trustformers-models 0.1.1

Model implementations for TrustformeRS
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
use crate::starcoder2::config::StarCoder2Config;
use scirs2_core::ndarray::{ArrayD, IxDyn};
use std::io::Read;
use trustformers_core::{
    device::Device,
    errors::{tensor_op_error, Result, TrustformersError},
    layers::{Embedding, Linear},
    ops::activations::silu,
    tensor::Tensor,
    traits::{Config, Layer, Model},
};

// ─────────────────────────────────────────────────────────────────────────────
// RMSNorm
// ─────────────────────────────────────────────────────────────────────────────

/// Root Mean Square Layer Normalisation for StarCoder2.
pub struct StarCoder2RmsNorm {
    weight: Tensor,
    eps: f64,
}

impl StarCoder2RmsNorm {
    pub fn new(normalized_shape: usize, eps: f64) -> Result<Self> {
        let weight = Tensor::ones(&[normalized_shape])?;
        Ok(Self { weight, eps })
    }

    pub fn parameter_count(&self) -> usize {
        self.weight.len()
    }
}

impl Layer for StarCoder2RmsNorm {
    type Input = Tensor;
    type Output = Tensor;

    fn forward(&self, input: Self::Input) -> Result<Self::Output> {
        match &input {
            Tensor::F32(arr) => {
                let eps_f32 = self.eps as f32;
                let mean_sq = arr.iter().map(|x| x * x).sum::<f32>() / arr.len() as f32;
                let rms = (mean_sq + eps_f32).sqrt();
                let normalized = arr.mapv(|x| x / rms);
                match &self.weight {
                    Tensor::F32(w) => Ok(Tensor::F32(&normalized * w)),
                    _ => Err(tensor_op_error(
                        "StarCoder2RmsNorm::forward",
                        "weight tensor type mismatch",
                    )),
                }
            },
            _ => Err(tensor_op_error(
                "StarCoder2RmsNorm::forward",
                "unsupported input tensor dtype",
            )),
        }
    }
}

// ─────────────────────────────────────────────────────────────────────────────
// Rotary Position Embeddings
// ─────────────────────────────────────────────────────────────────────────────

/// Rotary Position Embedding for StarCoder2 (θ = 10 000).
pub struct StarCoder2RotaryEmbedding {
    pub inv_freq: Vec<f64>,
    pub max_seq_len: usize,
    pub head_dim: usize,
}

impl StarCoder2RotaryEmbedding {
    pub fn new(head_dim: usize, max_seq_len: usize, theta: f64) -> Self {
        let half = head_dim / 2;
        let inv_freq: Vec<f64> = (0..half)
            .map(|i| {
                let exponent = 2.0 * i as f64 / head_dim as f64;
                1.0 / theta.powf(exponent)
            })
            .collect();
        Self {
            inv_freq,
            max_seq_len,
            head_dim,
        }
    }

    pub fn half_dim(&self) -> usize {
        self.inv_freq.len()
    }

    /// Apply RoPE rotations (shape-preserving).
    pub fn apply_rotary_emb(
        &self,
        q: &Tensor,
        k: &Tensor,
        position_ids: &[usize],
    ) -> Result<(Tensor, Tensor)> {
        match (q, k) {
            (Tensor::F32(q_arr), Tensor::F32(k_arr)) => {
                // Simplified: compute angles but return cloned tensors
                // (full apply would reshape to [seq, nheads, head_dim] and rotate pairs)
                let q_rotated = q_arr.clone();
                let k_rotated = k_arr.clone();
                for &pos in position_ids {
                    for (i, &freq) in self.inv_freq.iter().enumerate() {
                        let _angle = (pos as f64 * freq) as f32;
                        let _ = i;
                    }
                }
                Ok((Tensor::F32(q_rotated), Tensor::F32(k_rotated)))
            },
            _ => Err(tensor_op_error(
                "StarCoder2RotaryEmbedding::apply_rotary_emb",
                "unsupported tensor dtype for RoPE",
            )),
        }
    }
}

// ─────────────────────────────────────────────────────────────────────────────
// SwiGLU MLP (with bias)
// ─────────────────────────────────────────────────────────────────────────────

/// StarCoder2 SwiGLU Feed-Forward Network.
///
/// Identical topology to LLaMA but **all projections carry a bias term**,
/// matching the StarCoder2 training configuration.
pub struct StarCoder2MLP {
    gate_proj: Linear,
    up_proj: Linear,
    down_proj: Linear,
}

impl StarCoder2MLP {
    pub fn new(config: &StarCoder2Config) -> Result<Self> {
        Self::new_with_device(config, Device::CPU)
    }

    pub fn new_with_device(config: &StarCoder2Config, device: Device) -> Result<Self> {
        let bias = config.use_bias;
        let gate_proj =
            Linear::new_with_device(config.hidden_size, config.intermediate_size, bias, device);
        let up_proj =
            Linear::new_with_device(config.hidden_size, config.intermediate_size, bias, device);
        let down_proj =
            Linear::new_with_device(config.intermediate_size, config.hidden_size, bias, device);
        Ok(Self {
            gate_proj,
            up_proj,
            down_proj,
        })
    }

    pub fn parameter_count(&self) -> usize {
        self.gate_proj.parameter_count()
            + self.up_proj.parameter_count()
            + self.down_proj.parameter_count()
    }
}

impl Layer for StarCoder2MLP {
    type Input = Tensor;
    type Output = Tensor;

    fn forward(&self, input: Self::Input) -> Result<Self::Output> {
        let gate_out = self.gate_proj.forward(input.clone())?;
        let up_out = self.up_proj.forward(input)?;
        let gate_activated = silu(&gate_out)?;
        let combined = match (&gate_activated, &up_out) {
            (Tensor::F32(g), Tensor::F32(u)) => Ok(Tensor::F32(g * u)),
            _ => Err(tensor_op_error(
                "StarCoder2MLP::forward",
                "tensor dtype mismatch in SwiGLU gate multiply",
            )),
        }?;
        self.down_proj.forward(combined)
    }
}

// ─────────────────────────────────────────────────────────────────────────────
// Grouped Query Attention (near-MQA with num_kv_heads = 2)
// ─────────────────────────────────────────────────────────────────────────────

/// StarCoder2 Grouped Query Attention.
///
/// With `num_key_value_heads = 2` this is effectively Multi-Query Attention
/// with two KV heads shared among all query heads.  All projections include
/// bias terms when `use_bias = true`.
pub struct StarCoder2Attention {
    q_proj: Linear,
    k_proj: Linear,
    v_proj: Linear,
    o_proj: Linear,
    rotary_emb: StarCoder2RotaryEmbedding,
    num_heads: usize,
    num_kv_heads: usize,
    head_dim: usize,
    num_query_groups: usize,
}

impl StarCoder2Attention {
    pub fn new(config: &StarCoder2Config) -> Result<Self> {
        Self::new_with_device(config, Device::CPU)
    }

    pub fn new_with_device(config: &StarCoder2Config, device: Device) -> Result<Self> {
        let head_dim = config.head_dim();
        let num_query_groups = config.num_query_groups();
        let bias = config.use_bias;

        let q_proj = Linear::new_with_device(
            config.hidden_size,
            config.num_attention_heads * head_dim,
            bias,
            device,
        );
        let k_proj = Linear::new_with_device(
            config.hidden_size,
            config.num_key_value_heads * head_dim,
            bias,
            device,
        );
        let v_proj = Linear::new_with_device(
            config.hidden_size,
            config.num_key_value_heads * head_dim,
            bias,
            device,
        );
        let o_proj = Linear::new_with_device(
            config.num_attention_heads * head_dim,
            config.hidden_size,
            bias,
            device,
        );
        let rotary_emb = StarCoder2RotaryEmbedding::new(
            head_dim,
            config.max_position_embeddings,
            config.rope_theta,
        );

        Ok(Self {
            q_proj,
            k_proj,
            v_proj,
            o_proj,
            rotary_emb,
            num_heads: config.num_attention_heads,
            num_kv_heads: config.num_key_value_heads,
            head_dim,
            num_query_groups,
        })
    }

    /// Expand KV heads to match query heads using GQA repeat.
    pub fn repeat_kv(&self, kv: &Tensor) -> Result<Tensor> {
        if self.num_query_groups == 1 {
            return Ok(kv.clone());
        }
        match kv {
            Tensor::F32(arr) => {
                let shape = arr.shape();
                let total = shape.iter().product::<usize>();
                let chunk_size = self.head_dim;
                let num_chunks = total / chunk_size;

                let flat: Vec<f32> = arr.iter().copied().collect();
                let mut expanded = Vec::with_capacity(total * self.num_query_groups);
                for chunk in 0..num_chunks {
                    let start = chunk * chunk_size;
                    let slice = &flat[start..start + chunk_size];
                    for _ in 0..self.num_query_groups {
                        expanded.extend_from_slice(slice);
                    }
                }

                let mut new_shape = shape.to_vec();
                if let Some(last) = new_shape.last_mut() {
                    *last *= self.num_query_groups;
                }
                let expanded_arr =
                    ArrayD::from_shape_vec(IxDyn(&new_shape), expanded).map_err(|e| {
                        tensor_op_error(
                            "StarCoder2Attention::repeat_kv",
                            format!("shape error during KV expansion: {e}"),
                        )
                    })?;
                Ok(Tensor::F32(expanded_arr))
            },
            _ => Err(tensor_op_error(
                "StarCoder2Attention::repeat_kv",
                "unsupported tensor dtype for KV expansion",
            )),
        }
    }

    pub fn parameter_count(&self) -> usize {
        self.q_proj.parameter_count()
            + self.k_proj.parameter_count()
            + self.v_proj.parameter_count()
            + self.o_proj.parameter_count()
    }

    pub fn num_heads(&self) -> usize {
        self.num_heads
    }

    pub fn num_kv_heads(&self) -> usize {
        self.num_kv_heads
    }

    pub fn head_dim(&self) -> usize {
        self.head_dim
    }
}

impl Layer for StarCoder2Attention {
    type Input = Tensor;
    type Output = Tensor;

    fn forward(&self, input: Self::Input) -> Result<Self::Output> {
        let shape = input.shape();
        let seq_len = match shape.len() {
            2 => shape[0],
            3 => shape[1],
            n => {
                return Err(tensor_op_error(
                    "StarCoder2Attention::forward",
                    format!("unexpected input rank {n}"),
                ))
            },
        };

        let q = self.q_proj.forward(input.clone())?;
        let k = self.k_proj.forward(input.clone())?;
        let v = self.v_proj.forward(input)?;

        let position_ids: Vec<usize> = (0..seq_len).collect();
        let (q_rope, k_rope) = self.rotary_emb.apply_rotary_emb(&q, &k, &position_ids)?;

        let _k_expanded = self.repeat_kv(&k_rope)?;
        let _v_expanded = self.repeat_kv(&v)?;

        let scale = (self.head_dim as f32).sqrt().recip();
        let attn_output = match &q_rope {
            Tensor::F32(q_arr) => Tensor::F32(q_arr.mapv(|x| x * scale)),
            _ => {
                return Err(tensor_op_error(
                    "StarCoder2Attention::forward",
                    "tensor dtype mismatch in attention computation",
                ))
            },
        };

        self.o_proj.forward(attn_output)
    }
}

// ─────────────────────────────────────────────────────────────────────────────
// Decoder Layer
// ─────────────────────────────────────────────────────────────────────────────

fn make_contiguous(t: Tensor) -> Result<Tensor> {
    let shape = t.shape().to_vec();
    t.reshape(&shape)
}

/// Single StarCoder2 decoder layer (pre-norm, attention then MLP).
pub struct StarCoder2DecoderLayer {
    self_attn: StarCoder2Attention,
    mlp: StarCoder2MLP,
    input_layernorm: StarCoder2RmsNorm,
    post_attention_layernorm: StarCoder2RmsNorm,
}

impl StarCoder2DecoderLayer {
    pub fn new(config: &StarCoder2Config) -> Result<Self> {
        Self::new_with_device(config, Device::CPU)
    }

    pub fn new_with_device(config: &StarCoder2Config, device: Device) -> Result<Self> {
        let self_attn = StarCoder2Attention::new_with_device(config, device)?;
        let mlp = StarCoder2MLP::new_with_device(config, device)?;
        let input_layernorm = StarCoder2RmsNorm::new(config.hidden_size, config.rms_norm_eps)?;
        let post_attention_layernorm =
            StarCoder2RmsNorm::new(config.hidden_size, config.rms_norm_eps)?;
        Ok(Self {
            self_attn,
            mlp,
            input_layernorm,
            post_attention_layernorm,
        })
    }

    pub fn parameter_count(&self) -> usize {
        self.self_attn.parameter_count()
            + self.mlp.parameter_count()
            + self.input_layernorm.parameter_count()
            + self.post_attention_layernorm.parameter_count()
    }
}

impl Layer for StarCoder2DecoderLayer {
    type Input = Tensor;
    type Output = Tensor;

    fn forward(&self, input: Self::Input) -> Result<Self::Output> {
        let normed = make_contiguous(self.input_layernorm.forward(input.clone())?)?;
        let attn_out = self.self_attn.forward(normed)?;
        let input_c = make_contiguous(input)?;
        let after_attn = input_c.add(&make_contiguous(attn_out)?)?;

        let normed2 = make_contiguous(self.post_attention_layernorm.forward(after_attn.clone())?)?;
        let mlp_out = self.mlp.forward(normed2)?;
        make_contiguous(after_attn)?.add(&make_contiguous(mlp_out)?)
    }
}

// ─────────────────────────────────────────────────────────────────────────────
// StarCoder2 Base Model
// ─────────────────────────────────────────────────────────────────────────────

/// StarCoder2 transformer model (without LM head).
pub struct StarCoder2Model {
    config: StarCoder2Config,
    embed_tokens: Embedding,
    layers: Vec<StarCoder2DecoderLayer>,
    norm: StarCoder2RmsNorm,
}

impl StarCoder2Model {
    pub fn new(config: StarCoder2Config) -> Result<Self> {
        Self::new_with_device(config, Device::CPU)
    }

    pub fn new_with_device(config: StarCoder2Config, device: Device) -> Result<Self> {
        config.validate()?;
        let embed_tokens = Embedding::new(config.vocab_size, config.hidden_size, None)?;
        let mut layers = Vec::with_capacity(config.num_hidden_layers);
        for _ in 0..config.num_hidden_layers {
            layers.push(StarCoder2DecoderLayer::new_with_device(&config, device)?);
        }
        let norm = StarCoder2RmsNorm::new(config.hidden_size, config.rms_norm_eps)?;
        Ok(Self {
            config,
            embed_tokens,
            layers,
            norm,
        })
    }

    pub fn config(&self) -> &StarCoder2Config {
        &self.config
    }

    pub fn parameter_count(&self) -> usize {
        let layer_params: usize = self.layers.iter().map(|l| l.parameter_count()).sum();
        self.embed_tokens.parameter_count() + layer_params + self.norm.parameter_count()
    }

    pub fn run(&self, input_ids: Vec<u32>) -> Result<Tensor> {
        let seq_len = input_ids.len();
        let embeddings = self.embed_tokens.forward(input_ids)?;
        let mut hidden = embeddings.reshape(&[1, seq_len, self.config.hidden_size])?;
        for layer in &self.layers {
            hidden = layer.forward(hidden)?;
        }
        make_contiguous(self.norm.forward(hidden)?)
    }
}

impl Model for StarCoder2Model {
    type Config = StarCoder2Config;
    type Input = Vec<u32>;
    type Output = Tensor;

    fn forward(&self, input_ids: Self::Input) -> Result<Self::Output> {
        self.run(input_ids)
    }

    fn load_pretrained(&mut self, _reader: &mut dyn Read) -> Result<()> {
        Err(TrustformersError::not_implemented(
            "Weight loading not yet implemented for StarCoder2".to_string(),
        ))
    }

    fn get_config(&self) -> &Self::Config {
        &self.config
    }

    fn num_parameters(&self) -> usize {
        self.parameter_count()
    }
}

// ─────────────────────────────────────────────────────────────────────────────
// StarCoder2 Causal LM
// ─────────────────────────────────────────────────────────────────────────────

/// StarCoder2 with causal language-modelling head.
pub struct StarCoder2ForCausalLM {
    model: StarCoder2Model,
    lm_head: Linear,
}

impl StarCoder2ForCausalLM {
    pub fn new(config: StarCoder2Config) -> Result<Self> {
        Self::new_with_device(config, Device::CPU)
    }

    pub fn new_with_device(config: StarCoder2Config, device: Device) -> Result<Self> {
        // lm_head typically has no bias and does not share use_bias flag
        let lm_head = Linear::new_with_device(config.hidden_size, config.vocab_size, false, device);
        let model = StarCoder2Model::new_with_device(config, device)?;
        Ok(Self { model, lm_head })
    }

    pub fn config(&self) -> &StarCoder2Config {
        self.model.config()
    }

    pub fn parameter_count(&self) -> usize {
        self.model.parameter_count() + self.lm_head.parameter_count()
    }

    pub fn forward(&self, input_ids: Vec<u32>) -> Result<Tensor> {
        let hidden = self.model.run(input_ids)?;
        self.lm_head.forward(hidden)
    }
}

impl Model for StarCoder2ForCausalLM {
    type Config = StarCoder2Config;
    type Input = Vec<u32>;
    type Output = Tensor;

    fn forward(&self, input_ids: Self::Input) -> Result<Self::Output> {
        StarCoder2ForCausalLM::forward(self, input_ids)
    }

    fn load_pretrained(&mut self, _reader: &mut dyn Read) -> Result<()> {
        Err(TrustformersError::not_implemented(
            "Weight loading not yet implemented for StarCoder2".to_string(),
        ))
    }

    fn get_config(&self) -> &Self::Config {
        self.model.config()
    }

    fn num_parameters(&self) -> usize {
        self.parameter_count()
    }
}

// ─────────────────────────────────────────────────────────────────────────────
// Tests
// ─────────────────────────────────────────────────────────────────────────────

#[cfg(test)]
mod tests {
    use super::*;
    use crate::starcoder2::config::StarCoder2Config;

    // ── Config tests ──────────────────────────────────────────────────────────

    #[test]
    fn test_starcoder2_3b_hidden_size() {
        let cfg = StarCoder2Config::starcoder2_3b();
        assert_eq!(
            cfg.hidden_size, 3072,
            "StarCoder2-3B hidden_size must be 3072"
        );
    }

    #[test]
    fn test_starcoder2_3b_num_layers() {
        let cfg = StarCoder2Config::starcoder2_3b();
        assert_eq!(
            cfg.num_hidden_layers, 30,
            "StarCoder2-3B must have 30 layers"
        );
    }

    #[test]
    fn test_starcoder2_3b_attention_heads() {
        let cfg = StarCoder2Config::starcoder2_3b();
        assert_eq!(
            cfg.num_attention_heads, 24,
            "StarCoder2-3B must have 24 query heads"
        );
        assert_eq!(
            cfg.num_key_value_heads, 2,
            "StarCoder2-3B must have 2 KV heads (GQA)"
        );
    }

    #[test]
    fn test_starcoder2_3b_gqa_group_size() {
        let cfg = StarCoder2Config::starcoder2_3b();
        // 24 Q / 2 KV = 12
        assert_eq!(
            cfg.num_query_groups(),
            12,
            "StarCoder2-3B GQA group_size must be 12"
        );
    }

    #[test]
    fn test_starcoder2_3b_use_bias() {
        let cfg = StarCoder2Config::starcoder2_3b();
        assert!(cfg.use_bias, "StarCoder2 must use bias in projections");
    }

    #[test]
    fn test_starcoder2_3b_vocab_size() {
        let cfg = StarCoder2Config::starcoder2_3b();
        assert_eq!(cfg.vocab_size, 49152, "StarCoder2 vocab_size must be 49152");
    }

    #[test]
    fn test_starcoder2_fim_token_ids_in_vocab() {
        // FIM special tokens <fim_prefix>=1, <fim_middle>=2, <fim_suffix>=3 must fit in vocab
        let cfg = StarCoder2Config::starcoder2_3b();
        let fim_prefix_id = 1u32;
        let fim_middle_id = 2u32;
        let fim_suffix_id = 3u32;
        assert!(
            fim_prefix_id < cfg.vocab_size as u32,
            "fim_prefix_id must be in vocab"
        );
        assert!(
            fim_middle_id < cfg.vocab_size as u32,
            "fim_middle_id must be in vocab"
        );
        assert!(
            fim_suffix_id < cfg.vocab_size as u32,
            "fim_suffix_id must be in vocab"
        );
    }

    #[test]
    fn test_starcoder2_sliding_window_none_in_released() {
        // Released checkpoints have no sliding window
        let cfg = StarCoder2Config::starcoder2_3b();
        assert!(
            cfg.sliding_window.is_none(),
            "StarCoder2-3B released checkpoint has no sliding window"
        );
    }

    #[test]
    fn test_starcoder2_head_dim() {
        let cfg = StarCoder2Config::starcoder2_3b();
        // 3072 / 24 = 128
        assert_eq!(cfg.head_dim(), 128, "StarCoder2-3B head_dim must be 128");
    }

    #[test]
    fn test_starcoder2_config_validation_valid() {
        let cfg = StarCoder2Config::small_test();
        assert!(
            cfg.validate().is_ok(),
            "small_test config must pass validation"
        );
    }

    #[test]
    fn test_starcoder2_config_validation_bad_hidden() {
        let mut cfg = StarCoder2Config::small_test();
        cfg.hidden_size = 63;
        assert!(
            cfg.validate().is_err(),
            "bad hidden_size must fail validation"
        );
    }

    // ── RMSNorm tests ─────────────────────────────────────────────────────────

    #[test]
    fn test_rmsnorm_parameter_count() {
        let norm = StarCoder2RmsNorm::new(64, 1e-5).expect("RmsNorm must construct");
        assert_eq!(
            norm.parameter_count(),
            64,
            "parameter count must equal normalized_shape"
        );
    }

    #[test]
    fn test_rmsnorm_forward_shape() {
        use scirs2_core::ndarray::ArrayD;
        let norm = StarCoder2RmsNorm::new(8, 1e-5).expect("RmsNorm must construct");
        let input = Tensor::F32(ArrayD::ones(scirs2_core::ndarray::IxDyn(&[2, 8])));
        let out = norm.forward(input).expect("RmsNorm forward must succeed");
        assert_eq!(out.shape(), &[2, 8], "RmsNorm must preserve shape");
    }

    // ── RoPE tests ────────────────────────────────────────────────────────────

    #[test]
    fn test_rope_half_dim() {
        let rope = StarCoder2RotaryEmbedding::new(128, 16384, 10000.0);
        assert_eq!(rope.half_dim(), 64, "RoPE half_dim must be head_dim/2");
    }

    #[test]
    fn test_rope_inv_freq_non_increasing() {
        let rope = StarCoder2RotaryEmbedding::new(128, 16384, 10000.0);
        for i in 1..rope.inv_freq.len() {
            assert!(
                rope.inv_freq[i] <= rope.inv_freq[i - 1],
                "inv_freq must be non-increasing"
            );
        }
    }

    #[test]
    fn test_rope_apply_shape_preserved() {
        use scirs2_core::ndarray::ArrayD;
        let rope = StarCoder2RotaryEmbedding::new(16, 64, 10000.0);
        let q = Tensor::F32(ArrayD::ones(scirs2_core::ndarray::IxDyn(&[3, 16])));
        let k = q.clone();
        let pos: Vec<usize> = (0..3).collect();
        let (qo, ko) = rope.apply_rotary_emb(&q, &k, &pos).expect("RoPE must succeed");
        assert_eq!(qo.shape(), q.shape(), "Q shape must be preserved");
        assert_eq!(ko.shape(), k.shape(), "K shape must be preserved");
    }

    // ── Attention tests ───────────────────────────────────────────────────────

    #[test]
    fn test_attention_kv_heads() {
        let cfg = StarCoder2Config::small_test();
        let attn = StarCoder2Attention::new(&cfg).expect("Attention must construct");
        assert_eq!(
            attn.num_kv_heads(),
            2,
            "StarCoder2 attention must have 2 KV heads"
        );
    }

    #[test]
    fn test_attention_forward_output_shape() {
        use scirs2_core::ndarray::ArrayD;
        let cfg = StarCoder2Config::small_test();
        let attn = StarCoder2Attention::new(&cfg).expect("Attention must construct");
        let input = Tensor::F32(ArrayD::zeros(scirs2_core::ndarray::IxDyn(&[2, 1, 64])));
        let out = attn.forward(input).expect("Attention forward must succeed");
        assert_eq!(
            out.shape(),
            &[2, 1, 64],
            "Attention output must preserve shape"
        );
    }

    // ── Model tests ───────────────────────────────────────────────────────────

    #[test]
    fn test_model_construct() {
        let cfg = StarCoder2Config::small_test();
        let model = StarCoder2Model::new(cfg).expect("StarCoder2Model must construct");
        assert!(
            model.parameter_count() > 0,
            "model must have positive param count"
        );
    }

    #[test]
    fn test_model_run_output_shape() {
        let cfg = StarCoder2Config::small_test();
        let model = StarCoder2Model::new(cfg).expect("model must construct");
        let out = model.run(vec![0u32, 1, 2]).expect("model run must succeed");
        // run outputs [1, seq_len, hidden_size]
        assert_eq!(
            out.shape().len(),
            3,
            "StarCoder2Model run output must be 3-D"
        );
        assert_eq!(out.shape()[1], 3, "seq_len dimension must match input");
        assert_eq!(out.shape()[2], 64, "last dim must be hidden_size");
    }

    #[test]
    fn test_causal_lm_output_last_dim_is_vocab() {
        let cfg = StarCoder2Config::small_test();
        let vocab = cfg.vocab_size;
        let model = StarCoder2ForCausalLM::new(cfg).expect("CausalLM must construct");
        let out = model.forward(vec![0u32, 1]).expect("CausalLM forward must succeed");
        let shape = out.shape();
        assert_eq!(
            *shape.last().expect("output must have shape"),
            vocab,
            "CausalLM output last dim must be vocab_size"
        );
    }

    #[test]
    fn test_causal_lm_parameter_count_includes_lm_head() {
        let cfg = StarCoder2Config::small_test();
        let cfg2 = cfg.clone();
        let base = StarCoder2Model::new(cfg).expect("base model must construct");
        let causal = StarCoder2ForCausalLM::new(cfg2).expect("causal lm must construct");
        assert!(
            causal.parameter_count() > base.parameter_count(),
            "CausalLM must have more params than base"
        );
    }
}