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.

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

[metax] add llama3-70B flagscale continue train case #758

Merged
merged 3 commits into from
Oct 5, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
Original file line number Diff line number Diff line change
Expand Up @@ -176,7 +176,7 @@ def replace_yamls(scale_home, config_module, args):
print("Maybe some errors")
if len(info_line) == getattr(module, "steps"):
break
time.sleep(300)
time.sleep(300) if args.vendor != 'metax' else time.sleep(10)

infos = []
for line in info_line:
Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
此测例为FlagScale相关项目测例
Original file line number Diff line number Diff line change
@@ -0,0 +1,25 @@
# scale_parent must under FlagPerf/ or data_dir/, otherwise you cannot mount it into baremetal, therefore cannot use shared storage
scale_parent = "/share/project/FlagPerf/FlagScale/build/metax_C500"
scale_home = f"{scale_parent}/FlagScale"

# this cmd should install scale at <scale_home>. <scale_home> is set by flagperf.training.benchmarks.llava1.5_7b.flagscale.run_pretraining.py
scale_download_cmd = f"cd {scale_parent}"

# NV need nothing because all requirements have been established in base docker image. vendor can do anything related here
scale_install_cmd = ""

scale_conf_dir = f"{scale_home}/examples/llama/conf"
configyaml = f"{scale_conf_dir}/config.yaml"
trainyaml = f"{scale_conf_dir}/train/train_llama3_70b_finetune.yaml"
dataset = f"SAMPLE50B/llama3/llama3_dataset"
tokenizer = f"SAMPLE50B/llama3/llama3_tokenizer"
ckpt = f"llama3_ckpt"

cmds = {"before_start": ""}
# flagscale's requirements
flagscale_chip_type = "C500"
flagscale_ssh_port = 1234
flops = -1

# for llava's algorithm
steps = 500
1 change: 1 addition & 0 deletions training/run_benchmarks/config/test_conf.py
Original file line number Diff line number Diff line change
Expand Up @@ -174,6 +174,7 @@
# "llama2_7b:deepspeed:S4000:1:8:1": "/data/flagperf/llama/openwebtext",

# metax cases
#"llama3_70B_continuetrain:flagscale:C500:4:8:1": "/metax/dataset"
#"llava1.5_7b_continuetrain:flagscale:C500:4:8:1": "/data/dataset/llava"
#"llama3_8B:megatron_core060:C500:1:8:1": "/data/llama3_8b"
# "llama2_70B:megatron:C500:4:8:1": "/data/llama2-70B"
Expand Down