Function quantize_activations 

pub fn quantize_activations(
    activations: &Tensor,
    config: &BitNetConfig,
) -> Result<QuantizedActivations>

Quantize activations using per-token AbsMax scaling to INT8.

Algorithm

For each token (row):

1. Compute scale = max(|X|) / 127
2. Compute X_q = round(X / scale) clamped to [-127, 127]

Arguments

• activations - Input tensor [batch, seq_len, hidden_dim] or [batch, hidden_dim]
• config - BitNet configuration

Errors

Returns error if quantization fails.