-method {verbal,file}]
[--adult-content]
[--provenance-note PROVENANCE_NOTE]
[--mode {simple,expert}]
[--allow-count-outside-range]
[--crop {subject,whole}] [--captions CAPTIONS]
[--caption-model {qwen3-vl-4b-instruct,florence2-large}]
[--min-side MIN_SIDE]
[--blur-threshold BLUR_THRESHOLD]
[--dup-threshold DUP_THRESHOLD] [--fetch-missing |
--no-fetch]
options:
-h, --help show this help message and exit
--project PROJECT Project folder.
--photos PHOTOS Folder of JPG, PNG or WebP photos.
--photo PHOTO One photo. Repeat as needed.
--exclude EXCLUDE Leave this photo out.
--kind {person,object,style}
What the photos show.
--name NAME Name of the subject.
--stage-name STAGE_NAME
Optional alias; becomes the trigger word.
--trigger TRIGGER Word that calls the model in a prompt.
--consent-obtained Person: consent was obtained.
--subject-adult Person: the subject is 18 or older.
--subject-age SUBJECT_AGE
Optional whole number from 18 to 99. Only for a person
confirmed as 18 or older.
--rights-confirmed Object or style: you have the rights.
--release RELEASE Optional model release file.
--consent-method {verbal,file}
Person: verbal (no file) or file (needs --release).
Default: file with --release, else verbal.
--adult-content Person, 18+ only: the photos contain adult content (a
label).
--provenance-note PROVENANCE_NOTE
Optional note on where the photos come from.
--mode {simple,expert}
Photo limits: 10-30, or 5-60 in Expert.
--allow-count-outside-range
Expert limits: warn outside 10-30, allowed 5-60.
--crop {subject,whole}
How photos are cropped.
--captions CAPTIONS JSONL caption overrides: id, long, short.
--caption-model {qwen3-vl-4b-instruct,florence2-large}
Captioner.
--min-side MIN_SIDE Smallest short side in pixels (Expert).
--blur-threshold BLUR_THRESHOLD
Blur limit (Expert).
--dup-threshold DUP_THRESHOLD
Duplicate limit (Expert).
--fetch-missing Fetch missing model ids before this command.
--no-fetch Do not fetch model ids.
forge lora export
usage: forge lora export [-h] --project PROJECT [--checkpoint CHECKPOINT]
[--out OUT] [--install-to-mdl] [--tag TAG]
options:
-h, --help show this help message and exit
--project PROJECT Project folder.
--checkpoint CHECKPOINT
Step of the version to save. Default: last.
--out OUT Target .safetensors. Default: <project>/export/.
--install-to-mdl Also add it to your models for Canvas.
--tag TAG Plain-word tag for Canvas. Repeat as needed.
forge lora train
usage: forge lora train [-h] --project PROJECT
[--preset {balanced,best,quick}]
[--likeness {balanced,close,flexible}]
[--vram {auto,16,24}]
[--te-quant {auto,none,int8,nf4}]
[--te-device {gpu,cpu}] [--steps STEPS] [--lr LR]
[--scheduler {constant,cosine}] [--warmup WARMUP]
[--batch BATCH] [--grad-accum GRAD_ACCUM]
[--precision {bf16,fp16}]
[--grad-checkpointing | --no-grad-checkpointing]
[--caption-mode {long,short,both}]
[--caption-dropout CAPTION_DROPOUT]
[--captions CAPTIONS] [--seed SEED]
[--save-every SAVE_EVERY] [--previews | --no-previews]
[--preview-every PREVIEW_EVERY]
[--preview-prompts PREVIEW_PROMPTS] [--resume]
[--dry-run] [--fetch-missing | --no-fetch]
[--review-rounds REVIEW_ROUNDS]
[--first-round-steps FIRST_ROUND_STEPS]
[--first-round-samples FIRST_ROUND_SAMPLES]
[--samples-per-round SAMPLES_PER_ROUND]
[--sample-steps SAMPLE_STEPS] [--base BASE]
[--rank RANK] [--alpha ALPHA]
[--resolution RESOLUTION] [--train-text-encoder]
options:
-h, --help show this help message and exit
--project PROJECT Project folder.
--preset {balanced,best,quick}
Ready-made settings.
--likeness {balanced,close,flexible}
How close to the photos.
--vram {auto,16,24} Graphics memory tier in GB.
--te-quant {auto,none,int8,nf4}
MOSS text encoder format (Expert). Default auto: bf16
on 24 GB, int8 on 16 GB.
--te-device {gpu,cpu}
Where the MOSS text encoder runs (Expert). Default
gpu; cpu is a slower fallback.
--steps STEPS Training steps (Expert).
--lr LR Learning rate (Expert).
--scheduler {constant,cosine}
Learning-rate schedule (Expert).
--warmup WARMUP Warm-up steps (Expert).
--batch BATCH Photos per step (Expert).
--grad-accum GRAD_ACCUM
Gradient accumulation (Expert).
--precision {bf16,fp16}
Number format (Expert).
--grad-checkpointing, --no-grad-checkpointing
Save memory, slower.
--caption-mode {long,short,both}
Which descriptions to use.
--caption-dropout CAPTION_DROPOUT
Share of steps without a description.
--captions CAPTIONS JSONL caption overrides: id, long, short.
--seed SEED Seed for repeatable runs.
--save-every SAVE_EVERY
Save a version every N steps.
--previews, --no-previews
Draw test pictures.
--preview-every PREVIEW_EVERY
Draw test pictures every N steps.
--preview-prompts PREVIEW_PROMPTS
Text file, one test prompt per line.
--resume Continue a paused run from its last version.
--dry-run Print the plan and time estimate; train nothing.
--fetch-missing Fetch missing model ids before this command.
--no-fetch Do not fetch model ids.
--review-rounds REVIEW_ROUNDS
How many review pauses. From 1 to 6. Default: 3. 0
trains straight through.
--first-round-steps FIRST_ROUND_STEPS
Steps before the first review. From 100 to 500.
Default: 200.
--first-round-samples FIRST_ROUND_SAMPLES
Pictures drawn at the first review, at the training
resolution. Default: 6.
--samples-per-round SAMPLES_PER_ROUND
Pictures drawn after the first review. Default: 8.
--sample-steps SAMPLE_STEPS
Inference steps for each review picture. Default: 28.
--base BASE Base model MDL id. Default: z-image.
--rank RANK Adapter size (Expert).
--alpha ALPHA Adapter scale, usually equal to rank (Expert).
--resolution RESOLUTION
Training picture size, a multiple of 64.
--train-text-encoder Also train the text encoder.
forge lora extend
usage: forge lora extend [-h] --project PROJECT
[--preset {balanced,best,quick}]
[--likeness {balanced,close,flexible}]
[--vram {auto,16,24}]
[--te-quant {auto,none,int8,nf4}]
[--te-device {gpu,cpu}] [--steps STEPS] [--lr LR]
[--scheduler {constant,cosine}] [--warmup WARMUP]
[--batch BATCH] [--grad-accum GRAD_ACCUM]
[--precision {bf16,fp16}]
[--grad-checkpointing | --no-grad-checkpointing]
[--caption-mode {long,short,both}]
[--caption-dropout CAPTION_DROPOUT]
[--captions CAPTIONS] [--seed SEED]
[--save-every SAVE_EVERY]
[--previews | --no-previews]
[--preview-every PREVIEW_EVERY]
[--preview-prompts PREVIEW_PROMPTS] [--resume]
[--dry-run] [--fetch-missing | --no-fetch]
--from FROM_LORA [--replay REPLAY]
options:
-h, --help show this help message and exit
--project PROJECT Project folder.
--preset {balanced,best,quick}
Ready-made settings.
--likeness {balanced,close,flexible}
How close to the photos.
--vram {auto,16,24} Graphics memory tier in GB.
--te-quant {auto,none,int8,nf4}
MOSS text encoder format (Expert). Default auto: bf16
on 24 GB, int8 on 16 GB.
--te-device {gpu,cpu}
Where the MOSS text encoder runs (Expert). Default
gpu; cpu is a slower fallback.
--steps STEPS Training steps (Expert).
--lr LR Learning rate (Expert).
--scheduler {constant,cosine}
Learning-rate schedule (Expert).
--warmup WARMUP Warm-up steps (Expert).
--batch BATCH Photos per step (Expert).
--grad-accum GRAD_ACCUM
Gradient accumulation (Expert).
--precision {bf16,fp16}
Number format (Expert).
--grad-checkpointing, --no-grad-checkpointing
Save memory, slower.
--caption-mode {long,short,both}
Which descriptions to use.
--caption-dropout CAPTION_DROPOUT
Share of steps without a description.
--captions CAPTIONS JSONL caption overrides: id, long, short.
--seed SEED Seed for repeatable runs.
--save-every SAVE_EVERY
Save a version every N steps.
--previews, --no-previews
Draw test pictures.
--preview-every PREVIEW_EVERY
Draw test pictures every N steps.
--preview-prompts PREVIEW_PROMPTS
Text file, one test prompt per line.
--resume Continue a paused run from its last version.
--dry-run Print the plan and time estimate; train nothing.
--fetch-missing Fetch missing model ids before this command.
--no-fetch Do not fetch model ids.
--from FROM_LORA Existing LoRA .safetensors.
--replay REPLAY Share of older photos kept, 0.0-1.0.
forge lora refset
usage: forge lora refset [-h] {round,finish} ...
positional arguments:
{round,finish}
round Check the references (first time), then make and check
candidates.
finish Copy the references and kept candidates into
refset/approved.
options:
-h, --help show this help message and exit
forge lora refset round
usage: forge lora refset round [-h] --project PROJECT [--ref REF]
[--kind {person,object,style}] [--name NAME]
[--stage-name STAGE_NAME] [--trigger TRIGGER]
[--consent-obtained] [--subject-adult]
[--subject-age SUBJECT_AGE]
[--rights-confirmed] [--release RELEASE]
[--consent-method {verbal,file}]
[--adult-content]
[--provenance-note PROVENANCE_NOTE]
[--candidates CANDIDATES] [--strength STRENGTH]
[--steps STEPS] [--seed SEED]
[--fetch-missing | --no-fetch]
options:
-h, --help show this help message and exit
--project PROJECT Project folder.
--ref REF One reference picture. 1-10.
--kind {person,object,style}
What the photos show.
--name NAME Name of the subject.
--stage-name STAGE_NAME
Optional alias; becomes the trigger word.
--trigger TRIGGER Word that calls the model in a prompt.
--consent-obtained Person: consent was obtained.
--subject-adult Person: the subject is 18 or older.
--subject-age SUBJECT_AGE
Optional whole number from 18 to 99. Only for a person
confirmed as 18 or older.
--rights-confirmed Object or style: you have the rights.
--release RELEASE Optional model release file.
--consent-method {verbal,file}
Person: verbal (no file) or file (needs --release).
Default: file with --release, else verbal.
--adult-content Person, 18+ only: the photos contain adult content (a
label).
--provenance-note PROVENANCE_NOTE
Optional note on where the photos come from.
--candidates CANDIDATES
Candidates per round, 8-64. Default: 24.
--strength STRENGTH Variation strength (img2img denoise), 0.25-0.9.
Default: 0.5.
--steps STEPS Turbo steps, 4-16. Default: 9.
--seed SEED First seed of the round. Default: 1000 x round.
--fetch-missing Fetch missing model ids before this command.
--no-fetch Do not fetch model ids.
forge lora refset finish
usage: forge lora refset finish [-h] --project PROJECT
[--mode {simple,expert}] [--base BASE]
[--fetch-missing | --no-fetch]
options:
-h, --help show this help message and exit
--project PROJECT Project folder.
--mode {simple,expert}
Set size: 10-30, or 5-60 in Expert.
--base BASE Base model the set is for. Default: z-image.
--fetch-missing Fetch missing model ids before this command.
--no-fetch Do not fetch model ids.
forge lora review
usage: forge lora review [-h] [--like ID] [--dislike ID]
[--from-file FROM_FILE]
project
positional arguments:
project Project folder.
options:
-h, --help show this help message and exit
--like ID Mark this sample as liked.
--dislike ID Mark this sample as disliked.
--from-file FROM_FILE
Verdicts from a text or JSON file.
forge lora continue
usage: forge lora continue [-h] [--unattended] project
positional arguments:
project Project folder.
options:
-h, --help show this help message and exit
--unattended Keep the liked pictures and train every remaining step with no
more pauses.
forge lora finish
usage: forge lora finish [-h] project
positional arguments:
project Project folder.
options:
-h, --help show this help message and exit
forge lora status
usage: forge lora status [-h] project
positional arguments:
project Project folder.
options:
-h, --help show this help message and exit
forge models
usage: forge models [-h] {status,fetch} ...
positional arguments:
{status,fetch}
status List model ids.
fetch Fetch model ids and build folders.
options:
-h, --help show this help message and exit
forge models status
usage: forge models status [-h] [--required] [--model MODEL] [--gui]
options:
-h, --help show this help message and exit
--required Required model ids.
--model MODEL Model id. Repeat as needed.
--gui Show the progress window.
forge models fetch
usage: forge models fetch [-h] [--required] [--model MODEL] [--gui]
options:
-h, --help show this help message and exit
--required Required model ids.
--model MODEL Model id. Repeat as needed.
--gui Show the progress window.
forge index
usage: forge index [-h] {build,push,host,backfill} ...
positional arguments:
{build,push,host,backfill}
build Write index.json from local voxprint-model.json files.
Does not upload.
push Sign index.json and upload it with index.json.sig.
Does nothing unless --push is set.
host Upload official-source voxprint-model.json files into
the index dataset. Needs --push.
backfill Write voxprint-model.json for published voxprint-*
repos. Does not upload.
options:
-h, --help show this help message and exit
forge index build
usage: forge index build [-h] --from MANIFEST_FROM --out OUT
[--revision REPO=SHA] [--serial SERIAL]
options:
-h, --help show this help message and exit
--from MANIFEST_FROM Directory to scan.
--out OUT Where to write index.json.
--revision REPO=SHA Commit of voxprint-model.json in REPO, 40 hex digits.
Repeat for each repo.
--serial SERIAL Index serial. Default: the published serial plus one,
at least 2.
forge index push
usage: forge index push [-h] [--push] --index INDEX [--key-file KEY_FILE]
options:
-h, --help show this help message and exit
--push Upload to Mitroshenkov87/voxprint-model-index.
Required; omitted runs do not upload.
--index INDEX index.json to upload.
--key-file KEY_FILE Ed25519 private key PEM. Default:
VOXPRINT_MDL_INDEX_KEY_FILE. Never printed.
forge index host
usage: forge index host [-h] --from MANIFEST_FROM [--push]
options:
-h, --help show this help message and exit
--from MANIFEST_FROM Directory to scan.
--push Upload to Mitroshenkov87/voxprint-model-index under
descriptors/<id>/. Required.
forge index backfill
usage: forge index backfill [-h] --out OUT [--user USER]
options:
-h, --help show this help message and exit
--out OUT Directory for the generated files.
--user USER Hugging Face user to list.