-
Notifications
You must be signed in to change notification settings - Fork 47
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Merge branch 'feat/internlm2' into v0.2.3-internlm2
- Loading branch information
Showing
12 changed files
with
1,406 additions
and
85 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
|
@@ -128,6 +128,7 @@ aim_logs/ | |
nvmelogs/ | ||
run_backup/ | ||
runs/ | ||
RUN/ | ||
runs_bak/ | ||
LLM_ALERT | ||
small_demo/ | ||
|
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,37 @@ | ||
# Copyright (c) InternLM. All rights reserved. | ||
|
||
cudnn_deterministic = False | ||
cudnn_benchmark = False | ||
|
||
enable_tb = True | ||
|
||
grad_profiling = dict( | ||
# calculate layer norms and parameter norms, and show them on tensorboard | ||
grad_norm_profiling=False, | ||
# count zero gradients, and show them on tensorboard | ||
zero_grad_profiling=False, | ||
# [optional] layers displayed on tensorboard, default: layers=["ScaleColumnParallelLinear"] | ||
# if not set, display all layers | ||
layers=["ScaleColumnParallelLinear"], | ||
vocab_grad_norm_profiling=False, | ||
interval_steps=5, | ||
) | ||
|
||
grad_scaler = dict( | ||
fp16=dict( | ||
# the initial loss scale, defaults to 2**16 | ||
initial_scale=2**16, | ||
# the minimum loss scale, defaults to None | ||
min_scale=1, | ||
# the number of steps to increase loss scale when no overflow occurs | ||
growth_interval=1000, | ||
), | ||
# the multiplication factor for increasing loss scale, defaults to 2 | ||
growth_factor=2, | ||
# the multiplication factor for decreasing loss scale, defaults to 0.5 | ||
backoff_factor=0.5, | ||
# the maximum loss scale, defaults to None | ||
max_scale=2**24, | ||
# the number of overflows before decreasing loss scale, defaults to 2 | ||
hysteresis=2, | ||
) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,59 @@ | ||
# Copyright (c) InternLM. All rights reserved. | ||
|
||
model_type = "INTERNLM2" | ||
|
||
VOCAB_SIZE = 92544 | ||
HIDDEN_SIZE = 6144 | ||
NUM_ATTENTION_HEAD = 48 | ||
NUM_KV_ATTENTION_HEAD = 8 | ||
MLP_RATIO = 8 / 3 | ||
NUM_LAYER = 48 | ||
|
||
model = dict( | ||
num_chunks=1, | ||
checkpoint=1.0, | ||
dtype="torch.bfloat16", | ||
embed_split_hidden=True, | ||
num_layers=NUM_LAYER, | ||
hidden_size=HIDDEN_SIZE, | ||
vocab_size=VOCAB_SIZE, | ||
embed_grad_scale=1, | ||
parallel_output=True, | ||
num_attention_heads=NUM_ATTENTION_HEAD, | ||
num_kv_attention_heads=NUM_KV_ATTENTION_HEAD, | ||
mlp_ratio=MLP_RATIO, | ||
norm_type="rmsnorm", | ||
adapt_hf=True, | ||
apply_post_layer_norm=False, | ||
no_bias=True, | ||
layer_norm_epsilon=1e-5, | ||
rope_base=1000000, | ||
) | ||
|
||
hybrid_zero_optimizer = dict( | ||
# Enable low_level_optimzer overlap_communication | ||
overlap_sync_grad=True, | ||
overlap_sync_param=False, | ||
# bucket size for nccl communication params | ||
reduce_bucket_size=512 * 1024 * 1024, | ||
# grad clipping | ||
clip_grad_norm=1.0, | ||
) | ||
|
||
# zero1 parallel: | ||
# 1. if zero1 <= 0, The size of the zero process group is equal to the size of the dp process group, | ||
# so parameters will be divided within the range of dp. | ||
# 2. if zero1 == 1, zero is not used, and all dp groups retain the full amount of model parameters. | ||
# 3. zero1 > 1 and zero1 <= dp world size, the world size of zero is a subset of dp world size. | ||
# For smaller models, it is usually a better choice to split the parameters within nodes with a setting <= 8. | ||
# 4. fsdp: bool, whether to use fsdp in pytorch, which can be a subsitution of ZeRO1. | ||
# pipeline parallel (dict): | ||
# 1. size: int, the size of pipeline parallel. | ||
# 2. interleaved_overlap: bool, enable/disable communication overlap when using interleaved pipeline scheduler. | ||
# tensor parallel: tensor parallel size, usually the number of GPUs per node. | ||
parallel = dict( | ||
zero1=dict(size=16, fsdp=False), | ||
tensor=2, | ||
pipeline=dict(size=1, interleaved_overlap=True), | ||
sequence_parallel=True, | ||
) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,59 @@ | ||
# Copyright (c) InternLM. All rights reserved. | ||
|
||
model_type = "INTERNLM2" | ||
|
||
VOCAB_SIZE = 92544 | ||
HIDDEN_SIZE = 4096 | ||
NUM_ATTENTION_HEAD = 32 | ||
NUM_KV_ATTENTION_HEAD = 8 | ||
MLP_RATIO = 3.5 | ||
NUM_LAYER = 32 | ||
|
||
model = dict( | ||
num_chunks=1, | ||
checkpoint=0.2, | ||
dtype="torch.bfloat16", | ||
embed_split_hidden=True, | ||
num_layers=NUM_LAYER, | ||
hidden_size=HIDDEN_SIZE, | ||
vocab_size=VOCAB_SIZE, | ||
embed_grad_scale=1, | ||
parallel_output=True, | ||
num_attention_heads=NUM_ATTENTION_HEAD, | ||
num_kv_attention_heads=NUM_KV_ATTENTION_HEAD, | ||
mlp_ratio=MLP_RATIO, | ||
norm_type="rmsnorm", | ||
adapt_hf=False, | ||
apply_post_layer_norm=False, | ||
no_bias=True, | ||
layer_norm_epsilon=1e-5, | ||
rope_base=1000000, | ||
) | ||
|
||
hybrid_zero_optimizer = dict( | ||
# Enable low_level_optimzer overlap_communication | ||
overlap_sync_grad=True, | ||
overlap_sync_param=False, | ||
# bucket size for nccl communication params | ||
reduce_bucket_size=512 * 1024 * 1024, | ||
# grad clipping | ||
clip_grad_norm=1.0, | ||
) | ||
|
||
# zero1 parallel: | ||
# 1. if zero1 <= 0, The size of the zero process group is equal to the size of the dp process group, | ||
# so parameters will be divided within the range of dp. | ||
# 2. if zero1 == 1, zero is not used, and all dp groups retain the full amount of model parameters. | ||
# 3. zero1 > 1 and zero1 <= dp world size, the world size of zero is a subset of dp world size. | ||
# For smaller models, it is usually a better choice to split the parameters within nodes with a setting <= 8. | ||
# 4. fsdp: bool, whether to use fsdp in pytorch, which can be a subsitution of ZeRO1. | ||
# pipeline parallel (dict): | ||
# 1. size: int, the size of pipeline parallel. | ||
# 2. interleaved_overlap: bool, enable/disable communication overlap when using interleaved pipeline scheduler. | ||
# tensor parallel: tensor parallel size, usually the number of GPUs per node. | ||
parallel = dict( | ||
zero1=dict(size=8, fsdp=False), | ||
tensor=1, | ||
pipeline=dict(size=1, interleaved_overlap=True), | ||
sequence_parallel=False, | ||
) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,54 @@ | ||
# Copyright (c) InternLM. All rights reserved. | ||
|
||
model_type = "INTERNLM" | ||
|
||
VOCAB_SIZE = 103168 | ||
HIDDEN_SIZE = 5120 | ||
NUM_ATTENTION_HEAD = 40 | ||
MLP_RATIO = 8 / 3 | ||
NUM_LAYER = 60 | ||
|
||
model = dict( | ||
num_chunks=1, | ||
checkpoint=False, | ||
dtype="torch.bfloat16", | ||
embed_split_hidden=True, | ||
num_layers=NUM_LAYER, | ||
hidden_size=HIDDEN_SIZE, | ||
vocab_size=VOCAB_SIZE, | ||
embed_grad_scale=1, | ||
parallel_output=True, | ||
num_attention_heads=NUM_ATTENTION_HEAD, | ||
mlp_ratio=MLP_RATIO, | ||
norm_type="rmsnorm", | ||
apply_post_layer_norm=False, | ||
layer_norm_epsilon=1e-5, | ||
) | ||
|
||
hybrid_zero_optimizer = dict( | ||
# Enable overlap_communication | ||
overlap_sync_grad=True, | ||
overlap_sync_param=False, | ||
# bucket size for nccl communication params | ||
reduce_bucket_size=512 * 1024 * 1024, | ||
# grad clipping | ||
clip_grad_norm=1.0, | ||
) | ||
|
||
# zero1 parallel: | ||
# 1. if zero1 <= 0, The size of the zero process group is equal to the size of the dp process group, | ||
# so parameters will be divided within the range of dp. | ||
# 2. if zero1 == 1, zero is not used, and all dp groups retain the full amount of model parameters. | ||
# 3. zero1 > 1 and zero1 <= dp world size, the world size of zero is a subset of dp world size. | ||
# For smaller models, it is usually a better choice to split the parameters within nodes with a setting <= 8. | ||
# 4. fsdp: bool, whether to use fsdp in pytorch, which can be a subsitution of ZeRO1. | ||
# pipeline parallel (dict): | ||
# 1. size: int, the size of pipeline parallel. | ||
# 2. interleaved_overlap: bool, enable/disable communication overlap when using interleaved pipeline scheduler. | ||
# tensor parallel: tensor parallel size, usually the number of GPUs per node. | ||
parallel = dict( | ||
zero1=dict(size=8, fsdp=False), | ||
tensor=4, | ||
pipeline=dict(size=1, interleaved_overlap=True), | ||
sequence_parallel=False, | ||
) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,54 @@ | ||
# Copyright (c) InternLM. All rights reserved. | ||
|
||
model_type = "INTERNLM" | ||
|
||
VOCAB_SIZE = 103168 | ||
HIDDEN_SIZE = 4096 | ||
NUM_ATTENTION_HEAD = 32 | ||
MLP_RATIO = 8 / 3 | ||
NUM_LAYER = 32 | ||
|
||
model = dict( | ||
num_chunks=1, | ||
checkpoint=False, | ||
dtype="torch.bfloat16", | ||
embed_split_hidden=True, | ||
num_layers=NUM_LAYER, | ||
hidden_size=HIDDEN_SIZE, | ||
vocab_size=VOCAB_SIZE, | ||
embed_grad_scale=1, | ||
parallel_output=True, | ||
num_attention_heads=NUM_ATTENTION_HEAD, | ||
mlp_ratio=MLP_RATIO, | ||
norm_type="rmsnorm", | ||
apply_post_layer_norm=False, | ||
layer_norm_epsilon=1e-5, | ||
) | ||
|
||
hybrid_zero_optimizer = dict( | ||
# Enable overlap_communication | ||
overlap_sync_grad=True, | ||
overlap_sync_param=False, | ||
# bucket size for nccl communication params | ||
reduce_bucket_size=512 * 1024 * 1024, | ||
# grad clipping | ||
clip_grad_norm=1.0, | ||
) | ||
|
||
# zero1 parallel: | ||
# 1. if zero1 <= 0, The size of the zero process group is equal to the size of the dp process group, | ||
# so parameters will be divided within the range of dp. | ||
# 2. if zero1 == 1, zero is not used, and all dp groups retain the full amount of model parameters. | ||
# 3. zero1 > 1 and zero1 <= dp world size, the world size of zero is a subset of dp world size. | ||
# For smaller models, it is usually a better choice to split the parameters within nodes with a setting <= 8. | ||
# 4. fsdp: bool, whether to use fsdp in pytorch, which can be a subsitution of ZeRO1. | ||
# pipeline parallel (dict): | ||
# 1. size: int, the size of pipeline parallel. | ||
# 2. interleaved_overlap: bool, enable/disable communication overlap when using interleaved pipeline scheduler. | ||
# tensor parallel: tensor parallel size, usually the number of GPUs per node. | ||
parallel = dict( | ||
zero1=dict(size=8, fsdp=False), | ||
tensor=1, | ||
pipeline=dict(size=1, interleaved_overlap=True), | ||
sequence_parallel=False, | ||
) |
Oops, something went wrong.