Skip to content
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

Jump to bottom

update grok model param #86

Merged
merged 2 commits into from
May 28, 2024
Merged

update grok model param #86

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
be9c91d
update grok model param
hx89 May 9, 2024
6a54450
add rope
hx89 May 15, 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
26 changes: 22 additions & 4 deletions paxml/tasks/lm/params/nvidia.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -750,12 +750,12 @@ class Grok(NVIDIA1_3B):
DIMS_PER_HEAD = 128
CHECKPOINT_POLICY = layers.AutodiffCheckpointType.SAVE_QKV_OUT_PROJ
MODEL_DIMS = 6144
HIDDEN_DIMS = 49152
HIDDEN_DIMS = 32768
COMBINE_QKV = False
PERCORE_BATCH_SIZE = 1
MAX_SEQ_LEN = 8192
USE_FP8 = False
VOCAB_SIZE = 134144
VOCAB_SIZE = 131072

NUM_EXPERTS = 8
NUM_GROUPS = 256
Expand All @@ -769,6 +769,8 @@ class Grok(NVIDIA1_3B):
ICI_MESH_SHAPE = [1, 1, 8, 1]
DCN_MESH_SHAPE = [1, 32, 1, 1]

USE_ROPE = True

def task(self) -> pax_fiddle.Config[tasks_lib.SingleTask]:
task_p = super().task()
task_p.train.num_train_steps = self.MAX_STEPS
Expand Down Expand Up @@ -812,6 +814,13 @@ def task(self) -> pax_fiddle.Config[tasks_lib.SingleTask]:
mesh_axis_names=['replica', 'data', 'data_expert', 'mdl'],
training_optimized=self.TRAINING_OPTIMIZED_SHARDING,
)

stacked_p = task_p.model.lm_tpl.stacked_transformer_tpl.block
transformer_layer_p = stacked_p.transformer_layer_params_tpl
if self.USE_ROPE:
transformer_layer_p.tr_atten_tpl.use_rotary_position_emb = True
task_p.model.lm_tpl.position_emb_tpl = None

return task_p


Expand All @@ -827,11 +836,11 @@ class Grok_Proxy(NVIDIA1_3B):
DIMS_PER_HEAD = 128
CHECKPOINT_POLICY = layers.AutodiffCheckpointType.SAVE_QKV_OUT_PROJ
MODEL_DIMS = 6144
HIDDEN_DIMS = 49152
HIDDEN_DIMS = 32768
COMBINE_QKV = False
PERCORE_BATCH_SIZE = 1
MAX_SEQ_LEN = 8192
VOCAB_SIZE = 134144
VOCAB_SIZE = 131072
USE_FP8 = False

NUM_EXPERTS = 8
Expand All @@ -846,6 +855,8 @@ class Grok_Proxy(NVIDIA1_3B):
ICI_MESH_SHAPE = [1, 1, 8, 1]
DCN_MESH_SHAPE = [1, 8, 1, 1]

USE_ROPE = True

def task(self) -> pax_fiddle.Config[tasks_lib.SingleTask]:
task_p = super().task()
task_p.train.num_train_steps = self.MAX_STEPS
Expand Down Expand Up @@ -889,4 +900,11 @@ def task(self) -> pax_fiddle.Config[tasks_lib.SingleTask]:
mesh_axis_names=['replica', 'data', 'data_expert', 'mdl'],
training_optimized=self.TRAINING_OPTIMIZED_SHARDING,
)

stacked_p = task_p.model.lm_tpl.stacked_transformer_tpl.block
transformer_layer_p = stacked_p.transformer_layer_params_tpl
if self.USE_ROPE:
transformer_layer_p.tr_atten_tpl.use_rotary_position_emb = True
task_p.model.lm_tpl.position_emb_tpl = None

return task_p