SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

[modular] fixes! (#33820)

Browse Source 
* fix converter for function definitions

* small changes

* no prints

* style
This commit is contained in:
Arthur
2024-09-30 16:43:55 +02:00
committed by GitHub
parent 1d29a75a6a
commit 1dba608df9
9 changed files with 322 additions and 247 deletions
Download Patch File Download Diff File Expand all files Collapse all files

58
examples/modular-transformers/modeling_dummy.py
View File

@@ -1,11 +1,11 @@
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
# This file was automatically generated from <path_to_diff_file.py>.
# This file was automatically generated from <path_to_modular_file.py>.
# Do NOT edit this file manually as any edits will be overwritten by the generation of
# the file from the diff. If any change should be done, please apply the change to the
# diff.py file directly. One of our CI enforces this
# the file from the modular. If any change should be done, please apply the change to the
# modular_xxx.py file directly. One of our CI enforces this
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
import math
from math import log
from typing import List, Optional, Tuple, Union
import torch
@@ -31,11 +31,6 @@ from ...utils import (
from .configuration_dummy import DummyConfig
def _pre_process_input(input_ids):
print(log(input_ids))
return input_ids
logger = logging.get_logger(__name__)
@@ -129,7 +124,7 @@ class DummyRotaryEmbedding(nn.Module):
if config is None:
logger.warning_once(
"`DummyRotaryEmbedding` can now be fully parameterized by passing the model config through the "
"`config` argument. All other arguments will be removed in v4.45"
"`config` argument. All other arguments will be removed in v4.46"
)
self.rope_kwargs = {
"rope_type": rope_type,
@@ -201,8 +196,8 @@ class DummyRotaryEmbedding(nn.Module):
def rotate_half(x):
"""Rotates half the hidden dims of the input."""
x1 = x[..., : x.shape[-1] // 2]
x2 = x[..., x.shape[-1] // 2 :]
x1 = x[..., : x.shape[-1] // 4]
x2 = x[..., x.shape[-1] // 4 :]
return torch.cat((-x2, x1), dim=-1)
@@ -308,7 +303,7 @@ class DummyAttention(nn.Module):
self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias)
self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=config.attention_bias)
# TODO (joao): remove in v4.45 (RoPE is computed in the model, not in the decoder layers)
# TODO (joao): remove in v4.46 (RoPE is computed in the model, not in the decoder layers)
self.rotary_emb = DummyRotaryEmbedding(config=self.config)
def forward(
@@ -320,7 +315,7 @@ class DummyAttention(nn.Module):
output_attentions: bool = False,
use_cache: bool = False,
cache_position: Optional[torch.LongTensor] = None,
position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # will become mandatory in v4.45
position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # will become mandatory in v4.46
**kwargs,
) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
bsz, q_len, _ = hidden_states.size()
@@ -355,7 +350,7 @@ class DummyAttention(nn.Module):
logger.warning_once(
"The attention layers in this model are transitioning from computing the RoPE embeddings internally "
"through `position_ids` (2D tensor with the indexes of the tokens), to using externally computed "
"`position_embeddings` (Tuple of tensors, containing cos and sin). In v4.45 `position_ids` will be "
"`position_embeddings` (Tuple of tensors, containing cos and sin). In v4.46 `position_ids` will be "
"removed and `position_embeddings` will be mandatory."
)
cos, sin = self.rotary_emb(value_states, position_ids)
@@ -428,7 +423,7 @@ class DummyFlashAttention2(DummyAttention):
output_attentions: bool = False,
use_cache: bool = False,
cache_position: Optional[torch.LongTensor] = None,
position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # will become mandatory in v4.45
position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # will become mandatory in v4.46
) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
if isinstance(past_key_value, StaticCache):
raise ValueError(
@@ -455,7 +450,7 @@ class DummyFlashAttention2(DummyAttention):
logger.warning_once(
"The attention layers in this model are transitioning from computing the RoPE embeddings internally "
"through `position_ids` (2D tensor with the indexes of the tokens), to using externally computed "
"`position_embeddings` (Tuple of tensors, containing cos and sin). In v4.45 `position_ids` will be "
"`position_embeddings` (Tuple of tensors, containing cos and sin). In v4.46 `position_ids` will be "
"removed and `position_embeddings` will be mandatory."
)
cos, sin = self.rotary_emb(value_states, position_ids)
@@ -541,7 +536,7 @@ class DummySdpaAttention(DummyAttention):
output_attentions: bool = False,
use_cache: bool = False,
cache_position: Optional[torch.LongTensor] = None,
position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # will become mandatory in v4.45
position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # will become mandatory in v4.46
**kwargs,
) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
if output_attentions:
@@ -575,7 +570,7 @@ class DummySdpaAttention(DummyAttention):
logger.warning_once(
"The attention layers in this model are transitioning from computing the RoPE embeddings internally "
"through `position_ids` (2D tensor with the indexes of the tokens), to using externally computed "
"`position_embeddings` (Tuple of tensors, containing cos and sin). In v4.45 `position_ids` will be "
"`position_embeddings` (Tuple of tensors, containing cos and sin). In v4.46 `position_ids` will be "
"removed and `position_embeddings` will be mandatory."
)
cos, sin = self.rotary_emb(value_states, position_ids)
@@ -650,7 +645,7 @@ class DummyDecoderLayer(nn.Module):
output_attentions: Optional[bool] = False,
use_cache: Optional[bool] = False,
cache_position: Optional[torch.LongTensor] = None,
position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # will become mandatory in v4.45
position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # will become mandatory in v4.46
**kwargs,
) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
"""
@@ -796,7 +791,8 @@ DUMMY_INPUTS_DOCSTRING = r"""
returned by the model at a previous stage of decoding, when `use_cache=True` or `config.use_cache=True`.
Two formats are allowed:
- a [`~cache_utils.Cache`] instance;
- a [`~cache_utils.Cache`] instance, see our
[kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache);
- Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of
shape `(batch_size, num_heads, sequence_length, embed_size_per_head)`). This is also known as the legacy
cache format.
@@ -877,7 +873,6 @@ class DummyModel(DummyPreTrainedModel):
return_dict: Optional[bool] = None,
cache_position: Optional[torch.LongTensor] = None,
) -> Union[Tuple, BaseModelOutputWithPast]:
input_ids = _pre_process_input(input_ids)
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
output_hidden_states = (
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
@@ -899,16 +894,19 @@ class DummyModel(DummyPreTrainedModel):
if inputs_embeds is None:
inputs_embeds = self.embed_tokens(input_ids)
# kept for BC (non `Cache` `past_key_values` inputs)
return_legacy_cache = False
if (
use_cache and not isinstance(past_key_values, Cache) and not self.training
): # kept for BC (non `Cache` `past_key_values` inputs)
if use_cache and not isinstance(past_key_values, Cache):
return_legacy_cache = True
past_key_values = DynamicCache.from_legacy_cache(past_key_values)
logger.warning_once(
"We detected that you are passing `past_key_values` as a tuple and this is deprecated and will be removed in v4.43. "
"Please use an appropriate `Cache` class (https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
)
if past_key_values is None:
past_key_values = DynamicCache()
else:
past_key_values = DynamicCache.from_legacy_cache(past_key_values)
logger.warning_once(
"We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
"will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
"(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
)
if cache_position is None:
past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0