-
Notifications
You must be signed in to change notification settings - Fork 525
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
* add 30b yaml * Update scripts/train/yamls/finetune/mpt-30b-instruct.yaml --------- Co-authored-by: Vitaliy Chiley <[email protected]>
- Loading branch information
Showing
1 changed file
with
175 additions
and
0 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 |
---|---|---|
@@ -0,0 +1,175 @@ | ||
tokenizer_name: mosaicml/mpt-30b | ||
max_seq_len: 8192 | ||
global_seed: 17 | ||
|
||
# Run Name | ||
run_name: # If left blank, will be read from env var $COMPOSER_RUN_NAME | ||
|
||
# Model | ||
model: | ||
name: hf_causal_lm | ||
pretrained: true | ||
pretrained_model_name_or_path: mosaicml/mpt-30b | ||
init_device: mixed | ||
config_overrides: | ||
max_seq_len: ${max_seq_len} | ||
attn_config: | ||
attn_impl: triton | ||
# Note: we still use packing, but turn this off for memory. | ||
# Initial tests suggest this does not hurt performance too much. | ||
attn_uses_sequence_id: false | ||
|
||
# Tokenizer | ||
tokenizer: | ||
name: ${tokenizer_name} | ||
kwargs: | ||
model_max_length: ${max_seq_len} | ||
|
||
# Dataloaders | ||
train_loader: | ||
name: finetuning | ||
dataset: | ||
hf_name: mosaicml/instruct-v3 | ||
split: train | ||
max_seq_len: ${max_seq_len} | ||
allow_pad_trimming: false | ||
decoder_only_format: true | ||
packing_ratio: 9 | ||
shuffle: true | ||
drop_last: true | ||
num_workers: 8 | ||
pin_memory: false | ||
prefetch_factor: 2 | ||
persistent_workers: true | ||
timeout: 0 | ||
|
||
eval_loader: | ||
name: finetuning | ||
dataset: | ||
hf_name: mosaicml/instruct-v3 | ||
split: test | ||
max_seq_len: ${max_seq_len} | ||
allow_pad_trimming: false | ||
decoder_only_format: true | ||
packing_ratio: 9 | ||
shuffle: true | ||
drop_last: true | ||
num_workers: 8 | ||
pin_memory: false | ||
prefetch_factor: 2 | ||
persistent_workers: true | ||
timeout: 0 | ||
|
||
# Optimization | ||
scheduler: | ||
name: linear_decay_with_warmup # linear no warmup is HF default which dolly used | ||
t_warmup: 50ba | ||
alpha_f: 0 | ||
|
||
optimizer: | ||
# mimic InstructGPT | ||
name: decoupled_adamw | ||
lr: 9.65e-6 | ||
betas: | ||
- 0.9 | ||
- 0.95 | ||
eps: 1.0e-8 | ||
weight_decay: 0 | ||
|
||
algorithms: | ||
gradient_clipping: | ||
clipping_type: norm | ||
clipping_threshold: 1.0 | ||
|
||
max_duration: 8ep | ||
eval_interval: 1ep | ||
eval_first: true | ||
global_train_batch_size: 72 | ||
|
||
# System | ||
seed: ${global_seed} | ||
device_eval_batch_size: 4 | ||
device_train_microbatch_size: 1 | ||
precision: amp_bf16 | ||
|
||
# FSDP | ||
fsdp_config: | ||
sharding_strategy: FULL_SHARD | ||
mixed_precision: PURE | ||
activation_checkpointing: true | ||
activation_checkpointing_reentrant: false | ||
activation_cpu_offload: false | ||
limit_all_gathers: true | ||
sync_module_states: true | ||
state_dict_type: local | ||
verbose: false | ||
|
||
# Logging | ||
progress_bar: false | ||
log_to_console: true | ||
console_log_interval: 20ba | ||
|
||
callbacks: | ||
speed_monitor: | ||
window_size: 10 | ||
runtime_estimator: {} | ||
lr_monitor: {} | ||
|
||
# loggers: | ||
# wandb: {} | ||
|
||
# save_folder: | ||
# save_interval: 3ep | ||
# save_num_checkpoints_to_keep: 1 | ||
|
||
# need to use converted checkpoint with llm-foundry code | ||
# load_path: | ||
autoresume: false | ||
load_weights_only: false | ||
python_log_level: debug | ||
|
||
|
||
icl_max_seq_len: 2048 | ||
|
||
# YOU MUST ADD YOUR OWN DATASET URIs | ||
# this section can be removed if you do not want to track these metrics | ||
icl_tasks: | ||
- | ||
label: piqa | ||
dataset_uri: # ADD YOUR OWN DATASET URI | ||
num_fewshot: | ||
- 0 | ||
batch_size: 4 | ||
max_seq_len: ${icl_max_seq_len} | ||
icl_task_type: multiple_choice | ||
metric_names: | ||
- InContextLearningMultipleChoiceAccuracy | ||
prompt_string: '' # this goes at the beginning of each input | ||
example_delimiter: '\n' # this goes between fewshot examples | ||
continuation_delimiter: ' ' # this separates questions from answers | ||
- | ||
label: hellaswag | ||
dataset_uri: # ADD YOUR OWN DATASET URI | ||
num_fewshot: | ||
- 0 | ||
batch_size: 4 | ||
max_seq_len: ${icl_max_seq_len} | ||
icl_task_type: multiple_choice | ||
metric_names: | ||
- InContextLearningMultipleChoiceAccuracy | ||
prompt_string: '' # this goes at the beginning of each input | ||
example_delimiter: '\n' # this goes between fewshot examples | ||
continuation_delimiter: ' ' # this separates questions from answers | ||
- | ||
label: arc_easy | ||
dataset_uri: # ADD YOUR OWN DATASET URI | ||
num_fewshot: | ||
- 0 | ||
batch_size: 4 | ||
max_seq_len: ${icl_max_seq_len} | ||
icl_task_type: multiple_choice | ||
metric_names: | ||
- InContextLearningMultipleChoiceAccuracy | ||
prompt_string: '' # this goes at the beginning of each input | ||
example_delimiter: '\n' # this goes between fewshot examples | ||
continuation_delimiter: ' ' # this separates questions from answers |