fuel-transformers 0.19.1

// SPDX-License-Identifier: MIT OR Apache-2.0
//! StableLM decoder ported to the lazy-graph API.
//!
//! Phase D LLM port. StableLM (Stability AI) is structurally
//! `Mistral + LayerNorm + partial-rotary`:
//! - **LayerNorm** with bias on the pre-attention + pre-FFN paths
//!   (not RmsNorm).
//! - **Partial rotary** — apply RoPE to only the first
//!   `(head_dim * partial_rotary_factor)` dimensions of each head.
//!   StableLM-1 uses 0.25 (25% of head_dim gets RoPE); StableLM-2
//!   uses 1.0 (full rotary, equivalent to LLaMA's path).
//! - **Optional Q/K/V biases** via `cfg.use_qkv_bias` (StableLM-2
//!   uses biases on Q/K/V only, not O).
//! - GQA + SwiGLU + bias-free O.
//!
//! The partial-rotary helper applies RoPE to a head-dim prefix and
//! passes the rest through unchanged. v1 builds the prefix's
//! cos/sin tables from `rope_dim` (not `head_dim`); the suffix
//! never sees RoPE.

use fuel_ir::Shape;
use fuel_ir::error::Result;
use fuel_tensor::Device;
use fuel_tensor::{lazy::Tensor, lazy::WeightStorage};
use std::sync::Arc;

#[derive(Debug, Clone, PartialEq)]
pub struct StableLmConfig {
    pub vocab_size: usize,
    pub hidden_size: usize,
    pub intermediate_size: usize,
    pub num_hidden_layers: usize,
    pub num_attention_heads: usize,
    pub num_key_value_heads: usize,
    pub head_dim: usize,
    pub layer_norm_eps: f64,
    pub rope_theta: f64,
    pub max_position_embeddings: usize,
    /// Fraction of `head_dim` to which RoPE is applied. Stability's
    /// StableLM-1 ships 0.25; StableLM-2 uses 1.0.
    pub partial_rotary_factor: f64,
    /// StableLM-2 has Q/K/V biases; StableLM-1 doesn't.
    pub use_qkv_bias: bool,
}

impl StableLmConfig {
    pub fn rope_dim(&self) -> usize {
        // Round down to an even number so the half-split RoPE layout
        // (cos/sin pair-of-dims) divides cleanly. `head_dim` is
        // already even, so flooring partial_rotary_factor * head_dim
        // and rounding down to even gives a valid prefix.
        let rd = (self.head_dim as f64 * self.partial_rotary_factor) as usize;
        (rd / 2) * 2
    }
}

// ROADMAP item 8 (II): config-from-path on the #57 template. StableLM uses LayerNorm
// (layer_norm_eps, not rms) and partial rotary + optional qkv bias — the raw carries
// those, and resolve routes num_key_value_heads + head_dim through the shared rules.
#[derive(Debug, Clone, serde::Deserialize)]
struct StableLmConfigRaw {
    vocab_size: usize,
    hidden_size: usize,
    intermediate_size: usize,
    num_hidden_layers: usize,
    num_attention_heads: usize,
    #[serde(default)]
    num_key_value_heads: Option<usize>,
    #[serde(default)]
    head_dim: Option<usize>,
    #[serde(default = "default_stablelm_layer_norm_eps")]
    layer_norm_eps: f64,
    #[serde(default = "default_stablelm_rope_theta")]
    rope_theta: f64,
    max_position_embeddings: usize,
    #[serde(default = "default_stablelm_partial_rotary_factor")]
    partial_rotary_factor: f64,
    #[serde(default)]
    use_qkv_bias: bool,
}

fn default_stablelm_layer_norm_eps() -> f64 {
    1e-5
}
fn default_stablelm_rope_theta() -> f64 {
    10_000.0
}
fn default_stablelm_partial_rotary_factor() -> f64 {
    0.25
}

impl StableLmConfigRaw {
    fn from_json_str(json: &str) -> Result<Self> {
        serde_json::from_str(json)
            .map_err(|e| fuel_ir::error::Error::Msg(format!("parsing StableLM config.json: {e}")))
    }

    fn resolve(self) -> Result<StableLmConfig> {
        Ok(StableLmConfig {
            vocab_size: self.vocab_size,
            hidden_size: self.hidden_size,
            intermediate_size: self.intermediate_size,
            num_hidden_layers: self.num_hidden_layers,
            num_attention_heads: self.num_attention_heads,
            num_key_value_heads: fuel_loaders::hf_config::num_key_value_heads(
                self.num_key_value_heads,
                self.num_attention_heads,
            )?,
            head_dim: fuel_loaders::hf_config::head_dim(
                self.head_dim,
                self.hidden_size,
                self.num_attention_heads,
            )?,
            layer_norm_eps: self.layer_norm_eps,
            rope_theta: self.rope_theta,
            max_position_embeddings: self.max_position_embeddings,
            partial_rotary_factor: self.partial_rotary_factor,
            use_qkv_bias: self.use_qkv_bias,
        })
    }
}

impl StableLmConfig {
    /// Parse a HuggingFace `config.json` string into a [`StableLmConfig`].
    ///
    /// ROADMAP item 8 (II): reads the artifact — see the born-red
    /// `stablelm_config_from_hf_json_parses_the_artifact_not_a_preset`.
    pub fn from_hf_json_str(json: &str) -> Result<Self> {
        StableLmConfigRaw::from_json_str(json)?.resolve()
    }
}

#[derive(Debug, Clone)]
pub struct StableLmWeights {
    pub token_embedding: Arc<[f32]>,
    pub layers: Vec<StableLmLayerWeights>,
    pub final_ln_gain: Arc<[f32]>,
    pub final_ln_bias: Arc<[f32]>,
    pub output: WeightStorage,
}

/// Per-layer weights. LayerNorm has both gain + bias on both norm
/// positions; Q/K/V optionally have biases; O has none; MLP is
/// SwiGLU with no biases.
#[derive(Debug, Clone)]
pub struct StableLmLayerWeights {
    pub input_ln_gain: Arc<[f32]>,
    pub input_ln_bias: Arc<[f32]>,
    pub post_attn_ln_gain: Arc<[f32]>,
    pub post_attn_ln_bias: Arc<[f32]>,
    pub attn_q: WeightStorage,
    pub attn_q_bias: Option<Arc<[f32]>>,
    pub attn_k: WeightStorage,
    pub attn_k_bias: Option<Arc<[f32]>>,
    pub attn_v: WeightStorage,
    pub attn_v_bias: Option<Arc<[f32]>>,
    pub attn_o: WeightStorage,
    pub ffn_gate: WeightStorage,
    pub ffn_up: WeightStorage,
    pub ffn_down: WeightStorage,
}

#[derive(Debug, Clone)]
pub struct StableLmModel {
    pub config: StableLmConfig,
    pub weights: StableLmWeights,
}

impl StableLmModel {
    pub fn forward(&self, tokens: &[u32], start_pos: usize) -> Result<Tensor> {
        let h_norm = self.run_backbone(tokens, start_pos)?;
        self.apply_lm_head(&h_norm)
    }

    /// Run the decoder forward up to the final LayerNorm and
    /// return per-token hidden states `(1, seq, hidden_size)`.
    /// Skips the `lm_head` projection. Mirrors the
    /// `forward_hidden` pattern across the LLM family.
    pub fn forward_hidden(&self, tokens: &[u32], start_pos: usize) -> Result<Tensor> {
        self.run_backbone(tokens, start_pos)
    }

    /// Multimodal entry point. Skips token embedding; runs the decoder
    /// over pre-embedded inputs. StableLM does NOT scale embeddings.
    pub fn forward_embeds(&self, embeds: &Tensor, start_pos: usize) -> Result<Tensor> {
        let h_norm = self.run_backbone_embeds(embeds, start_pos)?;
        self.apply_lm_head(&h_norm)
    }

    /// Hidden-state variant of [`Self::forward_embeds`].
    pub fn forward_hidden_embeds(&self, embeds: &Tensor, start_pos: usize) -> Result<Tensor> {
        self.run_backbone_embeds(embeds, start_pos)
    }

    /// Build per-token embeddings without running the decoder.
    pub fn embed_tokens_anchored(&self, anchor: &Tensor, tokens: &[u32]) -> Result<Tensor> {
        let cfg = &self.config;
        anchor.embed_tokens_anchored(
            self.weights.token_embedding.clone(),
            cfg.vocab_size,
            cfg.hidden_size,
            tokens,
        )
    }

    fn apply_lm_head(&self, h_norm: &Tensor) -> Result<Tensor> {
        let cfg = &self.config;
        self.weights
            .output
            .apply_linear(h_norm, cfg.hidden_size, cfg.vocab_size)
    }

    fn run_backbone(&self, tokens: &[u32], start_pos: usize) -> Result<Tensor> {
        let cfg = &self.config;
        let weights = &self.weights;
        let seq = tokens.len();
        assert!(seq > 0);

        let h = Tensor::embed_tokens(
            weights.token_embedding.clone(),
            cfg.vocab_size,
            cfg.hidden_size,
            tokens,
            &Device::cpu(),
        )?;
        self.run_backbone_embeds(&h, start_pos)
    }

    fn run_backbone_embeds(&self, embeds: &Tensor, start_pos: usize) -> Result<Tensor> {
        let cfg = &self.config;
        let weights = &self.weights;
        let dims = embeds.shape();
        let dims = dims.dims();
        if dims.len() != 3 || dims[0] != 1 || dims[2] != cfg.hidden_size {
            return Err(fuel_ir::error::Error::Msg(format!(
                "StableLmModel::forward_embeds: expected embeds shape (1, seq, hidden_size={}), got {:?}",
                cfg.hidden_size, dims,
            )).bt());
        }
        let seq = dims[1];
        if seq == 0 {
            return Err(fuel_ir::error::Error::Msg(
                "StableLmModel::forward_embeds: seq must be > 0".into(),
            )
            .bt());
        }
        if cfg.num_attention_heads * cfg.head_dim != cfg.hidden_size {
            return Err(fuel_ir::error::Error::Msg(
                "StableLmConfig: num_attention_heads * head_dim must equal hidden_size".into(),
            )
            .bt());
        }
        let mut h = embeds.clone();

        let rope_dim = cfg.rope_dim();
        let (rope_cos, rope_sin) = h.rope_tables_const(cfg.rope_theta, start_pos, seq, rope_dim);

        for layer in &weights.layers {
            h = self.apply_layer(&h, layer, &rope_cos, &rope_sin)?;
        }
        h.layer_norm_affine(
            std::sync::Arc::clone(&weights.final_ln_gain),
            std::sync::Arc::clone(&weights.final_ln_bias),
            cfg.layer_norm_eps,
        )
    }

    fn apply_layer(
        &self,
        x: &Tensor,
        layer: &StableLmLayerWeights,
        rope_cos: &Tensor,
        rope_sin: &Tensor,
    ) -> Result<Tensor> {
        let cfg = &self.config;
        let head_dim = cfg.head_dim;
        let rope_dim = cfg.rope_dim();
        let x_shape = x.shape();
        let dims = x_shape.dims();
        let batch = dims[0];
        let seq = dims[1];
        let kv_dim = cfg.num_key_value_heads * head_dim;

        let x_norm = x.layer_norm_affine(
            std::sync::Arc::clone(&layer.input_ln_gain),
            std::sync::Arc::clone(&layer.input_ln_bias),
            cfg.layer_norm_eps,
        )?;

        let q = layer
            .attn_q
            .apply_linear(&x_norm, cfg.hidden_size, cfg.hidden_size)?
            .add_optional_trailing_bias(layer.attn_q_bias.as_ref())?;
        let k = layer
            .attn_k
            .apply_linear(&x_norm, cfg.hidden_size, kv_dim)?
            .add_optional_trailing_bias(layer.attn_k_bias.as_ref())?;
        let v = layer
            .attn_v
            .apply_linear(&x_norm, cfg.hidden_size, kv_dim)?
            .add_optional_trailing_bias(layer.attn_v_bias.as_ref())?;

        let _ = (batch, seq);
        let q = q.split_heads(cfg.num_attention_heads, head_dim)?;
        let k = k.split_heads(cfg.num_key_value_heads, head_dim)?;
        let v = v.split_heads(cfg.num_key_value_heads, head_dim)?;

        // Partial rotary: slice [..., :rope_dim] → apply RoPE → concat
        // with [..., rope_dim:].
        let q_r = q.rope_partial(rope_cos, rope_sin, rope_dim)?;
        let k_r = k.rope_partial(rope_cos, rope_sin, rope_dim)?;

        let n_rep = cfg.num_attention_heads / cfg.num_key_value_heads;
        let k_full = k_r.repeat_interleave(1_usize, n_rep)?;
        let v_full = v.repeat_interleave(1_usize, n_rep)?;

        let k_t = k_full.transpose()?;
        let scale = 1.0_f64 / (head_dim as f64).sqrt();
        let scores = q_r.matmul(&k_t)?;
        let scores_scaled = scores.mul_scalar(scale);
        let mut mask_data = vec![0.0_f32; seq * seq];
        for i in 0..seq {
            for j in (i + 1)..seq {
                mask_data[i * seq + j] = f32::NEG_INFINITY;
            }
        }
        let mask = x.const_f32_like(mask_data, Shape::from_dims(&[1, 1, seq, seq]))?;
        let scores_masked = scores_scaled.broadcast_add(&mask)?;
        let attn = scores_masked.softmax_last_dim()?;
        let attn_v = attn.matmul(&v_full)?;

        let merged = attn_v.merge_heads()?;
        let attn_out = layer
            .attn_o
            .apply_linear(&merged, cfg.hidden_size, cfg.hidden_size)?;

        let h1 = x.add(&attn_out)?;
        let h1_norm = h1.layer_norm_affine(
            std::sync::Arc::clone(&layer.post_attn_ln_gain),
            std::sync::Arc::clone(&layer.post_attn_ln_bias),
            cfg.layer_norm_eps,
        )?;
        let gate = layer
            .ffn_gate
            .apply_linear(&h1_norm, cfg.hidden_size, cfg.intermediate_size)?;
        let up = layer
            .ffn_up
            .apply_linear(&h1_norm, cfg.hidden_size, cfg.intermediate_size)?;
        let swiglu = gate.silu().mul(&up)?;
        let ffn_out =
            layer
                .ffn_down
                .apply_linear(&swiglu, cfg.intermediate_size, cfg.hidden_size)?;
        h1.add(&ffn_out)
    }
}

// ---- HuggingFace safetensors loader ----------------------------------------

impl StableLmWeights {
    /// Load StableLM weights (e.g. `stabilityai/stablelm-2-1_6b`).
    pub fn load_from_mmapped(
        st: &fuel_loaders::safetensors::MmapedSafetensors,
        cfg: &StableLmConfig,
    ) -> Result<Self> {
        use fuel_tensor::{lazy::load_tensor_as_f32, lazy::load_transposed_matrix_preserve_dtype};
        let h = cfg.hidden_size;
        let q_dim = cfg.num_attention_heads * cfg.head_dim;
        let kv_dim = cfg.num_key_value_heads * cfg.head_dim;
        let inter = cfg.intermediate_size;

        let token_embedding = Arc::from(load_tensor_as_f32(st, "model.embed_tokens.weight")?);
        let mut layers: Vec<StableLmLayerWeights> = Vec::with_capacity(cfg.num_hidden_layers);
        for i in 0..cfg.num_hidden_layers {
            let p = format!("model.layers.{i}");
            let input_ln_gain = Arc::from(load_tensor_as_f32(
                st,
                &format!("{p}.input_layernorm.weight"),
            )?);
            let input_ln_bias = Arc::from(load_tensor_as_f32(
                st,
                &format!("{p}.input_layernorm.bias"),
            )?);
            let post_attn_ln_gain = Arc::from(load_tensor_as_f32(
                st,
                &format!("{p}.post_attention_layernorm.weight"),
            )?);
            let post_attn_ln_bias = Arc::from(load_tensor_as_f32(
                st,
                &format!("{p}.post_attention_layernorm.bias"),
            )?);
            let attn_q = load_transposed_matrix_preserve_dtype(
                st,
                &format!("{p}.self_attn.q_proj.weight"),
                q_dim,
                h,
            )?;
            let attn_k = load_transposed_matrix_preserve_dtype(
                st,
                &format!("{p}.self_attn.k_proj.weight"),
                kv_dim,
                h,
            )?;
            let attn_v = load_transposed_matrix_preserve_dtype(
                st,
                &format!("{p}.self_attn.v_proj.weight"),
                kv_dim,
                h,
            )?;
            let attn_o = load_transposed_matrix_preserve_dtype(
                st,
                &format!("{p}.self_attn.o_proj.weight"),
                h,
                q_dim,
            )?;
            let (attn_q_bias, attn_k_bias, attn_v_bias) = if cfg.use_qkv_bias {
                (
                    Some(Arc::from(load_tensor_as_f32(
                        st,
                        &format!("{p}.self_attn.q_proj.bias"),
                    )?)),
                    Some(Arc::from(load_tensor_as_f32(
                        st,
                        &format!("{p}.self_attn.k_proj.bias"),
                    )?)),
                    Some(Arc::from(load_tensor_as_f32(
                        st,
                        &format!("{p}.self_attn.v_proj.bias"),
                    )?)),
                )
            } else {
                (None, None, None)
            };
            let ffn_gate = load_transposed_matrix_preserve_dtype(
                st,
                &format!("{p}.mlp.gate_proj.weight"),
                inter,
                h,
            )?;
            let ffn_up = load_transposed_matrix_preserve_dtype(
                st,
                &format!("{p}.mlp.up_proj.weight"),
                inter,
                h,
            )?;
            let ffn_down = load_transposed_matrix_preserve_dtype(
                st,
                &format!("{p}.mlp.down_proj.weight"),
                h,
                inter,
            )?;
            layers.push(StableLmLayerWeights {
                input_ln_gain,
                input_ln_bias,
                post_attn_ln_gain,
                post_attn_ln_bias,
                attn_q,
                attn_q_bias,
                attn_k,
                attn_k_bias,
                attn_v,
                attn_v_bias,
                attn_o,
                ffn_gate,
                ffn_up,
                ffn_down,
            });
        }
        let final_ln_gain = Arc::from(load_tensor_as_f32(st, "model.norm.weight")?);
        let final_ln_bias = Arc::from(load_tensor_as_f32(st, "model.norm.bias")?);
        let output =
            load_transposed_matrix_preserve_dtype(st, "lm_head.weight", cfg.vocab_size, h)?;
        Ok(Self {
            token_embedding,
            layers,
            final_ln_gain,
            final_ln_bias,
            output,
        })
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    fn tiny_weights(cfg: &StableLmConfig) -> StableLmWeights {
        let mut s: u32 = 88888;
        let next = || -> f32 {
            s = s.wrapping_mul(1103515245).wrapping_add(12345);
            ((s >> 16) as u16 as f32 / 65535.0 - 0.5) * 0.05
        };
        let vec_of = |n: usize, next: &mut dyn FnMut() -> f32| -> Arc<[f32]> {
            Arc::from((0..n).map(|_| next()).collect::<Vec<_>>())
        };
        let h = cfg.hidden_size;
        let i = cfg.intermediate_size;
        let kv = cfg.num_key_value_heads * cfg.head_dim;
        let mut nb: Box<dyn FnMut() -> f32> = Box::new(next);
        let token_embedding = vec_of(cfg.vocab_size * h, &mut *nb);
        let layers: Vec<StableLmLayerWeights> = (0..cfg.num_hidden_layers)
            .map(|_| StableLmLayerWeights {
                input_ln_gain: Arc::from(vec![1.0_f32; h]),
                input_ln_bias: Arc::from(vec![0.0_f32; h]),
                post_attn_ln_gain: Arc::from(vec![1.0_f32; h]),
                post_attn_ln_bias: Arc::from(vec![0.0_f32; h]),
                attn_q: WeightStorage::F32(vec_of(h * h, &mut *nb)),
                attn_q_bias: if cfg.use_qkv_bias {
                    Some(vec_of(h, &mut *nb))
                } else {
                    None
                },
                attn_k: WeightStorage::F32(vec_of(h * kv, &mut *nb)),
                attn_k_bias: if cfg.use_qkv_bias {
                    Some(vec_of(kv, &mut *nb))
                } else {
                    None
                },
                attn_v: WeightStorage::F32(vec_of(h * kv, &mut *nb)),
                attn_v_bias: if cfg.use_qkv_bias {
                    Some(vec_of(kv, &mut *nb))
                } else {
                    None
                },
                attn_o: WeightStorage::F32(vec_of(h * h, &mut *nb)),
                ffn_gate: WeightStorage::F32(vec_of(h * i, &mut *nb)),
                ffn_up: WeightStorage::F32(vec_of(h * i, &mut *nb)),
                ffn_down: WeightStorage::F32(vec_of(i * h, &mut *nb)),
            })
            .collect();
        let final_ln_gain = Arc::from(vec![1.0_f32; h]);
        let final_ln_bias = Arc::from(vec![0.0_f32; h]);
        let output = WeightStorage::F32(vec_of(h * cfg.vocab_size, &mut *nb));
        StableLmWeights {
            token_embedding,
            layers,
            final_ln_gain,
            final_ln_bias,
            output,
        }
    }

    #[test]
    fn forward_partial_rotary_factor_0_25() {
        let cfg = StableLmConfig {
            vocab_size: 32,
            hidden_size: 16,
            intermediate_size: 32,
            num_hidden_layers: 2,
            num_attention_heads: 4,
            num_key_value_heads: 4,
            head_dim: 4,
            layer_norm_eps: 1e-5,
            rope_theta: 10_000.0,
            max_position_embeddings: 64,
            partial_rotary_factor: 0.25,
            use_qkv_bias: false,
        };
        // 4 * 0.25 = 1 → rope_dim = 0 (rounded to even). Bump to 1.0
        // for this smoke test so RoPE actually fires.
        let cfg_full = StableLmConfig {
            partial_rotary_factor: 1.0,
            ..cfg
        };
        let model = StableLmModel {
            config: cfg_full.clone(),
            weights: tiny_weights(&cfg_full),
        };
        let logits = model.forward(&[1, 2, 3], 0).unwrap();
        assert_eq!(logits.shape().dims(), &[1, 3, cfg_full.vocab_size]);
        for &v in &logits.realize_f32() {
            assert!(v.is_finite());
        }
    }

    /// With a head_dim large enough that partial rotary is genuinely
    /// partial (rope_dim == 4 of 8), output must differ from the
    /// full-rotary path.
    #[test]
    fn partial_rotary_differs_from_full() {
        let mut cfg = StableLmConfig {
            vocab_size: 16,
            hidden_size: 16,
            intermediate_size: 32,
            num_hidden_layers: 1,
            num_attention_heads: 2,
            num_key_value_heads: 2,
            head_dim: 8,
            layer_norm_eps: 1e-5,
            rope_theta: 10_000.0,
            max_position_embeddings: 32,
            partial_rotary_factor: 0.5, // → rope_dim = 4
            use_qkv_bias: false,
        };
        let weights = tiny_weights(&cfg);
        let out_partial = StableLmModel {
            config: cfg.clone(),
            weights: weights.clone(),
        }
        .forward(&[1, 2, 3, 4], 0)
        .unwrap()
        .realize_f32();
        cfg.partial_rotary_factor = 1.0; // full rotary
        let out_full = StableLmModel {
            config: cfg,
            weights,
        }
        .forward(&[1, 2, 3, 4], 0)
        .unwrap()
        .realize_f32();
        let any_diff = out_partial
            .iter()
            .zip(out_full.iter())
            .any(|(&a, &b)| (a - b).abs() > 1e-7);
        assert!(any_diff, "partial vs full rotary must differ");
    }

    /// `forward_hidden` returns post-LayerNorm hidden states
    /// `(1, seq, hidden_size)` without the lm_head matmul.
    #[test]
    fn forward_hidden_shape_and_finite() {
        let cfg = StableLmConfig {
            vocab_size: 32,
            hidden_size: 16,
            intermediate_size: 32,
            num_hidden_layers: 2,
            num_attention_heads: 4,
            num_key_value_heads: 4,
            head_dim: 4,
            layer_norm_eps: 1e-5,
            rope_theta: 10_000.0,
            max_position_embeddings: 64,
            partial_rotary_factor: 1.0,
            use_qkv_bias: false,
        };
        let model = StableLmModel {
            config: cfg.clone(),
            weights: tiny_weights(&cfg),
        };
        let tokens: Vec<u32> = vec![1, 2, 3, 4];
        let hidden = model.forward_hidden(&tokens, 0).unwrap();
        assert_eq!(hidden.shape().dims(), &[1, tokens.len(), cfg.hidden_size]);
        for &v in &hidden.realize_f32() {
            assert!(v.is_finite(), "non-finite hidden: {v}");
        }
    }

    fn forward_embeds_test_cfg() -> StableLmConfig {
        StableLmConfig {
            vocab_size: 32,
            hidden_size: 16,
            intermediate_size: 32,
            num_hidden_layers: 2,
            num_attention_heads: 4,
            num_key_value_heads: 4,
            head_dim: 4,
            layer_norm_eps: 1e-5,
            rope_theta: 10_000.0,
            max_position_embeddings: 64,
            partial_rotary_factor: 1.0,
            use_qkv_bias: false,
        }
    }

    #[test]
    fn forward_embeds_matches_forward_after_token_lookup() {
        let cfg = forward_embeds_test_cfg();
        let model = StableLmModel {
            config: cfg.clone(),
            weights: tiny_weights(&cfg),
        };
        let tokens: Vec<u32> = vec![1, 2, 3];
        let logits_ref = model.forward(&tokens, 0).unwrap().realize_f32();
        let anchor =
            Tensor::from_f32(vec![0.0_f32], Shape::from_dims(&[1]), &Device::cpu()).unwrap();
        let embeds = model.embed_tokens_anchored(&anchor, &tokens).unwrap();
        let logits_via_embeds = model.forward_embeds(&embeds, 0).unwrap().realize_f32();
        let max_diff = logits_ref
            .iter()
            .zip(logits_via_embeds.iter())
            .map(|(a, b)| (a - b).abs())
            .fold(0.0_f32, f32::max);
        assert!(
            max_diff < 1e-5,
            "StableLm forward vs forward_embeds must agree (max diff {max_diff})"
        );
    }

    #[test]
    fn forward_embeds_rejects_bad_shape() {
        let cfg = forward_embeds_test_cfg();
        let model = StableLmModel {
            config: cfg.clone(),
            weights: tiny_weights(&cfg),
        };
        let bad = Tensor::from_f32(
            vec![0.0_f32; 3 * (cfg.hidden_size + 1)],
            Shape::from_dims(&[1, 3, cfg.hidden_size + 1]),
            &Device::cpu(),
        )
        .unwrap();
        assert!(model.forward_embeds(&bad, 0).is_err());
    }

    #[test]
    fn forward_hidden_embeds_matches_forward_hidden() {
        let cfg = forward_embeds_test_cfg();
        let model = StableLmModel {
            config: cfg.clone(),
            weights: tiny_weights(&cfg),
        };
        let tokens: Vec<u32> = vec![5, 7];
        let h_ref = model.forward_hidden(&tokens, 0).unwrap().realize_f32();
        let anchor =
            Tensor::from_f32(vec![0.0_f32], Shape::from_dims(&[1]), &Device::cpu()).unwrap();
        let embeds = model.embed_tokens_anchored(&anchor, &tokens).unwrap();
        let h_via_embeds = model
            .forward_hidden_embeds(&embeds, 0)
            .unwrap()
            .realize_f32();
        let max_diff = h_ref
            .iter()
            .zip(h_via_embeds.iter())
            .map(|(a, b)| (a - b).abs())
            .fold(0.0_f32, f32::max);
        assert!(
            max_diff < 1e-5,
            "StableLm forward_hidden vs forward_hidden_embeds must agree (max diff {max_diff})"
        );
    }

    // ROADMAP item 8 (II) born-red: config-from-path reads the artifact, routes GQA +
    // head_dim through the shared rules, and carries StableLM's partial-rotary + qkv-bias.
    #[test]
    fn stablelm_config_from_hf_json_parses_the_artifact_not_a_preset() {
        // StableLM-2-1.6B-ish shape WITHOUT head_dim (derived), explicit GQA kv=8.
        let json = r#"{
            "vocab_size": 100352, "hidden_size": 2048, "intermediate_size": 5632,
            "num_hidden_layers": 24, "num_attention_heads": 32, "num_key_value_heads": 8,
            "layer_norm_eps": 1e-5, "rope_theta": 10000.0, "max_position_embeddings": 4096,
            "partial_rotary_factor": 0.25, "use_qkv_bias": true
        }"#;
        let cfg = StableLmConfig::from_hf_json_str(json).expect("parse stablelm config.json");
        assert_eq!(cfg.vocab_size, 100352);
        assert_eq!(cfg.hidden_size, 2048);
        assert_eq!(cfg.intermediate_size, 5632);
        assert_eq!(cfg.num_hidden_layers, 24);
        assert_eq!(cfg.num_attention_heads, 32);
        assert_eq!(cfg.num_key_value_heads, 8); // explicit GQA via the rule
        assert_eq!(cfg.head_dim, 64); // derived: 2048 / 32
        assert_eq!(cfg.layer_norm_eps, 1e-5);
        assert_eq!(cfg.rope_theta, 10000.0);
        assert_eq!(cfg.partial_rotary_factor, 0.25);
        assert!(cfg.use_qkv_bias);
    }

    #[test]
    fn stablelm_config_reads_a_second_distinct_config() {
        // Different values: full rotary, no qkv bias, explicit head_dim.
        let json = r#"{
            "vocab_size": 50304, "hidden_size": 2560, "intermediate_size": 6912,
            "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 32,
            "head_dim": 80, "max_position_embeddings": 4096,
            "partial_rotary_factor": 1.0, "use_qkv_bias": false
        }"#;
        let cfg = StableLmConfig::from_hf_json_str(json).expect("parse second config");
        assert_eq!(cfg.hidden_size, 2560);
        assert_eq!(cfg.head_dim, 80); // explicit
        assert_eq!(cfg.partial_rotary_factor, 1.0);
        assert!(!cfg.use_qkv_bias);
        assert_eq!(cfg.layer_norm_eps, 1e-5); // default applied (absent)
    }

    #[test]
    fn stablelm_config_gqa_defaults_to_num_heads_when_absent() {
        let json = r#"{
            "vocab_size": 100352, "hidden_size": 2048, "intermediate_size": 5632,
            "num_hidden_layers": 24, "num_attention_heads": 32, "max_position_embeddings": 4096
        }"#;
        let cfg = StableLmConfig::from_hf_json_str(json).expect("parse");
        assert_eq!(cfg.num_key_value_heads, 32); // absent -> num_attention_heads (MHA)
    }

    #[test]
    fn stablelm_config_preserves_true_mqa() {
        let json = r#"{
            "vocab_size": 100352, "hidden_size": 2048, "intermediate_size": 5632,
            "num_hidden_layers": 24, "num_attention_heads": 32, "num_key_value_heads": 1,
            "max_position_embeddings": 4096
        }"#;
        let cfg = StableLmConfig::from_hf_json_str(json).expect("parse");
        assert_eq!(cfg.num_key_value_heads, 1); // true MQA survives
    }
}