@ondeinference / cli / commits / 41b6b44

Add publish-to-catalog flow after HF upload

After a fine-tuned GGUF finishes uploading to Hugging Face, let the user register it into the GresIQ model catalog and assign it straight to a running app from the same screen, so a fine-tune can go live without an App Store update. Depends on smbcloud-gresiq-sdk 0.4.9's new create_model() (not yet published to crates.io) and the matching backend endpoint in smbcloud-api (not yet deployed) -- this branch will not build from crates.io until 0.4.9 is published. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

keypair34 committed Jul 11, 2026 at 17:59 UTC 41b6b440c44aa2d8d0840c663265196df44f4676

5 files changed +294 -29

Cargo.lock

+3 -25

@@ -3351,7 +3351,7 @@ dependencies = [

3351 "reqwest 0.12.28",

3352 "serde",

3353 "serde_json",

3354 - "smbcloud-gresiq-sdk 0.4.8",

3354 + "smbcloud-gresiq-sdk",

3355 "thiserror 2.0.18",

3356 "tokio",

3357 "tsync",

@@ -3379,7 +3379,7 @@ dependencies = [

3379 "serde_json",

3380 "sha2",

3381 "smbcloud-auth-sdk",

3382 - "smbcloud-gresiq-sdk 0.3.41",

3382 + "smbcloud-gresiq-sdk",

3383 "smbcloud-model 0.3.41",

3384 "smbcloud-network 0.3.41",

3385 "tokenizers 0.20.4",

@@ -5143,25 +5143,7 @@ dependencies = [

5143

5144 [[package]]

5145 name = "smbcloud-gresiq-sdk"

5146 -version = "0.3.41"

5147 -source = "registry+https://github.com/rust-lang/crates.io-index"

5148 -checksum = "02a8c916737a712020c57191bcab728303a3963f3fd1c03bd3ddeaf67c4c0a64"

5149 -dependencies = [

5150 - "anyhow",

5151 - "log",

5152 - "reqwest 0.12.28",

5153 - "serde",

5154 - "serde_json",

5155 - "smbcloud-network 0.3.41",

5156 - "thiserror 2.0.18",

5157 - "tokio",

5158 -]

5159 -

5160 -[[package]]

5161 -name = "smbcloud-gresiq-sdk"

5162 -version = "0.4.8"

5163 -source = "registry+https://github.com/rust-lang/crates.io-index"

5164 -checksum = "27abf6a2b9d9db0246b3d68e4ce447b5569f43ede7ca4ceb08afeed4232f3c02"

5146 +version = "0.4.9"

5147 dependencies = [

5148 "anyhow",

5149 "log",

@@ -5194,8 +5176,6 @@ dependencies = [

5176 [[package]]

5177 name = "smbcloud-model"

5178 version = "0.4.8"

5197 -source = "registry+https://github.com/rust-lang/crates.io-index"

5198 -checksum = "21047dc7f858988cf73f0f70b0e4c087963464c4ba0f4d8c45fbcea4c6d9390a"

5179 dependencies = [

5180 "chrono",

5181 "log",

@@ -5227,8 +5207,6 @@ dependencies = [

5207 [[package]]

5208 name = "smbcloud-network"

5209 version = "0.4.8"

5230 -source = "registry+https://github.com/rust-lang/crates.io-index"

5231 -checksum = "5be96c7f572f0c91032c0a47818201f30481cdcf1965d86e5269a2cb8742ad2b"

5210 dependencies = [

5211 "clap 4.6.1",

5212 "log",

Cargo.toml

+1 -1

@@ -32,7 +32,7 @@ tokio = { version = "1", features = ["rt-multi-thread", "macros"] }

32 smbcloud-auth-sdk = "0.3.34"

33 smbcloud-model = "0.3.35"

34 smbcloud-network = "0.3.35"

35 -smbcloud-gresiq-sdk = "0.3.35"

35 +smbcloud-gresiq-sdk = "0.4.9"

36 candle-core = "0.10.2"

37 candle-nn = "0.10.2"

38 tokenizers = { version = "0.20", default-features = false, features = ["onig"] }

src/app.rs

+193

@@ -47,6 +47,8 @@ pub enum Screen {

47 GgufDetail,

48 CloneRepo,

49 Chat,

50 + /// Pick which app to assign a just-registered fine-tuned model to.

51 + PublishModel,

52 }

54 /// An entry in the Onde inference model picker.

@@ -247,6 +249,9 @@ pub enum AuthEvent {

249 model_id: String,

250 },

251 ModelAssignFailed(String),

252 + // publish: register a freshly HF-uploaded fine-tune into the catalog

253 + ModelRegistered(OndeModel),

254 + ModelRegisterFailed(String),

255 // downloads (catalog merged with local HF cache)

256 DownloadsLoaded(Vec<crate::hf::MergedModel>),

257 #[allow(dead_code)] // reserved for future explicit error reporting

@@ -597,6 +602,26 @@ impl App {

602 self.busy = false;

603 self.status = Status::error(msg);

604 }

605 + // publish

606 + AuthEvent::ModelRegistered(model) => {

607 + self.busy = false;

608 + let model_name = model

609 + .name

610 + .clone()

611 + .or_else(|| model.hf_repo_id.clone())

612 + .unwrap_or_else(|| "Model".to_string());

613 + self.models.push(model);

614 + self.models_cursor = self.models.len() - 1;

615 + self.screen = Screen::PublishModel;

616 + self.apps_cursor = 0;

617 + self.apps_offset = 0;

618 + self.status =

619 + Status::success(format!("{model_name} registered. Pick an app to assign it to."));

620 + }

621 + AuthEvent::ModelRegisterFailed(msg) => {

622 + self.busy = false;

623 + self.status = Status::error(format!("Registration failed: {msg}"));

624 + }

625 // downloads

626 AuthEvent::DownloadsLoaded(models) => {

627 self.busy = false;

@@ -927,6 +952,7 @@ fn handle_key(

952 Screen::FineTune => handle_key_finetune(app, key, tx),

953 Screen::CloneRepo => handle_key_clone_repo(app, key, tx),

954 Screen::Chat => handle_key_chat(app, key, tx),

955 + Screen::PublishModel => handle_key_publish_model(app, key, tx),

956 }

957 }

958

@@ -1629,6 +1655,9 @@ fn handle_key_gguf_detail(

1655 app.upload_progress = None;

1656 app.status = Status::success(format!("Uploaded: {url_owned}"));

1657 }

1658 + (Char('a'), KeyModifiers::NONE) => {

1659 + trigger_publish_model(app, tx);

1660 + }

1661 (Char('c'), KeyModifiers::CONTROL) => {

1662 app.should_quit = true;

1663 }

@@ -1718,6 +1747,131 @@ fn handle_key_gguf_detail(

1747 }

1748 }

1749

1750 +/// Registers the just-uploaded GGUF into the GresIQ catalog, private to this

1751 +/// account, then moves to the app picker so it can be assigned right away.

1752 +fn trigger_publish_model(app: &mut App, tx: mpsc::UnboundedSender<AuthEvent>) {

1753 + if app.busy {

1754 + return;

1755 + }

1756 + let Some(ref gguf) = app.selected_gguf else {

1757 + return;

1758 + };

1759 + let hf_repo_id = app.upload_repo_name.clone();

1760 + if hf_repo_id.is_empty() {

1761 + app.status = Status::error("No uploaded repo to register.");

1762 + return;

1763 + }

1764 + let gguf_file = gguf.file_name.clone();

1765 + let approx_size_bytes = std::fs::metadata(&gguf.path).ok().map(|m| m.len() as i64);

1766 + let base_model_id = app.current_project.as_ref().map(|p| p.base_model_id.clone());

1767 + let (base_name, family, parameter_class) = derive_base_model_meta(base_model_id.as_deref());

1768 + let display_name = format!("{base_name} (Fine-tuned)");

1769 + let token = token::load().unwrap_or_default();

1770 +

1771 + // Make sure the apps list is ready by the time the picker shows up.

1772 + if !app.apps_loaded {

1773 + trigger_load_apps(app, tx.clone());

1774 + }

1775 +

1776 + app.busy = true;

1777 + app.status = Status::neutral("Registering model…");

1778 +

1779 + tokio::spawn(async move {

1780 + match crate::gresiq::create_model(

1781 + &token,

1782 + &hf_repo_id,

1783 + &display_name,

1784 + &family,

1785 + &parameter_class,

1786 + Some(&gguf_file),

1787 + approx_size_bytes,

1788 + )

1789 + .await

1790 + {

1791 + Ok(model) => {

1792 + let _ = tx.send(AuthEvent::ModelRegistered(model));

1793 + }

1794 + Err(e) => {

1795 + let _ = tx.send(AuthEvent::ModelRegisterFailed(e.to_string()));

1796 + }

1797 + }

1798 + });

1799 +}

1800 +

1801 +/// Best-effort display name / family / parameter class for a fine-tune,

1802 +/// derived from the base model it was trained from. Falls back to generic

1803 +/// values when the base model isn't one of the officially supported ones.

1804 +fn derive_base_model_meta(base_model_id: Option<&str>) -> (String, String, String) {

1805 + let base_model_id = base_model_id.unwrap_or("");

1806 + let info = onde::inference::models::SUPPORTED_MODEL_INFO

1807 + .iter()

1808 + .find(|i| i.id == base_model_id);

1809 +

1810 + let base_name = info.map(|i| i.name.to_string()).unwrap_or_else(|| {

1811 + base_model_id

1812 + .rsplit('/')

1813 + .next()

1814 + .filter(|s| !s.is_empty())

1815 + .unwrap_or("Custom model")

1816 + .to_string()

1817 + });

1818 +

1819 + let haystack = format!("{base_model_id} {base_name}").to_lowercase();

1820 + let family = if haystack.contains("qwen3") {

1821 + "qwen3"

1822 + } else if haystack.contains("qwen2.5-coder") || haystack.contains("qwen2_5-coder") {

1823 + "qwen2.5-coder"

1824 + } else if haystack.contains("qwen2.5") || haystack.contains("qwen2_5") {

1825 + "qwen2.5"

1826 + } else {

1827 + "custom"

1828 + }

1829 + .to_string();

1830 +

1831 + const SIZES: &[&str] = &["0.5B", "1.5B", "3B", "4B", "7B", "14B", "32B", "72B"];

1832 + let haystack_upper = haystack.to_uppercase();

1833 + let parameter_class = SIZES

1834 + .iter()

1835 + .find(|s| haystack_upper.contains(*s))

1836 + .map(|s| s.to_string())

1837 + .unwrap_or_else(|| "unknown".to_string());

1838 +

1839 + (base_name, family, parameter_class)

1840 +}

1841 +

1842 +fn handle_key_publish_model(

1843 + app: &mut App,

1844 + key: crossterm::event::KeyEvent,

1845 + tx: mpsc::UnboundedSender<AuthEvent>,

1846 +) {

1847 + use KeyCode::*;

1848 +

1849 + match (key.code, key.modifiers) {

1850 + (Up, _) | (Char('k'), KeyModifiers::NONE) => {

1851 + app.apps_cursor = app.apps_cursor.saturating_sub(1);

1852 + clamp_apps_scroll(app, MAX_VISIBLE);

1853 + }

1854 + (Down, _) | (Char('j'), KeyModifiers::NONE) => {

1855 + if app.apps_cursor + 1 < app.apps.len() {

1856 + app.apps_cursor += 1;

1857 + }

1858 + clamp_apps_scroll(app, MAX_VISIBLE);

1859 + }

1860 + (Enter, _) if !app.apps.is_empty() => {

1861 + app.assigning_for_app_index = Some(app.apps_cursor);

1862 + submit_assign_model(app, tx);

1863 + }

1864 + (Esc, _) => {

1865 + app.screen = Screen::GgufDetail;

1866 + app.status = Status::neutral("Model registered — assign it later from Apps.");

1867 + }

1868 + (Char('c'), KeyModifiers::CONTROL) => {

1869 + app.should_quit = true;

1870 + }

1871 + _ => {}

1872 + }

1873 +}

1874 +

1875 fn format_adapter_size(bytes: u64) -> String {

1876 if bytes >= 1_000_000_000 {

1877 format!("{:.1}GB", bytes as f64 / 1e9)

@@ -2839,3 +2993,42 @@ fn trigger_inference_download(app: &mut App, tx: mpsc::UnboundedSender<AuthEvent

2993 }

2994 });

2995 }

2996 +

2997 +#[cfg(test)]

2998 +mod tests {

2999 + use super::*;

3000 +

3001 + #[test]

3002 + fn test_derive_base_model_meta_known_qwen3() {

3003 + let (name, family, parameter_class) =

3004 + derive_base_model_meta(Some("bartowski/Qwen_Qwen3-4B-GGUF"));

3005 + assert_eq!(family, "qwen3");

3006 + assert_eq!(parameter_class, "4B");

3007 + assert!(!name.is_empty());

3008 + }

3009 +

3010 + #[test]

3011 + fn test_derive_base_model_meta_known_qwen25_coder() {

3012 + let (_, family, parameter_class) = derive_base_model_meta(Some(

3013 + "bartowski/Qwen2.5-Coder-1.5B-Instruct-GGUF",

3014 + ));

3015 + assert_eq!(family, "qwen2.5-coder");

3016 + assert_eq!(parameter_class, "1.5B");

3017 + }

3018 +

3019 + #[test]

3020 + fn test_derive_base_model_meta_unknown_falls_back() {

3021 + let (name, family, parameter_class) = derive_base_model_meta(Some("someone/mystery-model"));

3022 + assert_eq!(family, "custom");

3023 + assert_eq!(parameter_class, "unknown");

3024 + assert_eq!(name, "mystery-model");

3025 + }

3026 +

3027 + #[test]

3028 + fn test_derive_base_model_meta_none() {

3029 + let (name, family, parameter_class) = derive_base_model_meta(None);

3030 + assert_eq!(family, "custom");

3031 + assert_eq!(parameter_class, "unknown");

3032 + assert_eq!(name, "Custom model");

3033 + }

3034 +}

src/gresiq.rs

+35 -1

@@ -1,6 +1,6 @@

1 //! Onde credentials baked in. Wraps `smbcloud_gresiq_sdk` so callers only

2 //! need to pass the user's access token.

3 -use smbcloud_gresiq_sdk::{Environment, OndeApp, OndeModel};

3 +use smbcloud_gresiq_sdk::{CreateModelParams, Environment, OndeApp, OndeModel};

5 fn api_key() -> &'static str {

6 crate::app::GRESIQ_API_KEY

@@ -34,6 +34,40 @@ pub async fn load_models(token: &str) -> anyhow::Result<Vec<OndeModel>> {

34 .map_err(|e| anyhow::anyhow!("{e}"))

35 }

37 +/// Registers a fine-tuned model (just uploaded to Hugging Face) into the

38 +/// catalog, private to this account, so it can then be assigned to an app.

39 +#[allow(clippy::too_many_arguments)]

40 +pub async fn create_model(

41 + token: &str,

42 + hf_repo_id: &str,

43 + name: &str,

44 + family: &str,

45 + parameter_class: &str,

46 + gguf_file: Option<&str>,

47 + approx_size_bytes: Option<i64>,

48 +) -> anyhow::Result<OndeModel> {

49 + smbcloud_gresiq_sdk::create_model(

50 + &Environment::Production,

51 + api_key(),

52 + api_secret(),

53 + token,

54 + CreateModelParams {

55 + hf_repo_id,

56 + name,

57 + org: "Custom",

58 + family,

59 + parameter_class,

60 + format: "gguf",

61 + gguf_file,

62 + modality: Some("text"),

63 + description: Some("Fine-tuned with onde-cli."),

64 + approx_size_bytes,

65 + },

66 + )

67 + .await

68 + .map_err(|e| anyhow::anyhow!("{e}"))

69 +}

70 +

71 pub async fn assign_model(token: &str, onde_app_id: &str, model_id: &str) -> anyhow::Result<()> {

72 smbcloud_gresiq_sdk::assign_model(

73 &Environment::Production,

src/ui.rs

+62 -2

@@ -89,6 +89,7 @@ fn render_card(frame: &mut Frame, app: &App, area: Rect) {

89 Screen::FineTune => render_finetune(frame, app, inner),

90 Screen::CloneRepo => render_clone_repo(frame, app, inner),

91 Screen::Chat => render_chat(frame, app, inner),

92 + Screen::PublishModel => render_publish_model(frame, app, inner),

93 }

94 }

@@ -461,6 +462,47 @@ fn resolve_model_name<'a>(app: &'a App, onde_app: &'a OndeApp) -> &'a str {

462 "No model assigned yet"

463 }

464

465 +// publish (assign a just-registered model) screen

466 +

467 +fn render_publish_model(frame: &mut Frame, app: &App, area: Rect) {

468 + let model_name = app

469 + .models

470 + .get(app.models_cursor)

471 + .and_then(|m| m.name.clone())

472 + .unwrap_or_else(|| "model".to_string());

473 + let heading = format!("Assign {model_name} to app");

474 +

475 + let rows = Layout::vertical([

476 + Constraint::Length(1), // heading

477 + Constraint::Length(1), // spacer

478 + Constraint::Length(1), // column header

479 + Constraint::Length(1), // divider

480 + Constraint::Min(0), // apps list

481 + Constraint::Length(1), // spacer

482 + Constraint::Length(1), // status

483 + ])

484 + .split(area);

485 +

486 + frame.render_widget(

487 + Paragraph::new(heading).style(Style::new().fg(C_INK).bold()),

488 + rows[0],

489 + );

490 +

491 + frame.render_widget(

492 + Paragraph::new(" Name Status Model").style(Style::new().fg(C_MUTED)),

493 + rows[2],

494 + );

495 +

496 + let divider = "─".repeat(area.width as usize);

497 + frame.render_widget(

498 + Paragraph::new(divider).style(Style::new().fg(C_LINE)),

499 + rows[3],

500 + );

501 +

502 + render_apps_list(frame, app, rows[4]);

503 + render_status(frame, app, rows[6]);

504 +}

505 +

506 // app detail screen

507

508 fn render_app_detail(frame: &mut Frame, app: &App, area: Rect) {

@@ -1428,7 +1470,7 @@ fn render_gguf_detail(frame: &mut Frame, app: &App, area: Rect) {

1470 app.upload_progress,

1471 Some(crate::hf_upload::UploadProgress::Done { .. })

1472 ) {

1431 - "Esc · back"

1473 + "a · assign to app Esc · back"

1474 } else if uploadable {

1475 "Enter · upload c · chat Esc · back"

1476 } else {

@@ -2608,11 +2650,19 @@ fn render_footer(frame: &mut Frame, app: &App, area: Rect) {

2650 .as_ref()

2651 .map(|g| g.is_uploadable())

2652 .unwrap_or(false);

2653 + let uploaded = matches!(

2654 + app.upload_progress,

2655 + Some(crate::hf_upload::UploadProgress::Done { .. })

2656 + );

2657 let mut keys = Vec::new();

2612 - if uploadable {

2658 + if uploadable && !uploaded {

2659 keys.push(Span::styled("Enter", Style::new().fg(C_NEON)));

2660 keys.push(Span::styled(" · upload ", Style::new().fg(C_MUTED)));

2661 }

2662 + if uploaded {

2663 + keys.push(Span::styled("a", Style::new().fg(C_NEON)));

2664 + keys.push(Span::styled(" · assign to app ", Style::new().fg(C_MUTED)));

2665 + }

2666 keys.push(Span::styled("c", Style::new().fg(C_NEON)));

2667 keys.push(Span::styled(" · chat ", Style::new().fg(C_MUTED)));

2668 keys.push(Span::styled("Esc", Style::new().fg(C_NEON)));

@@ -2664,6 +2714,16 @@ fn render_footer(frame: &mut Frame, app: &App, area: Rect) {

2714 Span::styled("Ctrl+C", Style::new().fg(C_NEON)),

2715 Span::styled(" · quit", Style::new().fg(C_MUTED)),

2716 ],

2717 + Screen::PublishModel => vec![

2718 + Span::styled("↑↓", Style::new().fg(C_NEON)),

2719 + Span::styled(" · navigate ", Style::new().fg(C_MUTED)),

2720 + Span::styled("Enter", Style::new().fg(C_NEON)),

2721 + Span::styled(" · assign ", Style::new().fg(C_MUTED)),

2722 + Span::styled("Esc", Style::new().fg(C_NEON)),

2723 + Span::styled(" · skip ", Style::new().fg(C_MUTED)),

2724 + Span::styled("Ctrl+C", Style::new().fg(C_NEON)),

2725 + Span::styled(" · quit", Style::new().fg(C_MUTED)),

2726 + ],

2727 };

2728

2729 frame.render_widget(