Commit ·
1feb0fe
1
Parent(s): 4c00383
Create axolotl_config.yaml
Browse files- axolotl_config.yaml +53 -0
axolotl_config.yaml
ADDED
|
@@ -0,0 +1,53 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
base_model: /workspace/mistral-11b-base
|
| 2 |
+
model_type: MistralForCausalLM
|
| 3 |
+
tokenizer_type: LlamaTokenizer
|
| 4 |
+
is_mistral_derived_model: true
|
| 5 |
+
|
| 6 |
+
load_in_8bit: false
|
| 7 |
+
load_in_4bit: false
|
| 8 |
+
strict: false
|
| 9 |
+
|
| 10 |
+
datasets:
|
| 11 |
+
- path: Open-Orca/SlimOrca
|
| 12 |
+
type: sharegpt
|
| 13 |
+
conversation: mistral
|
| 14 |
+
|
| 15 |
+
dataset_prepared_path:
|
| 16 |
+
val_set_size: 0.005
|
| 17 |
+
output_dir: ./out
|
| 18 |
+
|
| 19 |
+
sequence_len: 8192
|
| 20 |
+
sample_packing: true
|
| 21 |
+
pad_to_sequence_len: true
|
| 22 |
+
|
| 23 |
+
wandb_project: mistral-11b-noresidual
|
| 24 |
+
wandb_entity:
|
| 25 |
+
wandb_watch:
|
| 26 |
+
wandb_name:
|
| 27 |
+
wandb_log_model:
|
| 28 |
+
|
| 29 |
+
gradient_accumulation_steps: 16
|
| 30 |
+
micro_batch_size: 1
|
| 31 |
+
num_epochs: 2
|
| 32 |
+
optimizer: adamw_bnb_8bit
|
| 33 |
+
lr_scheduler: cosine
|
| 34 |
+
learning_rate: 0.000005
|
| 35 |
+
|
| 36 |
+
train_on_inputs: true
|
| 37 |
+
group_by_length: false
|
| 38 |
+
bf16: true
|
| 39 |
+
fp16: false
|
| 40 |
+
tf32: false
|
| 41 |
+
|
| 42 |
+
gradient_checkpointing: true
|
| 43 |
+
flash_attention: true
|
| 44 |
+
|
| 45 |
+
logging_steps: 1
|
| 46 |
+
warmup_steps: 10
|
| 47 |
+
eval_steps: 100
|
| 48 |
+
save_steps: 100
|
| 49 |
+
weight_decay: 0.0
|
| 50 |
+
special_tokens:
|
| 51 |
+
bos_token: "<s>"
|
| 52 |
+
eos_token: "</s>"
|
| 53 |
+
unk_token: "<unk>"
|