Generated GPT_OSS model files through porter script.

keras_hub/api/models/__init__.py

@@ -340,6 +340,18 @@
 from keras_hub.src.models.gpt_neo_x.gpt_neo_x_tokenizer import (
     GPTNeoXTokenizer as GPTNeoXTokenizer,
 )
+from keras_hub.src.models.gpt_oss.gpt_oss_backbone import (
+    GptOssBackbone as GptOssBackbone,
+)
+from keras_hub.src.models.gpt_oss.gpt_oss_causal_lm import (
+    GptOssCausalLM as GptOssCausalLM,
+)
+from keras_hub.src.models.gpt_oss.gpt_oss_causal_lm_preprocessor import (
+    GptOssCausalLMPreprocessor as GptOssCausalLMPreprocessor,
+)
+from keras_hub.src.models.gpt_oss.gpt_oss_tokenizer import (
+    GptOssTokenizer as GptOssTokenizer,
+)
 from keras_hub.src.models.hgnetv2.hgnetv2_backbone import (
     HGNetV2Backbone as HGNetV2Backbone,
 )

keras_hub/api/tokenizers/__init__.py

@@ -47,6 +47,9 @@
 from keras_hub.src.models.gpt_neo_x.gpt_neo_x_tokenizer import (
     GPTNeoXTokenizer as GPTNeoXTokenizer,
 )
+from keras_hub.src.models.gpt_oss.gpt_oss_tokenizer import (
+    GptOssTokenizer as GptOssTokenizer,
+)
 from keras_hub.src.models.llama.llama_tokenizer import (
     LlamaTokenizer as LlamaTokenizer,
 )

keras_hub/src/layers/modeling/rotary_embedding.py

@@ -25,6 +25,17 @@ class RotaryEmbedding(keras.layers.Layer):
             curves.
         scaling_factor: float. The scaling factor used to scale positions of
             the tokens.
+        rope_type: str. The type of RoPE scaling to apply. Supported types:
+            "linear", "dynamic", "yarn". Defaults to "linear".
+        beta_fast: float. Beta fast parameter for YaRN scaling. Only used
+            when rope_type="yarn". Defaults to 32.0.
+        beta_slow: float. Beta slow parameter for YaRN scaling. Only used
+            when rope_type="yarn". Defaults to 1.0.
+        original_max_position_embeddings: int. Original maximum position
+            embeddings for YaRN scaling. Only used when rope_type="yarn".
+            Defaults to 4096.
+        truncate: bool. Whether to apply truncation for YaRN scaling. Only used
+            when rope_type="yarn". Defaults to False.
         sequence_axis: int. Sequence axis in the input tensor.
         feature_axis: int. Feature axis in the input tensor.
         **kwargs: other keyword arguments passed to `keras.layers.Layer`,
@@ -69,33 +80,89 @@ def __init__(
         self,
         max_wavelength=10000,
         scaling_factor=1.0,
+        rope_type="linear",
+        beta_fast=32.0,
+        beta_slow=1.0,
+        original_max_position_embeddings=4096,
+        truncate=False,
         sequence_axis=1,
         feature_axis=-1,
         **kwargs,
     ):
         super().__init__(**kwargs)
         self.max_wavelength = max_wavelength
-        self.sequence_axis = sequence_axis
-        self.feature_axis = feature_axis
         self.scaling_factor = scaling_factor
-        self.built = True
+        self.rope_type = rope_type
+
+        # YaRN-specific parameters (only used when rope_type="yarn")
+        self.beta_fast = beta_fast
+        self.beta_slow = beta_slow
+        self.original_max_position_embeddings = original_max_position_embeddings
+        self.truncate = truncate
+
+        # Store original axis values for validation
+        self._original_sequence_axis = sequence_axis
+        self._original_feature_axis = feature_axis
+
+    def _normalize_axes(self, input_shape):
+        """Normalize and validate axis indices for the given input shape."""
+        rank = len(input_shape)
+
+        # Normalize negative indices
+        sequence_axis = self._original_sequence_axis
+        feature_axis = self._original_feature_axis
+
+        if sequence_axis < 0:
+            sequence_axis += rank
+        if feature_axis < 0:
+            feature_axis += rank
+
+        # Validate axis indices
+        if sequence_axis < 0 or sequence_axis >= rank:
+            raise ValueError(
+                f"sequence_axis {self._original_sequence_axis} "
+                f"is out of range for input with rank {rank}"
+            )
+        if feature_axis < 0 or feature_axis >= rank:
+            raise ValueError(
+                f"feature_axis {self._original_feature_axis} "
+                f"is out of range for input with rank {rank}"
+            )
+        if sequence_axis == feature_axis:
+            raise ValueError("sequence_axis and feature_axis must be different")
+
+        return sequence_axis, feature_axis
+
+    def _validate_rotary_dimension(self, rotary_dim):
+        """Validate that rotary dimension is even and handle odd dimensions."""
+        if rotary_dim % 2 != 0:
+            raise ValueError(
+                f"Rotary dimension must be even, got {rotary_dim}."
+                "The rotary embedding splits the feature dimension "
+                "into two halves. Consider using a different feature "
+                "dimension or padding."
+            )
 
     def call(self, inputs, start_index=0, positions=None):
+        # Normalize and validate axes
+        input_shape = ops.shape(inputs)
+        sequence_axis, feature_axis = self._normalize_axes(input_shape)
+
+        # Validate rotary dimension
+        rotary_dim = input_shape[feature_axis]
+        self._validate_rotary_dimension(rotary_dim)
+
         # Take care of unbatched `positions`.
         if positions is not None:
             if len(ops.shape(positions)) == 1:
                 positions = ops.expand_dims(positions, axis=0)
 
-        inputs = ops.moveaxis(
-            inputs, (self.feature_axis, self.sequence_axis), (-1, 1)
-        )
+        inputs = ops.moveaxis(inputs, (feature_axis, sequence_axis), (-1, 1))
         cos_emb, sin_emb = self._compute_cos_sin_embedding(
             inputs, start_index, positions
         )
         output = self._apply_rotary_pos_emb(inputs, cos_emb, sin_emb)
-        return ops.moveaxis(
-            output, (-1, 1), (self.feature_axis, self.sequence_axis)
-        )
+        return ops.moveaxis(output, (-1, 1), (feature_axis, sequence_axis))
 
     def _apply_rotary_pos_emb(self, tensor, cos_emb, sin_emb):
         x1, x2 = ops.split(tensor, 2, axis=-1)
@@ -113,51 +180,231 @@ def _compute_positions(self, inputs, start_index=0):
         return positions + ops.cast(start_index, dtype="float32")
 
     def _compute_cos_sin_embedding(self, inputs, start_index=0, positions=None):
+        """Compute cos & sin RoPE embeddings with optional YaRN scaling.
+        Uses tensor ops only to remain JIT/backends friendly.
+        """
         batch_axis = 0
-        feature_axis = len(inputs.shape) - 1
         sequence_axis = 1
+        feature_axis = len(inputs.shape) - 1
 
+        # rotary_dim should be half of the last
+        # feature axis (HF-style: rotate pairs)
         rotary_dim = ops.shape(inputs)[feature_axis]
+        # Validate evenness
+        try:
+            # best-effort check when running eagerly;
+            # if unavailable this will be a no-op
+            if int(rotary_dim) % 2 != 0:
+                raise ValueError(
+                    "Rotary embedding requires even feature "
+                    "dimension (last axis)."
+                )
+        except Exception:
+            pass
+
+        # Get inverse frequencies using the appropriate
+        # scaling method (linear, dynamic, yarn, etc.)
         inverse_freq = self._get_inverse_freq(rotary_dim)
 
+        # positions handling
         if positions is None:
             positions = self._compute_positions(inputs, start_index)
-            positions = ops.expand_dims(positions, axis=batch_axis)
+            positions = ops.expand_dims(
+                positions, axis=batch_axis
+            )  # shape (1, seq_len)
         else:
+            # ensure float dtype and batch dim
             positions = ops.cast(positions, "float32")
-        positions = positions / ops.cast(self.scaling_factor, "float32")
+            if len(ops.shape(positions)) == 1:
+                positions = ops.expand_dims(positions, axis=batch_axis)
+
+        # Apply truncation for YaRN if specified
+        if (
+            self.rope_type == "yarn"
+            and self.truncate
+            and self.original_max_position_embeddings is not None
+        ):
+            positions = ops.minimum(
+                positions,
+                ops.cast(self.original_max_position_embeddings, "float32"),
+            )
 
+        # compute outer product positions x inverse_freq ->
+        # shape (batch?, seq_len, rotary_dim//2)
+        # If positions has batch dim, einsum handles it.
         freq = ops.einsum("bi,j->bij", positions, inverse_freq)
 
+        # stack to interleave sin/cos dims and reshape to full rotary dim
         embedding = ops.stack((freq, freq), axis=-2)
         embedding = ops.reshape(
             embedding, (*ops.shape(freq)[:-1], ops.shape(freq)[-1] * 2)
         )
 
+        # Expand embedding to match inputs rank
+        # (insert axes for any non-batch/seq/feature dims)
         for axis in range(len(inputs.shape)):
             if axis not in (batch_axis, sequence_axis, feature_axis):
                 embedding = ops.expand_dims(embedding, axis)
 
         cos_emb = ops.cast(ops.cos(embedding), self.compute_dtype)
         sin_emb = ops.cast(ops.sin(embedding), self.compute_dtype)
+
+        # YaRN temperature scaling: implement in tensor ops
+        if self.rope_type == "yarn":
+            # t = (0.1 * ln(s) + 1)^2
+            # make sure s > 0
+            small = ops.cast(1e-6, self.compute_dtype)
+            s_safe = ops.maximum(
+                ops.cast(self.scaling_factor, self.compute_dtype), small
+            )
+            t = ops.square(
+                ops.add(
+                    ops.multiply(
+                        ops.cast(0.1, self.compute_dtype), ops.log(s_safe)
+                    ),
+                    ops.cast(1.0, self.compute_dtype),
+                )
+            )
+            sqrt_t = ops.sqrt(t)
+
+            # HF/YaRN descriptions indicate a temperature
+            # scaling applied to cos/sin embeddings, equivalently
+            # scaling the logits.We implement the sqrt scaling on cos/sin.
+            cos_emb = cos_emb * sqrt_t
+            sin_emb = sin_emb * sqrt_t
+
         return cos_emb, sin_emb
 
     def _get_inverse_freq(self, rotary_dim):
-        freq_range = ops.divide(
-            ops.arange(0, rotary_dim, 2, dtype="float32"),
-            ops.cast(rotary_dim, "float32"),
-        )
-        inverse_freq = 1.0 / (self.max_wavelength**freq_range)
-        return inverse_freq
+        """Return inverse frequencies."""
+        # rotary_dim expected to be python int or small tensor;
+        # create idx with dtype
+        idx = ops.arange(0, rotary_dim, 2, dtype="float32")
+        denom = ops.cast(rotary_dim, "float32")
+        freq_range = idx / denom
+        inv = ops.power(ops.cast(self.max_wavelength, "float32"), -freq_range)
+
+        # apply rope_scaling variants
+        if self.rope_type == "linear":
+            # linear: divide inverse freqs by factor
+            # (consistent with HF linear scaling semantics)
+            return inv / ops.cast(self.scaling_factor, "float32")
+        elif self.rope_type == "dynamic":
+            # dynamic (NTK-aware) fallback conservative implementation:
+            # HF dynamic implementation uses NTK-by-parts;
+            # use a practical scaling to approximate.
+            # Here we conservatively divide
+            # by scaling_factor^(rotary_dim/(rotary_dim-2))
+            exponent = ops.cast(rotary_dim, "float32") / ops.cast(
+                max(1, rotary_dim - 2), "float32"
+            )
+            return inv / ops.power(
+                ops.cast(self.scaling_factor, "float32"), exponent
+            )
+        elif self.rope_type == "yarn":
+            # Delegate to more advanced YaRN inverse freq routine
+            return self._get_yarn_inverse_freq(inv, rotary_dim)
+        else:
+            return inv
+
+    def _get_yarn_inverse_freq(self, base_inverse_freq, rotary_dim):
+        """YaRN NTK-by-parts style inverse frequency scaling
+        (tensor-friendly).This follows the YaRN paper and common
+        porting decisions used in HF forks.
+        """
+        s = ops.cast(self.scaling_factor, "float32")
+
+        # Get the base (rope_theta equivalent) from max_wavelength
+        base = ops.cast(self.max_wavelength, "float32")
+
+        # Compute base frequencies: base ** (idx / dim)
+        idx = ops.arange(0, rotary_dim, 2, dtype="float32")
+        pos_freqs = ops.power(base, idx / ops.cast(rotary_dim, "float32"))
+
+        # Compute interpolation and extrapolation frequencies
+        inv_freq_extrapolation = 1.0 / pos_freqs
+        inv_freq_interpolation = 1.0 / (s * pos_freqs)
+
+        # Find correction range (same logic as HuggingFace)
+        if (
+            self.beta_fast is not None
+            and self.beta_slow is not None
+            and self.original_max_position_embeddings is not None
+        ):
+            L = ops.cast(self.original_max_position_embeddings, "float32")
+            beta_fast = ops.cast(self.beta_fast, "float32")
+            beta_slow = ops.cast(self.beta_slow, "float32")
+
+            # Find correction dimensions for beta_fast and beta_slow
+            def find_correction_dim_tensor(
+                num_rotations, dim, base_val, max_pos
+            ):
+                return (
+                    dim
+                    * ops.log(max_pos / (num_rotations * 2 * 3.141592653589793))
+                ) / (2 * ops.log(base_val))
+
+            low = find_correction_dim_tensor(
+                beta_fast, ops.cast(rotary_dim, "float32"), base, L
+            )
+            high = find_correction_dim_tensor(
+                beta_slow, ops.cast(rotary_dim, "float32"), base, L
+            )
+
+            # Apply truncation if specified
+            if self.truncate:
+                low = ops.floor(low)
+                high = ops.ceil(high)
+
+            # Clamp to valid range
+            low = ops.maximum(low, ops.cast(0, "float32"))
+            high = ops.minimum(high, ops.cast(rotary_dim // 2 - 1, "float32"))
+
+            # Linear ramp function
+            dim_half = rotary_dim // 2
+            idx_half = ops.arange(0, dim_half, dtype="float32")
+
+            # Prevent singularity
+            diff = high - low
+            diff = ops.maximum(diff, ops.cast(0.001, "float32"))
+
+            linear_func = (idx_half - low) / diff
+            ramp_func = ops.clip(linear_func, 0, 1)
+
+            # Apply the ramp to get extrapolation factor
+            inv_freq_extrapolation_factor = 1 - ramp_func
+
+            # Combine interpolation and extrapolation
+            scaled_inverse_freq = (
+                inv_freq_interpolation * (1 - inv_freq_extrapolation_factor)
+                + inv_freq_extrapolation * inv_freq_extrapolation_factor
+            )
+        else:
+            # Fallback to simple scaling
+            alpha = ops.power(
+                s,
+                ops.cast(rotary_dim, "float32")
+                / ops.cast(max(1, rotary_dim - 2), "float32"),
+            )
+            scaled_inverse_freq = base_inverse_freq / alpha
+
+        return scaled_inverse_freq
 
     def get_config(self):
         config = super().get_config()
         config.update(
             {
                 "max_wavelength": self.max_wavelength,
                 "scaling_factor": self.scaling_factor,
-                "sequence_axis": self.sequence_axis,
-                "feature_axis": self.feature_axis,
+                "rope_type": self.rope_type,
+                "beta_fast": self.beta_fast,
+                "beta_slow": self.beta_slow,
+                "original_max_position_embeddings": (
+                    self.original_max_position_embeddings
+                ),
+                "truncate": self.truncate,
+                "sequence_axis": self._original_sequence_axis,
+                "feature_axis": self._original_feature_axis,
             }
         )
         return config