Skip to content

nemo-ux: Use kv_channels to enable cases where head_dim != hidden_size // head… #9994

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
cuichenx merged 3 commits into from
Aug 7, 2024
Merged

nemo-ux: Use kv_channels to enable cases where head_dim != hidden_size // head… #9994

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
b6bee06
Use kv_channels to enable cases where head_dim != hidden_size // head…
akoumpa Jul 31, 2024
794513e
Add head_dim to exporter
akoumpa Aug 5, 2024
0de127c
Drop default values for kv_channels
akoumpa Aug 6, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 4 additions & 2 deletions nemo/collections/llm/gpt/model/mistral.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,6 +111,7 @@ def make_vocab_size_divisible_by(mistral_vocab_size):
num_layers=source.num_hidden_layers,
hidden_size=source.hidden_size,
ffn_hidden_size=source.intermediate_size,
kv_channels=source.get('head_dim', source.hidden_size // source.num_attention_heads),
num_attention_heads=source.num_attention_heads,
# max_position_embeddings=source.max_position_embeddings,
init_method_std=source.initializer_range,
Expand Down Expand Up @@ -183,6 +184,7 @@ def config(self) -> "MistralConfig":
num_key_value_heads=source.num_query_groups,
rope_theta=source.rotary_base,
vocab_size=self.tokenizer.vocab_size,
head_dim=source.kv_channels,
)


Expand All @@ -202,7 +204,7 @@ def _import_qkv(ctx: io.TransformCTX, q, k, v):
heads_per_group = head_num // num_query_groups
hidden_size = megatron_config.hidden_size
head_num = megatron_config.num_attention_heads
head_size = hidden_size // head_num
head_size = megatron_config.kv_channels

old_tensor_shape = q.size()
new_q_tensor_shape = (head_num, head_size) + old_tensor_shape[1:]
Expand Down Expand Up @@ -244,7 +246,7 @@ def _export_qkv(ctx: io.TransformCTX, linear_qkv):
heads_per_group = head_num // num_query_groups
hidden_size = megatron_config.hidden_size
head_num = megatron_config.num_attention_heads
head_size = hidden_size // head_num
head_size = megatron_config.kv_channels
qkv_total_dim = head_num + 2 * num_query_groups

linear_qkv = linear_qkv.reshape([qkv_total_dim, head_size, hidden_size])
Expand Down
6 changes: 4 additions & 2 deletions nemo/collections/llm/gpt/model/mixtral.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -117,6 +117,7 @@ def config(self) -> MixtralConfig8x7B:
num_layers=config.num_hidden_layers,
hidden_size=config.hidden_size,
ffn_hidden_size=config.intermediate_size,
kv_channels=config.get('head_dim', config.hidden_size // config.num_attention_heads),
max_position_embeddings=config.max_position_embeddings, # TODO
seq_length=config.max_position_embeddings,
# RoPE
Expand Down Expand Up @@ -158,7 +159,7 @@ def _import_qkv(ctx: io.TransformCTX, q, k, v):
heads_per_group = head_num // num_query_groups
hidden_size = megatron_config.hidden_size
head_num = megatron_config.num_attention_heads
head_size = hidden_size // head_num
head_size = megatron_config.kv_channels

old_tensor_shape = q.size()
new_q_tensor_shape = (head_num, head_size) + old_tensor_shape[1:]
Expand Down Expand Up @@ -262,6 +263,7 @@ def config(self) -> "MixtralConfig":
initializer_range=source.init_method_std,
# vocab
vocab_size=self.tokenizer.vocab_size,
head_dim=source.kv_channels,
)


Expand All @@ -281,7 +283,7 @@ def _export_qkv(ctx: io.TransformCTX, linear_qkv):
heads_per_group = head_num // num_query_groups
hidden_size = megatron_config.hidden_size
head_num = megatron_config.num_attention_heads
head_size = hidden_size // head_num
head_size = megatron_config.kv_channels
qkv_total_dim = head_num + 2 * num_query_groups

linear_qkv = linear_qkv.reshape([qkv_total_dim, head_size, hidden_size])
Expand Down