[Docs] Replace rst style double-backtick with md single-backtick (#27091)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-10-17 02:47:34 -07:00
committed by GitHub
parent 483ea64611
commit 6c9fdbf725
31 changed files with 98 additions and 98 deletions
+3 -3
View File
@@ -64,7 +64,7 @@ from .utils import (
class OlmoAttention(nn.Module):
"""
This is the attention block where the output is computed as
``Attention(LN(x))`` in ``MLP(LN(x + Attention(LN(x))))``
`Attention(LN(x))` in `MLP(LN(x + Attention(LN(x))))`
(plus another skip connection).
"""
@@ -144,7 +144,7 @@ class OlmoAttention(nn.Module):
class OlmoMLP(nn.Module):
"""
This is the MLP block where the output is computed as
``MLP(LN(x))`` in ``MLP(LN(x + Attention(LN(x))))``
`MLP(LN(x))` in `MLP(LN(x + Attention(LN(x))))`
(plus another skip connection).
"""
@@ -193,7 +193,7 @@ class OlmoMLP(nn.Module):
class OlmoDecoderLayer(nn.Module):
"""
This is a typical transformer block where the output is
computed as ``MLP(LN(x + Attention(LN(x))))``
computed as `MLP(LN(x + Attention(LN(x))))`
(plus another skip connection).
"""
+3 -3
View File
@@ -69,7 +69,7 @@ from vllm.transformers_utils.configs import Olmo3Config
class Olmo2Attention(nn.Module):
"""
This is the attention block where the output is computed as
``Attention(LN(x))`` in ``MLP(LN(x + Attention(LN(x))))``
`Attention(LN(x))` in `MLP(LN(x + Attention(LN(x))))`
(plus another skip connection).
"""
@@ -190,7 +190,7 @@ class Olmo2Attention(nn.Module):
class Olmo2MLP(nn.Module):
"""
This is the MLP block where the output is computed as
``MLP(x)`` in ``LN(MLP(x + LN(Attention(x))))``
`MLP(x)` in `LN(MLP(x + LN(Attention(x))))`
(plus another skip connection).
"""
@@ -235,7 +235,7 @@ class Olmo2MLP(nn.Module):
class Olmo2DecoderLayer(nn.Module):
"""
This is a typical transformer block where the output is
computed as ``MLP(LN(x + Attention(LN(x))))``
computed as `MLP(LN(x + Attention(LN(x))))`
(plus another skip connection).
"""
+1 -1
View File
@@ -166,7 +166,7 @@ class VisualTokenizer(torch.nn.Module):
# e.g., for hidden_stride=2, this leads to a token length reduction:
# 1024 -> 256 for aimv2
if self.config.hidden_stride > 1:
# this `d` maybe different from the above `d``
# this `d` maybe different from the above `d`
n, L, d = features.shape
sqrt_l = int(L**0.5)
assert sqrt_l**2 == L, (
+16 -16
View File
@@ -99,13 +99,13 @@ class AutoWeightsLoader:
the weights only once.
The weight loading logic for individual modules can be overridden
by defining a ``load_weights`` method.
by defining a `load_weights` method.
Similarly, the weight loading logic for individual parameters can be
overridden by defining a ``weight_loader`` method.
overridden by defining a `weight_loader` method.
Detailed weight loading information can be viewed by setting the
environment variable ``VLLM_LOGGING_LEVEL=DEBUG``.
environment variable `VLLM_LOGGING_LEVEL=DEBUG`.
"""
# Models trained using early version ColossalAI
@@ -372,9 +372,9 @@ def flatten_bn(
concat: bool = False,
) -> list[torch.Tensor] | torch.Tensor:
"""
Flatten the ``B`` and ``N`` dimensions of batched multimodal inputs.
Flatten the `B` and `N` dimensions of batched multimodal inputs.
The input tensor should have shape ``(B, N, ...)```.
The input tensor should have shape `(B, N, ...)`.
"""
if isinstance(x, torch.Tensor):
return x.flatten(0, 1)
@@ -424,12 +424,12 @@ def _merge_multimodal_embeddings(
is_multimodal: torch.Tensor,
) -> torch.Tensor:
"""
Merge ``multimodal_embeddings`` into ``inputs_embeds`` by overwriting the
positions in ``inputs_embeds`` corresponding to placeholder tokens in
``input_ids``.
Merge `multimodal_embeddings` into `inputs_embeds` by overwriting the
positions in `inputs_embeds` corresponding to placeholder tokens in
`input_ids`.
Note:
This updates ``inputs_embeds`` in place.
This updates `inputs_embeds` in place.
"""
if len(multimodal_embeddings) == 0:
return inputs_embeds
@@ -475,14 +475,14 @@ def merge_multimodal_embeddings(
placeholder_token_id: int | list[int],
) -> torch.Tensor:
"""
Merge ``multimodal_embeddings`` into ``inputs_embeds`` by overwriting the
positions in ``inputs_embeds`` corresponding to placeholder tokens in
``input_ids``.
Merge `multimodal_embeddings` into `inputs_embeds` by overwriting the
positions in `inputs_embeds` corresponding to placeholder tokens in
`input_ids`.
``placeholder_token_id`` can be a list of token ids (e.g, token ids
`placeholder_token_id` can be a list of token ids (e.g, token ids
of img_start, img_break, and img_end tokens) when needed: This means
the order of these tokens in the ``input_ids`` MUST MATCH the order of
their embeddings in ``multimodal_embeddings`` since we need to
the order of these tokens in the `input_ids` MUST MATCH the order of
their embeddings in `multimodal_embeddings` since we need to
slice-merge instead of individually scattering.
For example, if input_ids is "TTTTTSIIIBIIIBIIIETTT", where
@@ -497,7 +497,7 @@ def merge_multimodal_embeddings(
input_ids for a correct embedding merge.
Note:
This updates ``inputs_embeds`` in place.
This updates `inputs_embeds` in place.
"""
if isinstance(placeholder_token_id, list):
is_multimodal = isin_list(input_ids, placeholder_token_id)