Crystalcareai
commited on
Commit
•
0470c5b
1
Parent(s):
2584bd2
Create axolotl.yml
Browse files- axolotl.yml +118 -0
axolotl.yml
ADDED
@@ -0,0 +1,118 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
base_model: unsloth/Phi-3-medium-4k-instruct
|
2 |
+
model_type: AutoModelForCausalLM
|
3 |
+
tokenizer_type: AutoTokenizer
|
4 |
+
|
5 |
+
trust_remote_code: true
|
6 |
+
|
7 |
+
# load_in_8bit: true
|
8 |
+
load_in_4bit: true
|
9 |
+
# strict: false
|
10 |
+
|
11 |
+
datasets:
|
12 |
+
- path: /workspace/datasets/dolphin-2.9.2/dolphin201-sharegpt2.jsonl
|
13 |
+
type: sharegpt
|
14 |
+
conversation: chatml
|
15 |
+
- path: /workspace/datasets/dolphin-2.9.2/dolphin-coder-codegen-sharegpt2.jsonl
|
16 |
+
type: sharegpt
|
17 |
+
conversation: chatml
|
18 |
+
- path: /workspace/datasets/dolphin-2.9.2/dolphin-coder-translate-sharegpt2.jsonl
|
19 |
+
type: sharegpt
|
20 |
+
conversation: chatml
|
21 |
+
- path: /workspace/datasets/dolphin-2.9.2/m-a-p_Code-Feedback-sharegpt-unfiltered.jsonl
|
22 |
+
type: sharegpt
|
23 |
+
conversation: chatml
|
24 |
+
- path: /workspace/datasets/dolphin-2.9.2/m-a-p_CodeFeedback-Filtered-Instruction-sharegpt-unfiltered.jsonl
|
25 |
+
type: sharegpt
|
26 |
+
conversation: chatml
|
27 |
+
- path: /workspace/datasets/dolphin-2.9.2/not_samantha_norefusals.jsonl
|
28 |
+
type: sharegpt
|
29 |
+
conversation: chatml
|
30 |
+
- path: /workspace/datasets/dolphin-2.9.2/openhermes200k_unfiltered.jsonl
|
31 |
+
type: sharegpt
|
32 |
+
conversation: chatml
|
33 |
+
- path: /workspace/datasets/dolphin-2.9.2/Orca-Math-resort-unfiltered.jsonl
|
34 |
+
type: sharegpt
|
35 |
+
conversation: chatml
|
36 |
+
- path: /workspace/datasets/dolphin-2.9.2/SystemChat_sharegpt.jsonl
|
37 |
+
type: sharegpt
|
38 |
+
conversation: chatml
|
39 |
+
- path: /workspace/datasets/dolphin-2.9.2/toolbench_instruct_j1s1_3k_unfiltered.jsonl
|
40 |
+
type: sharegpt
|
41 |
+
conversation: chatml
|
42 |
+
- path: /workspace/datasets/dolphin-2.9.2/toolbench_negative_unfiltered.jsonl
|
43 |
+
type: sharegpt
|
44 |
+
conversation: chatml
|
45 |
+
- path: /workspace/datasets/dolphin-2.9.2/toolbench_react_10p_unfiltered.jsonl
|
46 |
+
type: sharegpt
|
47 |
+
conversation: chatml
|
48 |
+
- path: /workspace/datasets/dolphin-2.9.2/toolbench_tflan_cot_30p_unfiltered.jsonl
|
49 |
+
type: sharegpt
|
50 |
+
conversation: chatml
|
51 |
+
- path: /workspace/datasets/dolphin-2.9.2/agent_instruct_react_unfiltered.jsonl
|
52 |
+
type: sharegpt
|
53 |
+
conversation: chatml
|
54 |
+
|
55 |
+
|
56 |
+
chat_template: chatml
|
57 |
+
dataset_prepared_path: phi-14-data
|
58 |
+
val_set_size: 0.01
|
59 |
+
output_dir: phi-14
|
60 |
+
|
61 |
+
sequence_len: 4096 # supports up to 8192
|
62 |
+
sample_packing: true
|
63 |
+
pad_to_sequence_len: true
|
64 |
+
|
65 |
+
adapter: qlora
|
66 |
+
lora_model_dir:
|
67 |
+
lora_r: 32
|
68 |
+
lora_alpha: 16
|
69 |
+
lora_dropout: 0.05
|
70 |
+
lora_modules_to_save: [embed_tokens, lm_head]
|
71 |
+
lora_target_linear: true
|
72 |
+
lora_fan_in_fan_out:
|
73 |
+
|
74 |
+
wandb_project: Phi-3-14b
|
75 |
+
wandb_entity:
|
76 |
+
wandb_watch:
|
77 |
+
wandb_name:
|
78 |
+
wandb_log_model:
|
79 |
+
|
80 |
+
gradient_accumulation_steps: 4
|
81 |
+
micro_batch_size: 4
|
82 |
+
num_epochs: 3
|
83 |
+
optimizer: adamw_torch_fused
|
84 |
+
lr_scheduler: cosine
|
85 |
+
learning_rate: 4e-4
|
86 |
+
|
87 |
+
train_on_inputs: false
|
88 |
+
group_by_length: false
|
89 |
+
bf16: auto
|
90 |
+
fp16:
|
91 |
+
tf32: false
|
92 |
+
|
93 |
+
gradient_checkpointing: true
|
94 |
+
early_stopping_patience:
|
95 |
+
resume_from_checkpoint:
|
96 |
+
local_rank:
|
97 |
+
logging_steps: 1
|
98 |
+
xformers_attention:
|
99 |
+
flash_attention: true
|
100 |
+
|
101 |
+
warmup_steps: 10
|
102 |
+
# evals_per_epoch: 2
|
103 |
+
# eval_table_size:
|
104 |
+
# eval_max_new_tokens: 128
|
105 |
+
saves_per_epoch: 2
|
106 |
+
save_total_limit: 1
|
107 |
+
debug:
|
108 |
+
deepspeed: /workspace/axolotl/deepspeed_configs/zero3_bf16.json
|
109 |
+
weight_decay: 0.1
|
110 |
+
fsdp:
|
111 |
+
fsdp_config:
|
112 |
+
special_tokens:
|
113 |
+
pad_token: "<|placeholder6|>"
|
114 |
+
eos_token: "<|im_end|>"
|
115 |
+
bos_token: "<s>"
|
116 |
+
unk_token: "<unk>"
|
117 |
+
tokens:
|
118 |
+
- <|im_start|>
|