riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-07-31 10:12:23 +06:00
Code Issues Actions 23 Packages Projects Releases Wiki Activity

Add Pytorch Tensor Parallel support for Qwen2, Qwen2Moe, Starcoder2 (#35007)

Browse Source 
* add base tp plan for qwen2 and qwen2moe

* add parallel tp for starcoder2

* fix modular conversion

* add infer dim for qkv states

* Update src/transformers/models/qwen2_moe/configuration_qwen2_moe.py

---------

Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
This commit is contained in:
Vladislav Bronzov 2024-12-04 14:43:36 +01:00 committed by GitHub
parent c7a109ec81
commit accb7204f9
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
8 changed files with 76 additions and 42 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
src/transformers/models/mistral/modeling_mistral.py
View File

@ -310,9 +310,9 @@ class MistralFlashAttention2(MistralAttention):
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
kv_seq_len = key_states.shape[-2]
if past_key_value is not None:
@ -423,9 +423,9 @@ class MistralSdpaAttention(MistralAttention):
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
cos, sin = self.rotary_emb(value_states, position_ids)
query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)

11
src/transformers/models/qwen2/configuration_qwen2.py
View File

@ -129,6 +129,17 @@ class Qwen2Config(PretrainedConfig):
model_type = "qwen2"
keys_to_ignore_at_inference = ["past_key_values"]
# Default tensor parallel plan for base model `Qwen2`
base_model_tp_plan = {
"layers.*.self_attn.q_proj": "colwise",
"layers.*.self_attn.k_proj": "colwise",
"layers.*.self_attn.v_proj": "colwise",
"layers.*.self_attn.o_proj": "rowwise",
"layers.*.mlp.gate_proj": "colwise",
"layers.*.mlp.up_proj": "colwise",
"layers.*.mlp.down_proj": "rowwise",
}
def __init__(
self,
vocab_size=151936,

19
src/transformers/models/qwen2/modeling_qwen2.py
View File

@ -292,9 +292,9 @@ class Qwen2Attention(nn.Module):
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
if position_embeddings is None:
logger.warning_once(
@ -378,9 +378,9 @@ class Qwen2FlashAttention2(Qwen2Attention):
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
if position_embeddings is None:
logger.warning_once(
@ -502,9 +502,9 @@ class Qwen2SdpaAttention(Qwen2Attention):
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
if position_embeddings is None:
logger.warning_once(
@ -1077,6 +1077,7 @@ class Qwen2Model(Qwen2PreTrainedModel):
class Qwen2ForCausalLM(Qwen2PreTrainedModel, GenerationMixin):
_tied_weights_keys = ["lm_head.weight"]
_tp_plan = {"lm_head": "colwise_rep"}
def __init__(self, config):
super().__init__(config)

11
src/transformers/models/qwen2_moe/configuration_qwen2_moe.py
View File

@ -150,6 +150,17 @@ class Qwen2MoeConfig(PretrainedConfig):
model_type = "qwen2_moe"
keys_to_ignore_at_inference = ["past_key_values"]
# Default tensor parallel plan for base model `Qwen2Moe`
base_model_tp_plan = {
"layers.*.self_attn.q_proj": "colwise",
"layers.*.self_attn.k_proj": "colwise",
"layers.*.self_attn.v_proj": "colwise",
"layers.*.self_attn.o_proj": "rowwise",
"layers.*.mlp.gate_proj": "colwise",
"layers.*.mlp.up_proj": "colwise",
"layers.*.mlp.down_proj": "rowwise",
}
def __init__(
self,
vocab_size=151936,

19
src/transformers/models/qwen2_moe/modeling_qwen2_moe.py
View File

@ -376,9 +376,9 @@ class Qwen2MoeAttention(nn.Module):
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
if position_embeddings is None:
logger.warning_once(
@ -465,9 +465,9 @@ class Qwen2MoeFlashAttention2(Qwen2MoeAttention):
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
if position_embeddings is None:
logger.warning_once(
@ -590,9 +590,9 @@ class Qwen2MoeSdpaAttention(Qwen2MoeAttention):
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
if position_embeddings is None:
logger.warning_once(
@ -1257,6 +1257,7 @@ class Qwen2MoeModel(Qwen2MoePreTrainedModel):
class Qwen2MoeForCausalLM(Qwen2MoePreTrainedModel, GenerationMixin):
_tied_weights_keys = ["lm_head.weight"]
_tp_plan = {"lm_head": "colwise_rep"}
def __init__(self, config):
super().__init__(config)

9
src/transformers/models/starcoder2/configuration_starcoder2.py
View File

@ -134,6 +134,15 @@ class Starcoder2Config(PretrainedConfig):
model_type = "starcoder2"
keys_to_ignore_at_inference = ["past_key_values"]
# Default tensor parallel plan for base model `Starcoder2`
base_model_tp_plan = {
"layers.*.self_attn.q_proj": "colwise",
"layers.*.self_attn.k_proj": "colwise",
"layers.*.self_attn.v_proj": "colwise",
"layers.*.self_attn.o_proj": "rowwise",
"layers.*.mlp.c_fc": "colwise",
"layers.*.mlp.c_proj": "colwise",
}
def __init__(
self,

19
src/transformers/models/starcoder2/modeling_starcoder2.py
View File

@ -271,9 +271,9 @@ class Starcoder2Attention(nn.Module):
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
if position_embeddings is None:
logger.warning_once(
@ -355,9 +355,9 @@ class Starcoder2FlashAttention2(Starcoder2Attention):
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
if position_embeddings is None:
logger.warning_once(
@ -470,9 +470,9 @@ class Starcoder2SdpaAttention(Starcoder2Attention):
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
if position_embeddings is None:
logger.warning_once(
@ -1043,6 +1043,7 @@ class Starcoder2Model(Starcoder2PreTrainedModel):
class Starcoder2ForCausalLM(Starcoder2PreTrainedModel, GenerationMixin):
_tied_weights_keys = ["lm_head.weight"]
_tp_plan = {"lm_head": "colwise_rep"}
def __init__(self, config):
super().__init__(config)

18
src/transformers/models/starcoder2/modular_starcoder2.py
View File

@ -136,9 +136,9 @@ class Starcoder2Attention(nn.Module):
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
if position_embeddings is None:
logger.warning_once(
@ -220,9 +220,9 @@ class Starcoder2FlashAttention2(Starcoder2Attention):
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
if position_embeddings is None:
logger.warning_once(
@ -335,9 +335,9 @@ class Starcoder2SdpaAttention(Starcoder2Attention):
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
if position_embeddings is None:
logger.warning_once(