nika_engine/core/
backend.rs

1//! Backend types for model management.
2//!
3//! These types provide a unified interface for local model management,
4//! including download progress, model info, and inference configuration.
5//!
6
7use serde::{Deserialize, Serialize};
8use std::fmt;
9use std::path::PathBuf;
10use thiserror::Error;
11
12// ============================================================================
13// Progress Types
14// ============================================================================
15
16/// Progress information during model pull/download.
17#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
18pub struct PullProgress {
19    /// Current status message (e.g., "pulling manifest", "downloading").
20    pub status: String,
21    /// Bytes completed.
22    pub completed: u64,
23    /// Total bytes to download.
24    pub total: u64,
25}
26
27impl PullProgress {
28    /// Create a new progress update.
29    #[must_use]
30    pub fn new(status: impl Into<String>, completed: u64, total: u64) -> Self {
31        Self {
32            status: status.into(),
33            completed,
34            total,
35        }
36    }
37
38    /// Get progress as a percentage (0.0 to 100.0).
39    #[must_use]
40    pub fn percent(&self) -> f64 {
41        if self.total == 0 {
42            0.0
43        } else {
44            (self.completed as f64 / self.total as f64) * 100.0
45        }
46    }
47
48    /// Check if download is complete.
49    #[must_use]
50    pub fn is_complete(&self) -> bool {
51        self.total > 0 && self.completed >= self.total
52    }
53}
54
55impl fmt::Display for PullProgress {
56    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
57        write!(f, "{}: {:.1}%", self.status, self.percent())
58    }
59}
60
61// ============================================================================
62// Model Info
63// ============================================================================
64
65/// Information about an installed model.
66#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
67pub struct ModelInfo {
68    /// Model name (e.g., "llama3.2:7b").
69    pub name: String,
70    /// Size in bytes.
71    pub size: u64,
72    /// Quantization level (e.g., "Q4_K_M", "Q8_0").
73    pub quantization: Option<String>,
74    /// Parameter count (e.g., "7B", "70B").
75    pub parameters: Option<String>,
76    /// Model digest/hash.
77    pub digest: Option<String>,
78}
79
80impl ModelInfo {
81    /// Get size in gigabytes.
82    #[must_use]
83    pub fn size_gb(&self) -> f64 {
84        self.size as f64 / 1_000_000_000.0
85    }
86
87    /// Get size as human-readable string.
88    #[must_use]
89    pub fn size_human(&self) -> String {
90        let gb = self.size_gb();
91        if gb >= 1.0 {
92            format!("{gb:.1} GB")
93        } else {
94            format!("{:.0} MB", self.size as f64 / 1_000_000.0)
95        }
96    }
97}
98
99impl fmt::Display for ModelInfo {
100    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
101        write!(f, "{} ({})", self.name, self.size_human())
102    }
103}
104
105// ============================================================================
106// Download Types
107// ============================================================================
108
109/// Request to download a model.
110#[derive(Debug, Clone)]
111pub struct DownloadRequest<'a> {
112    /// The model to download (curated).
113    pub model: Option<&'a super::models::KnownModel>,
114
115    /// HuggingFace repo (for passthrough).
116    pub hf_repo: Option<String>,
117
118    /// Specific filename to download.
119    pub filename: Option<String>,
120
121    /// Quantization level (for curated models).
122    pub quantization: Option<super::models::Quantization>,
123
124    /// Force re-download even if exists.
125    pub force: bool,
126}
127
128impl<'a> DownloadRequest<'a> {
129    /// Create a request for a curated model.
130    #[must_use]
131    pub fn curated(model: &'a super::models::KnownModel) -> Self {
132        Self {
133            model: Some(model),
134            hf_repo: None,
135            filename: None,
136            quantization: None,
137            force: false,
138        }
139    }
140
141    /// Create a request for a HuggingFace model.
142    #[must_use]
143    pub fn huggingface(repo: impl Into<String>, filename: impl Into<String>) -> Self {
144        Self {
145            model: None,
146            hf_repo: Some(repo.into()),
147            filename: Some(filename.into()),
148            quantization: None,
149            force: false,
150        }
151    }
152
153    /// Set the quantization level.
154    #[must_use]
155    pub fn with_quantization(mut self, quant: super::models::Quantization) -> Self {
156        self.quantization = Some(quant);
157        self
158    }
159
160    /// Force re-download.
161    #[must_use]
162    pub fn force(mut self) -> Self {
163        self.force = true;
164        self
165    }
166
167    /// Get the target filename for this download.
168    #[must_use]
169    pub fn target_filename(&self) -> Option<String> {
170        if let Some(filename) = &self.filename {
171            return Some(filename.clone());
172        }
173
174        if let Some(model) = self.model {
175            let quant = self
176                .quantization
177                .unwrap_or(super::models::Quantization::Q4_K_M);
178            // Find the filename for this quantization
179            return model
180                .quantizations
181                .iter()
182                .find(|(q, _)| *q == quant)
183                .map(|(_, f)| (*f).to_string());
184        }
185
186        None
187    }
188}
189
190/// Result of a model download.
191#[derive(Debug, Clone)]
192pub struct DownloadResult {
193    /// Local path to the downloaded model.
194    pub path: PathBuf,
195
196    /// Size of the downloaded file in bytes.
197    pub size: u64,
198
199    /// SHA256 checksum of the file.
200    pub checksum: Option<String>,
201
202    /// Whether the file was already cached.
203    pub cached: bool,
204}
205
206// ============================================================================
207// Native Model Kind
208// ============================================================================
209
210/// Specifies which kind of native model to load.
211///
212/// The native runtime supports two loading paths:
213/// - `TextGguf` — local GGUF files via `GgufModelBuilder` (text-only)
214/// - `VisionHf` — HuggingFace vision models via `VisionModelBuilder`
215///
216/// The default is `TextGguf` for backward compatibility. When `VisionHf`
217/// is used, `model_path` in `InferenceBackend::load()` is ignored -- the
218/// model is fetched from HuggingFace by `model_id`.
219#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
220#[serde(tag = "kind", rename_all = "snake_case")]
221pub enum NativeModelKind {
222    /// Load a local GGUF file (text-only, existing behavior).
223    #[default]
224    TextGguf,
225    /// Load a HuggingFace vision model by model ID.
226    VisionHf {
227        /// HuggingFace model ID (e.g., "HuggingFaceM4/Idefics3-8B-Llama3").
228        model_id: String,
229        /// Optional ISQ quantization type string (e.g., "Q4K", "Q8_0").
230        /// Parsed at load time via `mistralrs::parse_isq_value`.
231        isq: Option<String>,
232    },
233}
234
235impl NativeModelKind {
236    /// Returns true if this is a vision model kind.
237    #[must_use]
238    pub fn is_vision(&self) -> bool {
239        matches!(self, Self::VisionHf { .. })
240    }
241}
242
243// ============================================================================
244// Vision Image
245// ============================================================================
246
247/// An image to send to a vision model for inference.
248///
249/// Contains raw image bytes and the MIME media type. The native runtime
250/// will decode these bytes into a `DynamicImage` before passing them to
251/// the mistral.rs `VisionMessages` / `RequestBuilder` API.
252#[derive(Debug, Clone)]
253pub struct VisionImage {
254    /// Raw image bytes (PNG, JPEG, WebP, etc.).
255    pub bytes: Vec<u8>,
256    /// MIME type (e.g., "image/png", "image/jpeg").
257    pub media_type: String,
258}
259
260impl VisionImage {
261    /// Create a new vision image from bytes and media type.
262    #[must_use]
263    pub fn new(bytes: Vec<u8>, media_type: impl Into<String>) -> Self {
264        Self {
265            bytes,
266            media_type: media_type.into(),
267        }
268    }
269}
270
271// ============================================================================
272// Load Configuration
273// ============================================================================
274
275/// Configuration for loading a model.
276#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
277pub struct LoadConfig {
278    /// GPU IDs to use for this model (empty = auto).
279    pub gpu_ids: Vec<u32>,
280    /// Number of layers to offload to GPU (-1 = all, 0 = none).
281    pub gpu_layers: i32,
282    /// Context size (token window).
283    pub context_size: Option<u32>,
284    /// Keep model loaded in memory (prevent unload).
285    pub keep_alive: bool,
286    /// Which kind of native model to load.
287    /// Defaults to `TextGguf` for backward compatibility.
288    #[serde(default)]
289    pub model_kind: NativeModelKind,
290}
291
292impl Default for LoadConfig {
293    fn default() -> Self {
294        Self {
295            gpu_ids: Vec::new(),
296            gpu_layers: -1, // All layers on GPU by default
297            context_size: None,
298            keep_alive: false,
299            model_kind: NativeModelKind::default(),
300        }
301    }
302}
303
304impl LoadConfig {
305    /// Create a new load configuration.
306    #[must_use]
307    pub fn new() -> Self {
308        Self::default()
309    }
310
311    /// Set specific GPU IDs.
312    #[must_use]
313    pub fn with_gpus(mut self, gpu_ids: Vec<u32>) -> Self {
314        self.gpu_ids = gpu_ids;
315        self
316    }
317
318    /// Set GPU layers (-1 = all, 0 = CPU only).
319    #[must_use]
320    pub fn with_gpu_layers(mut self, layers: i32) -> Self {
321        self.gpu_layers = layers;
322        self
323    }
324
325    /// Set context size.
326    #[must_use]
327    pub fn with_context_size(mut self, size: u32) -> Self {
328        self.context_size = Some(size);
329        self
330    }
331
332    /// Set keep alive.
333    #[must_use]
334    pub fn with_keep_alive(mut self, keep: bool) -> Self {
335        self.keep_alive = keep;
336        self
337    }
338
339    /// Check if this is a CPU-only configuration.
340    #[must_use]
341    pub fn is_cpu_only(&self) -> bool {
342        self.gpu_layers == 0
343    }
344
345    /// Check if using all GPU layers.
346    #[must_use]
347    pub fn is_full_gpu(&self) -> bool {
348        self.gpu_layers < 0
349    }
350}
351
352// ============================================================================
353// Chat Types
354// ============================================================================
355
356/// Role in a chat conversation.
357#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
358#[serde(rename_all = "lowercase")]
359pub enum ChatRole {
360    /// System message (instructions).
361    System,
362    /// User message.
363    User,
364    /// Assistant response.
365    Assistant,
366}
367
368impl fmt::Display for ChatRole {
369    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
370        match self {
371            Self::System => write!(f, "system"),
372            Self::User => write!(f, "user"),
373            Self::Assistant => write!(f, "assistant"),
374        }
375    }
376}
377
378/// A message in a chat conversation.
379#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
380pub struct ChatMessage {
381    /// Role of the message sender.
382    pub role: ChatRole,
383    /// Content of the message.
384    pub content: String,
385}
386
387impl ChatMessage {
388    /// Create a new system message.
389    #[must_use]
390    pub fn system(content: impl Into<String>) -> Self {
391        Self {
392            role: ChatRole::System,
393            content: content.into(),
394        }
395    }
396
397    /// Create a new user message.
398    #[must_use]
399    pub fn user(content: impl Into<String>) -> Self {
400        Self {
401            role: ChatRole::User,
402            content: content.into(),
403        }
404    }
405
406    /// Create a new assistant message.
407    #[must_use]
408    pub fn assistant(content: impl Into<String>) -> Self {
409        Self {
410            role: ChatRole::Assistant,
411            content: content.into(),
412        }
413    }
414}
415
416/// Options for chat completion.
417#[derive(Debug, Clone, PartialEq, Default, Serialize, Deserialize)]
418pub struct ChatOptions {
419    /// Temperature for sampling (0.0 to 2.0).
420    pub temperature: Option<f32>,
421    /// Top-p (nucleus) sampling.
422    pub top_p: Option<f32>,
423    /// Top-k sampling.
424    pub top_k: Option<u32>,
425    /// Maximum tokens to generate.
426    pub max_tokens: Option<u32>,
427    /// Stop sequences.
428    pub stop: Vec<String>,
429    /// Seed for reproducibility.
430    pub seed: Option<u64>,
431}
432
433impl ChatOptions {
434    /// Create new chat options.
435    #[must_use]
436    pub fn new() -> Self {
437        Self::default()
438    }
439
440    /// Set temperature.
441    #[must_use]
442    pub fn with_temperature(mut self, temp: f32) -> Self {
443        self.temperature = Some(temp);
444        self
445    }
446
447    /// Set top-p sampling.
448    #[must_use]
449    pub fn with_top_p(mut self, top_p: f32) -> Self {
450        self.top_p = Some(top_p);
451        self
452    }
453
454    /// Set top-k sampling.
455    #[must_use]
456    pub fn with_top_k(mut self, top_k: u32) -> Self {
457        self.top_k = Some(top_k);
458        self
459    }
460
461    /// Set maximum tokens.
462    #[must_use]
463    pub fn with_max_tokens(mut self, max: u32) -> Self {
464        self.max_tokens = Some(max);
465        self
466    }
467
468    /// Add a stop sequence.
469    #[must_use]
470    pub fn with_stop(mut self, stop: impl Into<String>) -> Self {
471        self.stop.push(stop.into());
472        self
473    }
474
475    /// Set seed for reproducibility.
476    #[must_use]
477    pub fn with_seed(mut self, seed: u64) -> Self {
478        self.seed = Some(seed);
479        self
480    }
481}
482
483/// Response from a chat completion.
484#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
485pub struct ChatResponse {
486    /// The assistant's response message.
487    pub message: ChatMessage,
488    /// Whether the response is complete (not streaming).
489    pub done: bool,
490    /// Total duration in nanoseconds.
491    pub total_duration: Option<u64>,
492    /// Tokens generated.
493    pub eval_count: Option<u64>,
494    /// Prompt tokens.
495    pub prompt_eval_count: Option<u64>,
496}
497
498impl ChatResponse {
499    /// Get the response content.
500    #[must_use]
501    pub fn content(&self) -> &str {
502        &self.message.content
503    }
504
505    /// Get tokens per second (if metrics available).
506    #[must_use]
507    pub fn tokens_per_second(&self) -> Option<f64> {
508        match (self.eval_count, self.total_duration) {
509            (Some(count), Some(duration)) if duration > 0 => {
510                Some(count as f64 / (duration as f64 / 1_000_000_000.0))
511            }
512            _ => None,
513        }
514    }
515}
516
517// ============================================================================
518// Error Types
519// ============================================================================
520
521/// Error types for backend operations.
522#[derive(Error, Debug, Clone)]
523pub enum BackendError {
524    /// Backend server is not running.
525    #[error("Backend server is not running")]
526    NotRunning,
527
528    /// Model not found in registry or locally.
529    #[error("Model not found: {0}")]
530    ModelNotFound(String),
531
532    /// Model is already loaded.
533    #[error("Model already loaded: {0}")]
534    AlreadyLoaded(String),
535
536    /// Insufficient GPU/system memory.
537    #[error("Insufficient memory to load model")]
538    InsufficientMemory,
539
540    /// Network error during pull/API call.
541    #[error("Network error: {0}")]
542    NetworkError(String),
543
544    /// Process management error.
545    #[error("Process error: {0}")]
546    ProcessError(String),
547
548    /// Backend-specific error.
549    #[error("Backend error: {0}")]
550    BackendSpecific(String),
551
552    /// Missing API key for cloud provider.
553    #[error("Missing API key for provider: {0}")]
554    MissingApiKey(String),
555
556    /// API returned an error response.
557    #[error("API error (HTTP {status}): {message}")]
558    ApiError {
559        /// HTTP status code.
560        status: u16,
561        /// Error message from API.
562        message: String,
563    },
564
565    /// Failed to parse API response.
566    #[error("Parse error: {0}")]
567    ParseError(String),
568
569    /// Model loading failed.
570    #[error("Model load error: {0}")]
571    LoadError(String),
572
573    /// Inference failed.
574    #[error("Inference error: {0}")]
575    InferenceError(String),
576
577    /// Invalid model configuration.
578    #[error("Invalid configuration: {0}")]
579    InvalidConfig(String),
580
581    /// Storage/filesystem error.
582    #[error("Storage error: {0}")]
583    StorageError(String),
584
585    /// Download failed.
586    #[error("Download error: {0}")]
587    DownloadError(String),
588
589    /// Checksum verification failed.
590    #[error("Checksum mismatch: expected {expected}, got {actual}")]
591    ChecksumError {
592        /// Expected checksum.
593        expected: String,
594        /// Actual checksum.
595        actual: String,
596    },
597
598    /// Path traversal attack detected.
599    #[error("Path traversal detected: '{path}' escapes storage directory")]
600    PathTraversal {
601        /// The invalid path.
602        path: String,
603    },
604}
605
606impl BackendError {
607    /// Returns `true` if this error is transient and the operation should be retried.
608    #[must_use]
609    pub const fn is_retryable(&self) -> bool {
610        matches!(
611            self,
612            Self::NetworkError(_) | Self::NotRunning | Self::DownloadError(_)
613        )
614    }
615
616    /// Returns `true` if this is an authentication/authorization error.
617    #[must_use]
618    pub fn is_auth_error(&self) -> bool {
619        match self {
620            Self::MissingApiKey(_) => true,
621            Self::ApiError { status, .. } => *status == 401 || *status == 403,
622            _ => false,
623        }
624    }
625}
626
627// ============================================================================
628// Tests
629// ============================================================================
630
631#[cfg(test)]
632mod tests {
633    use super::*;
634
635    #[test]
636    fn test_pull_progress() {
637        let progress = PullProgress::new("downloading", 500, 1000);
638        assert_eq!(progress.percent(), 50.0);
639        assert!(!progress.is_complete());
640
641        let complete = PullProgress::new("complete", 1000, 1000);
642        assert!(complete.is_complete());
643    }
644
645    #[test]
646    fn test_pull_progress_display() {
647        let progress = PullProgress::new("pulling", 750, 1000);
648        assert_eq!(progress.to_string(), "pulling: 75.0%");
649    }
650
651    #[test]
652    fn test_model_info_size() {
653        let info = ModelInfo {
654            name: "llama3.2:7b".to_string(),
655            size: 4_500_000_000,
656            quantization: Some("Q4_K_M".to_string()),
657            parameters: Some("7B".to_string()),
658            digest: None,
659        };
660
661        assert!((info.size_gb() - 4.5).abs() < 0.01);
662        assert_eq!(info.size_human(), "4.5 GB");
663    }
664
665    #[test]
666    fn test_load_config_default() {
667        let config = LoadConfig::default();
668        assert!(config.gpu_ids.is_empty());
669        assert_eq!(config.gpu_layers, -1);
670        assert!(config.is_full_gpu());
671        assert!(!config.is_cpu_only());
672    }
673
674    #[test]
675    fn test_load_config_builder() {
676        let config = LoadConfig::new()
677            .with_gpus(vec![0, 1])
678            .with_gpu_layers(32)
679            .with_context_size(8192)
680            .with_keep_alive(true);
681
682        assert_eq!(config.gpu_ids, vec![0, 1]);
683        assert_eq!(config.gpu_layers, 32);
684        assert_eq!(config.context_size, Some(8192));
685        assert!(config.keep_alive);
686    }
687
688    #[test]
689    fn test_chat_message_constructors() {
690        let system = ChatMessage::system("You are helpful");
691        assert_eq!(system.role, ChatRole::System);
692        assert_eq!(system.content, "You are helpful");
693
694        let user = ChatMessage::user("Hello");
695        assert_eq!(user.role, ChatRole::User);
696
697        let assistant = ChatMessage::assistant("Hi there!");
698        assert_eq!(assistant.role, ChatRole::Assistant);
699    }
700
701    #[test]
702    fn test_chat_options_builder() {
703        let options = ChatOptions::new()
704            .with_temperature(0.7)
705            .with_top_p(0.9)
706            .with_max_tokens(100);
707
708        assert_eq!(options.temperature, Some(0.7));
709        assert_eq!(options.top_p, Some(0.9));
710        assert_eq!(options.max_tokens, Some(100));
711    }
712
713    #[test]
714    fn test_backend_error_is_retryable() {
715        assert!(BackendError::NetworkError("timeout".to_string()).is_retryable());
716        assert!(BackendError::NotRunning.is_retryable());
717        assert!(!BackendError::ModelNotFound("model".to_string()).is_retryable());
718        assert!(!BackendError::InsufficientMemory.is_retryable());
719    }
720
721    // ========================================================================
722    // NativeModelKind serde tests
723    // ========================================================================
724
725    #[test]
726    fn test_native_model_kind_serde_text_gguf() {
727        let kind = NativeModelKind::TextGguf;
728        let json = serde_json::to_string(&kind).unwrap();
729        assert!(json.contains("text_gguf"));
730        let roundtrip: NativeModelKind = serde_json::from_str(&json).unwrap();
731        assert_eq!(roundtrip, NativeModelKind::TextGguf);
732    }
733
734    #[test]
735    fn test_native_model_kind_serde_vision_hf() {
736        let kind = NativeModelKind::VisionHf {
737            model_id: "Qwen/Qwen2.5-VL-7B-Instruct".to_string(),
738            isq: Some("Q4K".to_string()),
739        };
740        let json = serde_json::to_string(&kind).unwrap();
741        assert!(json.contains("vision_hf"));
742        assert!(json.contains("Qwen/Qwen2.5-VL-7B-Instruct"));
743        assert!(json.contains("Q4K"));
744        let roundtrip: NativeModelKind = serde_json::from_str(&json).unwrap();
745        assert_eq!(roundtrip, kind);
746    }
747
748    #[test]
749    fn test_native_model_kind_serde_vision_hf_no_isq() {
750        let kind = NativeModelKind::VisionHf {
751            model_id: "google/gemma-3-4b-it".to_string(),
752            isq: None,
753        };
754        let json = serde_json::to_string(&kind).unwrap();
755        assert!(json.contains("vision_hf"));
756        assert!(json.contains("google/gemma-3-4b-it"));
757        // isq serializes as null (no skip_serializing_if), roundtrip still works
758        let roundtrip: NativeModelKind = serde_json::from_str(&json).unwrap();
759        assert_eq!(roundtrip, kind);
760        assert_eq!(
761            roundtrip,
762            NativeModelKind::VisionHf {
763                model_id: "google/gemma-3-4b-it".to_string(),
764                isq: None,
765            }
766        );
767    }
768
769    // ========================================================================
770    // LoadConfig serde tests (vision model kind)
771    // ========================================================================
772
773    #[test]
774    fn test_load_config_serde_default_model_kind() {
775        // Deserialize without model_kind → defaults to TextGguf
776        let json = r#"{"gpu_ids":[],"gpu_layers":-1,"context_size":null,"keep_alive":false}"#;
777        let config: LoadConfig = serde_json::from_str(json).unwrap();
778        assert_eq!(config.model_kind, NativeModelKind::TextGguf);
779    }
780
781    #[test]
782    fn test_load_config_serde_with_vision_hf() {
783        let json = r#"{"gpu_ids":[],"gpu_layers":-1,"context_size":4096,"keep_alive":false,"model_kind":{"kind":"vision_hf","model_id":"Qwen/Qwen2.5-VL-7B","isq":"Q4K"}}"#;
784        let config: LoadConfig = serde_json::from_str(json).unwrap();
785        assert!(config.model_kind.is_vision());
786        match &config.model_kind {
787            NativeModelKind::VisionHf { model_id, isq } => {
788                assert_eq!(model_id, "Qwen/Qwen2.5-VL-7B");
789                assert_eq!(isq.as_deref(), Some("Q4K"));
790            }
791            _ => panic!("Expected VisionHf"),
792        }
793    }
794
795    // ========================================================================
796    // VisionImage tests
797    // ========================================================================
798
799    #[test]
800    fn test_vision_image_construction() {
801        let img = VisionImage::new(vec![0x89, 0x50, 0x4E, 0x47], "image/png");
802        assert_eq!(img.bytes.len(), 4);
803        assert_eq!(img.media_type, "image/png");
804    }
805
806    #[test]
807    fn test_vision_image_empty_bytes() {
808        let img = VisionImage::new(vec![], "image/jpeg");
809        assert_eq!(img.bytes.len(), 0);
810        assert_eq!(img.media_type, "image/jpeg");
811    }
812}
nika_engine/core/backend.rs

nika_engine/core/
backend.rs
