Hugging Face's logo

kayrab
/
doktor-llama-3-cosmos-8b

Text Generation
Transformers
Safetensors
English
llama
text-generation-inference
unsloth
trl
Inference Endpoints
Model card Files Community
kayrab commited on
Commit
1abecb2
·
verified ·
1 Parent(s): 26db558

Upload 3 files

Browse files
Files changed (3) hide show
  1. training_args.bin +3 -0
  2. training_infos.md +71 -0
  3. training_log.json +146 -0
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:818381194015250cb41e61b91c2b9eb6d11b0adcf159a6638ecd1cb391b42bae
3
+ size 5176
training_infos.md ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Promt Format
2
+ ```
3
+ alpaca_prompt = """Sen bir doktorsun. Soruları buna göre cevapla.
4
+ ### <|reserved_special_token_0|>:
5
+ {}
6
+
7
+ ### <|reserved_special_token_1|>:
8
+ {}"""
9
+ ```
10
+
11
+ # Training args
12
+ ```
13
+ batch_size = 128
14
+ gradient_accumulation_steps = 32
15
+ num_train_epochs = 2
16
+ per_device_batch_size = int(batch_size / gradient_accumulation_steps)
17
+ training_args = TrainingArguments(
18
+ per_device_train_batch_size = per_device_batch_size,
19
+ per_device_eval_batch_size = per_device_batch_size,
20
+ gradient_accumulation_steps = gradient_accumulation_steps,
21
+ save_total_limit = 1,
22
+ warmup_steps = int(2000 / batch_size),
23
+ num_train_epochs = num_train_epochs,
24
+ learning_rate = 1e-4,
25
+ fp16 = not is_bfloat16_supported(),
26
+ bf16 = is_bfloat16_supported(),
27
+ optim = "adamw_8bit",
28
+ weight_decay = 0.01,
29
+ lr_scheduler_type = "linear",
30
+ seed = 3407,
31
+ output_dir = output_dir,
32
+ save_strategy = "steps",
33
+ eval_strategy = "steps",
34
+ logging_strategy = "steps",
35
+ save_steps = int(5000 / batch_size * num_train_epochs),
36
+ eval_steps = int(28900 / batch_size * num_train_epochs),
37
+ logging_steps = int(28900 / batch_size * num_train_epochs),
38
+ )
39
+ ```
40
+
41
+ # Trainer args
42
+
43
+ ```
44
+ max_seq_length = 8192
45
+ trainer = SFTTrainer(
46
+ model = model,
47
+ tokenizer = tokenizer,
48
+ train_dataset = train_dataset,
49
+ eval_dataset = eval_dataset,
50
+ dataset_text_field = "text",
51
+ max_seq_length = max_seq_length,
52
+ dataset_num_proc = 1,
53
+ packing = False, # Can make training 5x faster for short sequences.
54
+ args = training_args
55
+ )
56
+ ```
57
+
58
+ # From pretrained args
59
+
60
+ ```
61
+ from unsloth import FastLanguageModel
62
+ dtype = None
63
+ load_in_4bit = False
64
+
65
+ model, tokenizer = FastLanguageModel.from_pretrained(
66
+ model_name = output_dir,
67
+ max_seq_length = max_seq_length,
68
+ dtype = dtype,
69
+ load_in_4bit = load_in_4bit,
70
+ )
71
+ ```
training_log.json ADDED
@@ -0,0 +1,146 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "loss": 1.8184,
4
+ "grad_norm": 0.774284303188324,
5
+ "learning_rate": 8.923190911336132e-05,
6
+ "epoch": 0.2218976306523778,
7
+ "step": 451
8
+ },
9
+ {
10
+ "eval_loss": 1.749639630317688,
11
+ "eval_runtime": 1083.3199,
12
+ "eval_samples_per_second": 26.684,
13
+ "eval_steps_per_second": 6.671,
14
+ "epoch": 0.2218976306523778,
15
+ "step": 451
16
+ },
17
+ {
18
+ "loss": 1.7064,
19
+ "grad_norm": 0.7226008772850037,
20
+ "learning_rate": 7.809335638429242e-05,
21
+ "epoch": 0.4437952613047556,
22
+ "step": 902
23
+ },
24
+ {
25
+ "eval_loss": 1.6820720434188843,
26
+ "eval_runtime": 1087.8354,
27
+ "eval_samples_per_second": 26.573,
28
+ "eval_steps_per_second": 6.643,
29
+ "epoch": 0.4437952613047556,
30
+ "step": 902
31
+ },
32
+ {
33
+ "loss": 1.6623,
34
+ "grad_norm": 0.7210889458656311,
35
+ "learning_rate": 6.695480365522352e-05,
36
+ "epoch": 0.6656928919571334,
37
+ "step": 1353
38
+ },
39
+ {
40
+ "eval_loss": 1.6484241485595703,
41
+ "eval_runtime": 1084.0066,
42
+ "eval_samples_per_second": 26.667,
43
+ "eval_steps_per_second": 6.667,
44
+ "epoch": 0.6656928919571334,
45
+ "step": 1353
46
+ },
47
+ {
48
+ "loss": 1.6362,
49
+ "grad_norm": 0.7425829172134399,
50
+ "learning_rate": 5.581625092615461e-05,
51
+ "epoch": 0.8875905226095112,
52
+ "step": 1804
53
+ },
54
+ {
55
+ "eval_loss": 1.6261749267578125,
56
+ "eval_runtime": 1085.6726,
57
+ "eval_samples_per_second": 26.626,
58
+ "eval_steps_per_second": 6.657,
59
+ "epoch": 0.8875905226095112,
60
+ "step": 1804
61
+ },
62
+ {
63
+ "loss": 1.5914,
64
+ "grad_norm": 0.7476176023483276,
65
+ "learning_rate": 4.4677698197085704e-05,
66
+ "epoch": 1.109488153261889,
67
+ "step": 2255
68
+ },
69
+ {
70
+ "eval_loss": 1.6124544143676758,
71
+ "eval_runtime": 1084.6953,
72
+ "eval_samples_per_second": 26.65,
73
+ "eval_steps_per_second": 6.663,
74
+ "epoch": 1.109488153261889,
75
+ "step": 2255
76
+ },
77
+ {
78
+ "loss": 1.5557,
79
+ "grad_norm": 0.7473255395889282,
80
+ "learning_rate": 3.3539145468016795e-05,
81
+ "epoch": 1.3313857839142669,
82
+ "step": 2706
83
+ },
84
+ {
85
+ "eval_loss": 1.6011990308761597,
86
+ "eval_runtime": 1084.5737,
87
+ "eval_samples_per_second": 26.653,
88
+ "eval_steps_per_second": 6.663,
89
+ "epoch": 1.3313857839142669,
90
+ "step": 2706
91
+ },
92
+ {
93
+ "loss": 1.5468,
94
+ "grad_norm": 0.750347375869751,
95
+ "learning_rate": 2.240059273894789e-05,
96
+ "epoch": 1.5532834145666445,
97
+ "step": 3157
98
+ },
99
+ {
100
+ "eval_loss": 1.592125415802002,
101
+ "eval_runtime": 1086.1382,
102
+ "eval_samples_per_second": 26.614,
103
+ "eval_steps_per_second": 6.654,
104
+ "epoch": 1.5532834145666445,
105
+ "step": 3157
106
+ },
107
+ {
108
+ "loss": 1.539,
109
+ "grad_norm": 0.7452530860900879,
110
+ "learning_rate": 1.1262040009878982e-05,
111
+ "epoch": 1.7751810452190224,
112
+ "step": 3608
113
+ },
114
+ {
115
+ "eval_loss": 1.5855711698532104,
116
+ "eval_runtime": 1084.064,
117
+ "eval_samples_per_second": 26.665,
118
+ "eval_steps_per_second": 6.667,
119
+ "epoch": 1.7751810452190224,
120
+ "step": 3608
121
+ },
122
+ {
123
+ "loss": 1.5373,
124
+ "grad_norm": 0.7762174606323242,
125
+ "learning_rate": 1.2348728081007656e-07,
126
+ "epoch": 1.9970786758714003,
127
+ "step": 4059
128
+ },
129
+ {
130
+ "eval_loss": 1.5827687978744507,
131
+ "eval_runtime": 1086.4478,
132
+ "eval_samples_per_second": 26.607,
133
+ "eval_steps_per_second": 6.652,
134
+ "epoch": 1.9970786758714003,
135
+ "step": 4059
136
+ },
137
+ {
138
+ "train_runtime": 69830.4547,
139
+ "train_samples_per_second": 7.451,
140
+ "train_steps_per_second": 0.058,
141
+ "total_flos": 9.47865607059515e+18,
142
+ "train_loss": 1.5162640926171476,
143
+ "epoch": 1.9995387382954841,
144
+ "step": 4064
145
+ }
146
+ ]