[CI] Ignore ruff D209, rejoin the docstrings it split, and silence incompatible-rule warnings (#57212)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Harry Mellor
2026-09-16 10:57:24 -07:00
committed by GitHub
co-authored by Claude Opus 5
parent 1085b64425
commit 651a88c09a
693 changed files with 2088 additions and 4171 deletions
+3 -6
View File
@@ -564,8 +564,7 @@ def _engram_head_shard_weight_loader(
param: torch.nn.Parameter, loaded_weight: torch.Tensor
) -> None:
"""Load this rank's complete head buckets. ue8m0 scales arrive as
float8_e8m0fnu; keep the raw bytes (the param stores uint8).
"""
float8_e8m0fnu; keep the raw bytes (the param stores uint8)."""
part_rows = param.shape[0]
if loaded_weight.dtype == torch.float8_e8m0fnu:
loaded_weight = loaded_weight.view(torch.uint8)
@@ -740,8 +739,7 @@ class ParallelEngramEmbedding(nn.Module):
def forward(self, indices: torch.Tensor) -> torch.Tensor:
"""indices: [num_tokens, n_hash_cols] -> [num_tokens, n_hash_cols, dim]
bf16, gathered from all shards for this replica's tokens.
"""
bf16, gathered from all shards for this replica's tokens."""
out = torch.empty(
(indices.shape[0], self.part_n_hash_cols, self.dim),
dtype=torch.bfloat16,
@@ -1002,8 +1000,7 @@ class Engram(nn.Module):
) -> torch.Tensor:
"""hidden_states: [T, hc_mult, dim]; hash_ids: [T, n_hash_cols] (all
tokens, pre sequence-parallel shard); token_mask: [T], False shuts
the gate so those positions pass through untouched.
"""
the gate so those positions pass through untouched."""
kv = self.wkv(self.embed(hash_ids).flatten(-2))
num_kv_tokens = hash_ids.shape[0]
assert token_mask is None or token_mask.shape == (num_kv_tokens,)
@@ -82,8 +82,7 @@ def num_image_tokens(n_llm_h: int, n_llm_w: int) -> int:
def solve_resize_ratio(height, width, patch_size, downsample_ratio, max_n_token):
"""Largest aspect-preserving pixel size whose token grid still fits in
max_n_token. Returns (best_height, best_width).
"""
max_n_token. Returns (best_height, best_width)."""
r = height / width
max_w_float = math.sqrt((max_n_token - 2) / r + 0.25) - 0.5
max_h_float = max_w_float * r
@@ -106,8 +105,7 @@ def safe_resize(
height, width, best_height, best_width, patch_size, downsample_ratio, max_n_token
):
"""Shrink the pixel size until the image costs at most max_n_token LLM
tokens.
"""
tokens."""
n_llm_h, n_llm_w = llm_grid(best_height, best_width, patch_size, downsample_ratio)
if num_image_tokens(n_llm_h, n_llm_w) > max_n_token:
best_height, best_width = solve_resize_ratio(
@@ -173,8 +171,7 @@ def image_token_types(n_llm_h: int, n_llm_w: int) -> torch.Tensor:
class DeepseekV4VLImageProcessor:
"""Per-image transform (the PIL-input equivalent of the reference
``load_image``).
"""
``load_image``)."""
def __init__(self, config: DeepseekV41Config) -> None:
super().__init__()