Update kto-4b.yml
Browse files- kto-4b.yml +6 -24
kto-4b.yml
CHANGED
@@ -27,33 +27,13 @@ datasets:
|
|
27 |
- path: NewEden/KTO-IF-Dans
|
28 |
split: train
|
29 |
type: chatml.argilla
|
30 |
-
- path: NewEden/Purpura-Arkhaios-CC-KTO
|
31 |
-
split: train
|
32 |
-
type: chatml.argilla
|
33 |
dataset_prepared_path: last_run_prepared
|
34 |
|
35 |
shuffle_merged_datasets: true
|
36 |
val_set_size: 0.0
|
37 |
output_dir: ./outputs/out
|
38 |
|
39 |
-
|
40 |
-
lora_model_dir:
|
41 |
-
|
42 |
-
lora_r: 64
|
43 |
-
lora_alpha: 32
|
44 |
-
lora_dropout: 0.0
|
45 |
-
lora_target_linear: true
|
46 |
-
lora_fan_in_fan_out:
|
47 |
-
lora_target_modules:
|
48 |
-
- gate_proj
|
49 |
-
- down_proj
|
50 |
-
- up_proj
|
51 |
-
- q_proj
|
52 |
-
- v_proj
|
53 |
-
- k_proj
|
54 |
-
- o_proj
|
55 |
-
|
56 |
-
sequence_len: 32768
|
57 |
sample_packing: false
|
58 |
eval_sample_packing: false
|
59 |
pad_to_sequence_len: false
|
@@ -64,8 +44,8 @@ wandb_watch:
|
|
64 |
wandb_name: kto-1
|
65 |
wandb_log_model:
|
66 |
|
67 |
-
gradient_accumulation_steps:
|
68 |
-
micro_batch_size:
|
69 |
num_epochs: 1
|
70 |
optimizer: paged_adamw_8bit
|
71 |
lr_scheduler:
|
@@ -84,6 +64,8 @@ fp16:
|
|
84 |
tf32: true
|
85 |
|
86 |
gradient_checkpointing: true
|
|
|
|
|
87 |
remove_unused_columns: false
|
88 |
early_stopping_patience:
|
89 |
resume_from_checkpoint:
|
@@ -99,7 +81,7 @@ eval_max_new_tokens:
|
|
99 |
saves_per_epoch: 2
|
100 |
|
101 |
debug:
|
102 |
-
deepspeed:
|
103 |
fsdp:
|
104 |
fsdp_config:
|
105 |
fsdp:
|
|
|
27 |
- path: NewEden/KTO-IF-Dans
|
28 |
split: train
|
29 |
type: chatml.argilla
|
|
|
|
|
|
|
30 |
dataset_prepared_path: last_run_prepared
|
31 |
|
32 |
shuffle_merged_datasets: true
|
33 |
val_set_size: 0.0
|
34 |
output_dir: ./outputs/out
|
35 |
|
36 |
+
sequence_len: 8192
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
37 |
sample_packing: false
|
38 |
eval_sample_packing: false
|
39 |
pad_to_sequence_len: false
|
|
|
44 |
wandb_name: kto-1
|
45 |
wandb_log_model:
|
46 |
|
47 |
+
gradient_accumulation_steps: 16
|
48 |
+
micro_batch_size: 2
|
49 |
num_epochs: 1
|
50 |
optimizer: paged_adamw_8bit
|
51 |
lr_scheduler:
|
|
|
64 |
tf32: true
|
65 |
|
66 |
gradient_checkpointing: true
|
67 |
+
gradient_checkpointing_kwargs:
|
68 |
+
use_reentrant: true
|
69 |
remove_unused_columns: false
|
70 |
early_stopping_patience:
|
71 |
resume_from_checkpoint:
|
|
|
81 |
saves_per_epoch: 2
|
82 |
|
83 |
debug:
|
84 |
+
deepspeed:
|
85 |
fsdp:
|
86 |
fsdp_config:
|
87 |
fsdp:
|