Upload folder using huggingface_hub
Browse files
configs/dolphin-mistral-7b.yml
ADDED
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
base_model: mistralai/Mistral-7B-v0.1
|
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: /workspace/datasets/dolphin/dolphin201.jsonl
|
12 |
+
type: alpaca_w_system.load_open_orca_chatml
|
13 |
+
- path: /workspace/datasets/WizardLM_evol_instruct_cleaned.jsonl
|
14 |
+
type: sharegpt
|
15 |
+
conversation: chatml
|
16 |
+
- path: /workspace/datasets/not_samantha_norefusals.jsonl
|
17 |
+
type: sharegpt
|
18 |
+
conversation: chatml
|
19 |
+
dataset_prepared_path: last_run_prepared
|
20 |
+
val_set_size: 0
|
21 |
+
output_dir: /workspace/dolphin-2.2-mistral-7b
|
22 |
+
|
23 |
+
sequence_len: 8192
|
24 |
+
sample_packing: true
|
25 |
+
pad_to_sequence_len: true
|
26 |
+
|
27 |
+
wandb_project: dolphin
|
28 |
+
wandb_entity:
|
29 |
+
wandb_watch:
|
30 |
+
wandb_run_id:
|
31 |
+
wandb_log_model:
|
32 |
+
|
33 |
+
gradient_accumulation_steps: 4
|
34 |
+
micro_batch_size: 5
|
35 |
+
num_epochs: 4
|
36 |
+
adam_beta2: 0.95
|
37 |
+
adam_epsilon: 0.00001
|
38 |
+
max_grad_norm: 1.0
|
39 |
+
lr_scheduler: cosine
|
40 |
+
learning_rate: 0.000006
|
41 |
+
|
42 |
+
train_on_inputs: false
|
43 |
+
group_by_length: false
|
44 |
+
bf16: true
|
45 |
+
fp16: false
|
46 |
+
tf32: false
|
47 |
+
|
48 |
+
gradient_checkpointing: true
|
49 |
+
early_stopping_patience:
|
50 |
+
resume_from_checkpoint:
|
51 |
+
local_rank:
|
52 |
+
logging_steps: 1
|
53 |
+
xformers_attention:
|
54 |
+
flash_attention: true
|
55 |
+
|
56 |
+
warmup_steps: 100
|
57 |
+
eval_steps:
|
58 |
+
eval_table_size:
|
59 |
+
eval_table_max_new_tokens:
|
60 |
+
eval_sample_packing: false
|
61 |
+
save_steps: 0.25
|
62 |
+
debug:
|
63 |
+
deepspeed: deepspeed/zero2.json
|
64 |
+
weight_decay: 0.1
|
65 |
+
fsdp:
|
66 |
+
fsdp_config:
|
67 |
+
special_tokens:
|
68 |
+
eos_token: "<|im_end|>"
|
69 |
+
tokens:
|
70 |
+
- "<|im_start|>"
|
71 |
+
- "<|im_end|>"
|