Skip to content

Commit e4307fb

Browse files
authored
feat: add examples for deepcoder (axolotl-ai-cloud#2517)
1 parent dd8bad0 commit e4307fb

File tree

1 file changed

+58
-0
lines changed

1 file changed

+58
-0
lines changed
Lines changed: 58 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,58 @@
1+
base_model: agentica-org/DeepCoder-14B-Preview
2+
# Automatically upload checkpoint and final model to HF
3+
# hub_model_id: username/custom_model_name
4+
5+
load_in_8bit: true
6+
load_in_4bit: false
7+
strict: false
8+
9+
datasets:
10+
- path: fozziethebeat/alpaca_messages_2k_test
11+
type: chat_template
12+
field_messages: messages
13+
message_property_mappings:
14+
role: role
15+
content: content
16+
17+
dataset_prepared_path:
18+
val_set_size: 0.05
19+
output_dir: ./outputs/lora-out
20+
21+
sequence_len: 4096
22+
sample_packing: true
23+
eval_sample_packing: false
24+
pad_to_sequence_len: true
25+
26+
adapter: lora
27+
lora_model_dir:
28+
lora_r: 32
29+
lora_alpha: 16
30+
lora_dropout: 0.05
31+
lora_target_linear: true
32+
33+
wandb_project:
34+
wandb_entity:
35+
wandb_watch:
36+
wandb_name:
37+
wandb_log_model:
38+
39+
gradient_accumulation_steps: 2
40+
micro_batch_size: 2
41+
num_epochs: 4
42+
optimizer: adamw_bnb_8bit
43+
lr_scheduler: cosine
44+
learning_rate: 0.0002
45+
46+
bf16: auto
47+
tf32: true
48+
49+
gradient_checkpointing: true
50+
resume_from_checkpoint:
51+
logging_steps: 1
52+
flash_attention: true
53+
54+
warmup_steps: 10
55+
evals_per_epoch: 1
56+
saves_per_epoch: 1
57+
weight_decay: 0.0
58+
special_tokens:

0 commit comments

Comments
 (0)