Add pydocstyle to the ruff rules (#52136)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Signed-off-by: Harry Mellor <hej.mellor@gmail.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Opus 5
parent
804e5377aa
commit
8c1557a79c
@@ -140,8 +140,7 @@ class FalconH1SSMDecoderLayer(nn.Module):
|
||||
self._init_mup_vector()
|
||||
|
||||
def _init_mup_vector(self):
|
||||
"""
|
||||
Non learnable per-block scaling vector composed of element-wise
|
||||
"""Non learnable per-block scaling vector composed of element-wise
|
||||
multipliersapplied to each separate contiguous block of the output
|
||||
of the linear projection (in_proj) before further processing
|
||||
(gating, convolution, SSM):
|
||||
@@ -313,8 +312,7 @@ class FalconH1AttentionDecoderLayer(nn.Module):
|
||||
|
||||
|
||||
class FalconH1ParallelHybrid(nn.Module):
|
||||
"""
|
||||
A hybrid decoder layer for FalconH1 where the input is processed
|
||||
"""A hybrid decoder layer for FalconH1 where the input is processed
|
||||
in parallel through both the self-attention branch and the SSM (Mamba)
|
||||
branch. Their outputs are then summed to produce the final hidden state.
|
||||
|
||||
@@ -552,6 +550,7 @@ class FalconH1ForCausalLM(
|
||||
Tuple containing:
|
||||
- conv_state_shape: Shape for convolutional state cache
|
||||
- temporal_state_shape: Shape for state space model cache
|
||||
|
||||
"""
|
||||
parallel_config = vllm_config.parallel_config
|
||||
hf_config = vllm_config.model_config.hf_config
|
||||
|
||||
Reference in New Issue
Block a user