broccoli-ml 9.5.1__py3-none-any.whl

broccoli/rope.py

@@ -0,0 +1,407 @@
+"""
+This code is from https://github.com/lucidrains/rotary-embedding-torch
+
+It is provided along with the following copyright notice and license:
+
+MIT License
+
+Copyright (c) 2021 Phil Wang
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+"""
+
+from __future__ import annotations
+from math import pi
+
+import torch
+
+from torch.nn import Module
+from torch import nn, einsum, broadcast_tensors, is_tensor, tensor, Tensor
+
+# Gracefully find the best way to import autocast
+try:
+    from torch.amp import autocast as autocast_factory
+except ImportError:
+    # Fallback: For PyTorch 1.6 to 1.9
+    from torch.cuda.amp import autocast
+
+    def autocast_factory(_, enabled=True):
+        """
+        A wrapper that mimics the modern autocast signature but calls the older
+        torch.cuda.amp.autocast, ignoring the device_type argument.
+        """
+        return autocast(enabled=enabled)
+
+
+from einops import rearrange, repeat
+
+from typing import Literal
+
+# helper functions
+
+
+def exists(val):
+    return val is not None
+
+
+def default(val, d):
+    return val if exists(val) else d
+
+
+# broadcat, as tortoise-tts was using it
+
+
+def broadcat(tensors, dim=-1):
+    broadcasted_tensors = broadcast_tensors(*tensors)
+    return torch.cat(broadcasted_tensors, dim=dim)
+
+
+def slice_at_dim(t, dim_slice: slice, *, dim):
+    dim += t.ndim if dim < 0 else 0
+    colons = [slice(None)] * t.ndim
+    colons[dim] = dim_slice
+    return t[tuple(colons)]
+
+
+# rotary embedding helper functions
+
+
+def rotate_half(x):
+    x = rearrange(x, "... (d r) -> ... d r", r=2)
+    x1, x2 = x.unbind(dim=-1)
+    x = torch.stack((-x2, x1), dim=-1)
+    return rearrange(x, "... d r -> ... (d r)")
+
+
+@autocast_factory("cuda", enabled=False)
+def apply_rotary_emb(
+    freqs, t, start_index=0, scale=1.0, seq_dim=-2, freqs_seq_dim=None
+):
+    dtype = t.dtype
+
+    if not exists(freqs_seq_dim):
+        if freqs.ndim == 2 or t.ndim == 3:
+            freqs_seq_dim = 0
+
+    if t.ndim == 3 or exists(freqs_seq_dim):
+        seq_len = t.shape[seq_dim]
+        freqs = slice_at_dim(freqs, slice(-seq_len, None), dim=freqs_seq_dim)
+
+    rot_dim = freqs.shape[-1]
+    end_index = start_index + rot_dim
+
+    assert rot_dim <= t.shape[-1], (
+        f"feature dimension {t.shape[-1]} is not of sufficient size "
+        "to rotate in all the positions {rot_dim}"
+    )
+
+    # Split t into three parts: left, middle (to be transformed), and right
+    t_left = t[..., :start_index]
+    t_middle = t[..., start_index:end_index]
+    t_right = t[..., end_index:]
+
+    # Apply rotary embeddings without modifying t in place
+    t_transformed = (t_middle * freqs.cos() * scale) + (
+        rotate_half(t_middle) * freqs.sin() * scale
+    )
+
+    out = torch.cat((t_left, t_transformed, t_right), dim=-1)
+
+    return out.type(dtype)
+
+
+# learned rotation helpers
+
+
+def apply_learned_rotations(rotations, t, start_index=0, freq_ranges=None):
+    if exists(freq_ranges):
+        rotations = einsum("..., f -> ... f", rotations, freq_ranges)
+        rotations = rearrange(rotations, "... r f -> ... (r f)")
+
+    rotations = repeat(rotations, "... n -> ... (n r)", r=2)
+    return apply_rotary_emb(rotations, t, start_index=start_index)
+
+
+# classes
+
+
+class RotaryEmbedding(Module):
+    def __init__(
+        self,
+        dim,
+        custom_freqs: Tensor | None = None,
+        freqs_for: Literal["lang", "pixel", "constant"] = "lang",
+        theta=10000,
+        max_freq=10,
+        num_freqs=1,
+        learned_freq=False,
+        use_xpos=False,
+        xpos_scale_base=512,
+        interpolate_factor=1.0,
+        theta_rescale_factor=1.0,
+        seq_before_head_dim=False,
+        cache_if_possible=True,
+        cache_max_seq_len=8192,
+    ):
+        super().__init__()
+        # proposed by reddit user bloc97,
+        # to rescale rotary embeddings to longer sequence length without fine-tuning
+        # has some connection to NTK literature
+        # https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/
+
+        theta *= theta_rescale_factor ** (dim / (dim - 2))
+
+        self.freqs_for = freqs_for
+
+        if exists(custom_freqs):
+            freqs = custom_freqs
+        elif freqs_for == "lang":
+            freqs = 1.0 / (
+                theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim)
+            )
+        elif freqs_for == "pixel":
+            freqs = torch.linspace(1.0, max_freq / 2, dim // 2) * pi
+        elif freqs_for == "constant":
+            freqs = torch.ones(num_freqs).float()
+
+        self.cache_if_possible = cache_if_possible
+        self.cache_max_seq_len = cache_max_seq_len
+
+        self.register_buffer(
+            "cached_freqs", torch.zeros(cache_max_seq_len, dim), persistent=False
+        )
+        self.cached_freqs_seq_len = 0
+
+        self.freqs = nn.Parameter(freqs, requires_grad=learned_freq)
+
+        self.learned_freq = learned_freq
+
+        # dummy for device
+
+        self.register_buffer("dummy", torch.tensor(0), persistent=False)
+
+        # default sequence dimension
+
+        self.seq_before_head_dim = seq_before_head_dim
+        self.default_seq_dim = -3 if seq_before_head_dim else -2
+
+        # interpolation factors
+
+        assert interpolate_factor >= 1.0
+        self.interpolate_factor = interpolate_factor
+
+        # xpos
+
+        self.use_xpos = use_xpos
+
+        if not use_xpos:
+            return
+
+        scale = (torch.arange(0, dim, 2) + 0.4 * dim) / (1.4 * dim)
+        self.scale_base = xpos_scale_base
+
+        self.register_buffer("scale", scale, persistent=False)
+        self.register_buffer(
+            "cached_scales", torch.zeros(cache_max_seq_len, dim), persistent=False
+        )
+        self.cached_scales_seq_len = 0
+
+        # add apply_rotary_emb as static method
+
+        self.apply_rotary_emb = staticmethod(apply_rotary_emb)
+
+    @property
+    def device(self):
+        return self.dummy.device
+
+    def get_seq_pos(self, seq_len, device=None, dtype=None, offset=0):
+        device = default(device, self.device)
+        dtype = default(dtype, self.cached_freqs.dtype)
+
+        return (
+            torch.arange(seq_len, device=device, dtype=dtype) + offset
+        ) / self.interpolate_factor
+
+    def rotate_queries_or_keys(self, t, seq_dim=None, offset=0, scale=None):
+        seq_dim = default(seq_dim, self.default_seq_dim)
+
+        assert not self.use_xpos or exists(scale), (
+            "you must use `.rotate_queries_and_keys` method instead and pass "
+            "in both queries and keys, for length extrapolatable rotary embeddings"
+        )
+
+        device, dtype, seq_len = t.device, t.dtype, t.shape[seq_dim]
+
+        seq = self.get_seq_pos(seq_len, device=device, dtype=dtype, offset=offset)
+
+        freqs = self.forward(seq, seq_len=seq_len, offset=offset)
+
+        if seq_dim == -3:
+            freqs = rearrange(freqs, "n d -> n 1 d")
+
+        return apply_rotary_emb(freqs, t, scale=default(scale, 1.0), seq_dim=seq_dim)
+
+    def rotate_queries_with_cached_keys(self, q, k, seq_dim=None, offset=0):
+        dtype, device, seq_dim = (
+            q.dtype,
+            q.device,
+            default(seq_dim, self.default_seq_dim),
+        )
+
+        q_len, k_len = q.shape[seq_dim], k.shape[seq_dim]
+        assert q_len <= k_len
+
+        q_scale = k_scale = 1.0
+
+        if self.use_xpos:
+            seq = self.get_seq_pos(k_len, dtype=dtype, device=device)
+
+            q_scale = self.get_scale(seq[-q_len:]).type(dtype)
+            k_scale = self.get_scale(seq).type(dtype)
+
+        rotated_q = self.rotate_queries_or_keys(
+            q, seq_dim=seq_dim, scale=q_scale, offset=k_len - q_len + offset
+        )
+        rotated_k = self.rotate_queries_or_keys(k, seq_dim=seq_dim, scale=k_scale**-1)
+
+        rotated_q = rotated_q.type(q.dtype)
+        rotated_k = rotated_k.type(k.dtype)
+
+        return rotated_q, rotated_k
+
+    def rotate_queries_and_keys(self, q, k, seq_dim=None):
+        seq_dim = default(seq_dim, self.default_seq_dim)
+
+        assert self.use_xpos
+        device, dtype, seq_len = q.device, q.dtype, q.shape[seq_dim]
+
+        seq = self.get_seq_pos(seq_len, dtype=dtype, device=device)
+
+        freqs = self.forward(seq, seq_len=seq_len)
+        scale = self.get_scale(seq, seq_len=seq_len).to(dtype)
+
+        if seq_dim == -3:
+            freqs = rearrange(freqs, "n d -> n 1 d")
+            scale = rearrange(scale, "n d -> n 1 d")
+
+        rotated_q = apply_rotary_emb(freqs, q, scale=scale, seq_dim=seq_dim)
+        rotated_k = apply_rotary_emb(freqs, k, scale=scale**-1, seq_dim=seq_dim)
+
+        rotated_q = rotated_q.type(q.dtype)
+        rotated_k = rotated_k.type(k.dtype)
+
+        return rotated_q, rotated_k
+
+    def get_scale(self, t: Tensor, seq_len: int | None = None, offset=0):
+        assert self.use_xpos
+
+        should_cache = (
+            self.cache_if_possible
+            and exists(seq_len)
+            and (offset + seq_len) <= self.cache_max_seq_len
+        )
+
+        if (
+            should_cache
+            and exists(self.cached_scales)
+            and (seq_len + offset) <= self.cached_scales_seq_len
+        ):
+            return self.cached_scales[offset : (offset + seq_len)]
+
+        scale = 1.0
+        if self.use_xpos:
+            power = (t - len(t) // 2) / self.scale_base
+            scale = self.scale ** rearrange(power, "n -> n 1")
+            scale = repeat(scale, "n d -> n (d r)", r=2)
+
+        if should_cache and offset == 0:
+            self.cached_scales[:seq_len] = scale.detach()
+            self.cached_scales_seq_len = seq_len
+
+        return scale
+
+    def get_axial_freqs(
+        self, *dims, offsets: tuple[int | float, ...] | Tensor | None = None
+    ):
+        Colon = slice(None)
+        all_freqs = []
+
+        # handle offset
+
+        if exists(offsets):
+            if not is_tensor(offsets):
+                offsets = tensor(offsets)
+
+            assert len(offsets) == len(dims)
+
+        # get frequencies for each axis
+
+        for ind, dim in enumerate(dims):
+
+            offset = 0
+            if exists(offsets):
+                offset = offsets[ind]
+
+            if self.freqs_for == "pixel":
+                pos = torch.linspace(-1, 1, steps=dim, device=self.device)
+            else:
+                pos = torch.arange(dim, device=self.device)
+
+            pos = pos + offset
+
+            freqs = self.forward(pos, seq_len=dim)
+
+            all_axis = [None] * len(dims)
+            all_axis[ind] = Colon
+
+            new_axis_slice = (Ellipsis, *all_axis, Colon)
+            all_freqs.append(freqs[new_axis_slice])
+
+        # concat all freqs
+
+        all_freqs = broadcast_tensors(*all_freqs)
+        return torch.cat(all_freqs, dim=-1)
+
+    @autocast_factory("cuda", enabled=False)
+    def forward(self, t: Tensor, seq_len: int | None = None, offset=0):
+        should_cache = (
+            self.cache_if_possible
+            and not self.learned_freq
+            and exists(seq_len)
+            and self.freqs_for != "pixel"
+            and (offset + seq_len) <= self.cache_max_seq_len
+        )
+
+        if (
+            should_cache
+            and exists(self.cached_freqs)
+            and (offset + seq_len) <= self.cached_freqs_seq_len
+        ):
+            return self.cached_freqs[offset : (offset + seq_len)].detach()
+
+        freqs = self.freqs
+
+        freqs = einsum("..., f -> ... f", t.type(freqs.dtype), freqs)
+        freqs = repeat(freqs, "... n -> ... (n r)", r=2)
+
+        if should_cache and offset == 0:
+            self.cached_freqs[:seq_len] = freqs.detach()
+            self.cached_freqs_seq_len = seq_len
+
+        return freqs

broccoli/tensor.py

@@ -0,0 +1,128 @@
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+
+class SigmaReparamTensor(nn.Module):
+    """
+    Inspired by Apple's Spectral Normed Linear Layers
+        (https://github.com/apple/ml-sigma-reparam)
+    """
+
+    def __init__(self, init_tensor: torch.Tensor):
+        assert init_tensor.ndim == 2
+
+        super().__init__()
+
+        self.sigma_reparam_tensor = nn.Parameter(init_tensor, requires_grad=True)
+
+        with torch.no_grad():
+            _, sigma, v_transpose = torch.linalg.svd(
+                self.sigma_reparam_tensor, full_matrices=False
+            )
+
+        self.register_buffer("approx_spectral_norm", sigma[:1])
+        self.register_buffer("right_singular", v_transpose[0])
+        self.sigma_reparam_scale = nn.Parameter(
+            self.approx_spectral_norm.clone().detach(), requires_grad=True
+        )
+
+    def power_iteration(self):
+        with torch.no_grad():
+            approx_right_singular_transpose = self.sigma_reparam_tensor.mv(
+                self.right_singular
+            )
+            approx_right_singular_transpose = F.normalize(
+                approx_right_singular_transpose, dim=0
+            )
+            updated_right_singular = self.sigma_reparam_tensor.T.mv(
+                approx_right_singular_transpose
+            )
+            updated_right_singular = F.normalize(updated_right_singular, dim=0)
+            self.right_singular.data.copy_(updated_right_singular)
+            rayleigh_quotient = torch.einsum(
+                "m,mn,n->",
+                approx_right_singular_transpose,
+                self.sigma_reparam_tensor,
+                updated_right_singular,
+            )
+            self.approx_spectral_norm.data.copy_(rayleigh_quotient)
+
+    def forward(self):
+        if self.training:
+            self.power_iteration()
+        return self.sigma_reparam_scale * (
+            self.sigma_reparam_tensor / self.approx_spectral_norm
+        )
+
+
+class AnchoredReparamTensor(nn.Module):
+    """
+    Reparameterises a tensor by decoupling its magnitude and direction.
+
+    The direction is represented by a learnable weight tensor, normalised by the
+    Rayleigh quotient with respect to its initial dominant right-singular vector.
+    The magnitude is a separate learnable scalar.
+
+    The reparameterization is:
+
+        W_reparam = scale * (W / norm)
+
+    where the norm is the Rayleigh quotient uᵀWv₀, with v₀ being the dominant
+    right-singular vector of the initial tensor and u = normalize(Wv₀).
+    """
+
+    def __init__(self, init_tensor: torch.Tensor):
+        assert init_tensor.ndim == 2
+
+        super().__init__()
+
+        self.weight = nn.Parameter(init_tensor, requires_grad=True)
+
+        with torch.no_grad():
+            _, sigma, v_transpose = torch.linalg.svd(self.weight, full_matrices=False)
+
+        self.register_buffer("rayleigh_norm", sigma[:1])
+        self.register_buffer("initial_right_singular", v_transpose[0])
+        self.nondecay_scale = nn.Parameter(
+            sigma[:1].clone().detach(), requires_grad=True
+        )
+
+    def _update_rayleigh_norm(self):
+        with torch.no_grad():
+            product = self.weight.mv(self.initial_right_singular)
+            normed_product = F.normalize(product, dim=0)
+            rayleigh_norm = torch.einsum(
+                "m,mn,n->",
+                normed_product,
+                self.weight,
+                self.initial_right_singular,
+            )
+            self.rayleigh_norm.data.copy_(rayleigh_norm)
+
+    def forward(self):
+        if self.training:
+            self._update_rayleigh_norm()
+        return self.nondecay_scale * (self.weight / (self.rayleigh_norm + 1e-6))
+
+
+class NormReparamTensor(nn.Module):
+    """
+    Reparameterise a tensor as a normalised tensor of weights multiplied by a
+        learnable scaling factor.
+    """
+
+    def __init__(self, init_tensor: torch.Tensor):
+        assert init_tensor.ndim == 2, "Input tensor must be a 2D matrix."
+        super().__init__()
+
+        # Use the gradboard convention of calling something nondecay_* if we should
+        # exclude it from weight decay
+        self.weight = nn.Parameter(init_tensor.clone(), requires_grad=True)
+        self.nondecay_scale = nn.Parameter(
+            torch.linalg.norm(self.weight).clone().detach(), requires_grad=True
+        )
+
+    def forward(self) -> torch.Tensor:
+        norm = torch.linalg.norm(self.weight)
+        return self.nondecay_scale * (self.weight / (norm + 1e-6))