Skip to content

update MI355 yaml for v26.5 #893

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

Open
vidushi8 wants to merge 2 commits into
base: main
Choose a base branch
from
Open

update MI355 yaml for v26.5 #893

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5544283
update MI355 yaml for v26.5
vidushi8 Jul 20, 2026
f33111c
Update llama2_70B-BF16-pretrain.yaml
vidushi8 Jul 21, 2026
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
2 changes: 1 addition & 1 deletion examples/megatron/configs/MI355X/gpt_oss_20B-BF16-pretrain.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,7 +46,7 @@ modules:
profile_step_start: 6

# hyper parameters
train_iters: 10
train_iters: 50
micro_batch_size: 8
global_batch_size: 512
seq_length: ${PRIMUS_SEQ_LENGTH:4096}
Expand Down
7 changes: 4 additions & 3 deletions examples/megatron/configs/MI355X/gpt_oss_20B-FP8-pretrain.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,9 +46,9 @@ modules:
profile_step_start: 6

# hyper parameters
train_iters: 10
micro_batch_size: 8
global_batch_size: 512
train_iters: 50
micro_batch_size: 6
global_batch_size: 48
seq_length: ${PRIMUS_SEQ_LENGTH:4096}
max_position_embeddings: ${PRIMUS_MAX_POSITION_EMBEDDINGS:4096}
lr: 1.0e-5
Expand Down Expand Up @@ -106,6 +106,7 @@ modules:
eval_iters: 0

cross_entropy_loss_fusion: true
cross_entropy_fusion_impl: "te"

fp8: hybrid
# enable_primus_turbo: true
Expand Down
4 changes: 2 additions & 2 deletions examples/megatron/configs/MI355X/llama2_70B-BF16-pretrain.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,8 +17,8 @@ modules:
log_avg_reset_interval: 50

train_iters: 50
micro_batch_size: 14
global_batch_size: 224
micro_batch_size: 13
global_batch_size: 104

seq_length: 4096
max_position_embeddings: 4096
Expand Down
4 changes: 2 additions & 2 deletions examples/megatron/configs/MI355X/llama3.1_70B-FP8-pretrain.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,8 +17,8 @@ modules:
log_avg_reset_interval: 50

train_iters: 50
micro_batch_size: 4
global_batch_size: 32
micro_batch_size: 3
global_batch_size: 24

seq_length: 8192
max_position_embeddings: 8192
Expand Down
2 changes: 1 addition & 1 deletion examples/megatron/configs/MI355X/qwen3_30B_A3B-BF16-pretrain.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,7 @@ modules:
log_avg_reset_interval: 50

# hyper parameters
train_iters: 10
train_iters: 50
micro_batch_size: 8
global_batch_size: 512
seq_length: ${PRIMUS_SEQ_LENGTH:4096}
Expand Down
6 changes: 3 additions & 3 deletions examples/megatron/configs/MI355X/qwen3_30B_A3B-FP8-pretrain.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,9 +21,9 @@ modules:
log_avg_reset_interval: 50

# hyper parameters
train_iters: 10
micro_batch_size: 8
global_batch_size: 512
train_iters: 50
micro_batch_size: 6
global_batch_size: 48
seq_length: ${PRIMUS_SEQ_LENGTH:4096}
max_position_embeddings: ${PRIMUS_MAX_POSITION_EMBEDDINGS:4096}
lr: 1.0e-5
Expand Down
2 changes: 1 addition & 1 deletion examples/mlperf/gpt_oss_20b/config_MI355X_1x8x1_tp1pp1ep1_gbs32.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,7 +61,7 @@ export HSA_FORCE_FINE_GRAIN_PCIE=1
export HSA_KERNARG_POOL_SIZE=12582912
export TORCH_NCCL_HIGH_PRIORITY=1
export ENABLE_NUMA_BINDING=1
export PYTORCH_ALLOC_CONF=expandable_segments:True
export PYTORCH_ALLOC_CONF=expandable_segments:False
export HSA_NO_SCRATCH_RECLAIM=1
export HSA_ENABLE_SDMA=1
export HSA_ENABLE_INTERRUPT=0
Expand Down
4 changes: 2 additions & 2 deletions examples/mlperf/gpt_oss_20b/configs/MI355/gpt_oss_20B-FP8-mlperf-pretrain.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,8 +47,8 @@ modules:
# log_avg_reset_interval: 50

# profile
profile: ${PRIMUS_PROFILE:false}
use_pytorch_profiler: ${PRIMUS_PROFILE:false}
profile: ${PRIMUS_PROFILE:False}
use_pytorch_profiler: ${PRIMUS_PROFILE:False}
Comment on lines +50 to +51
profile_step_end: ${PRIMUS_PROFILE_STEP_END:32}
profile_step_start: ${PRIMUS_PROFILE_STEP_START:16}
profile_ranks: [0,1,2,3,4,5,6,7]
Expand Down
2 changes: 1 addition & 1 deletion examples/torchtitan/configs/MI355X/deepseek_v3_16b-BF16-pretrain.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,7 @@ modules:

training:
debug_moe_force_load_balance: true
local_batch_size: 13
local_batch_size: 9
seq_len: 4096
max_norm: 1.0 # grad norm clipping
steps: 50
Expand Down
Loading