danielgombas
commited on
danielgombas/llama_3b_step2_batch_v1
Browse files- .ipynb_checkpoints/args_v1-checkpoint.txt +71 -0
- README.md +169 -194
- args_v1.txt +71 -0
- training_args.bin +3 -0
.ipynb_checkpoints/args_v1-checkpoint.txt
ADDED
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
max_seq_length = 500
|
2 |
+
|
3 |
+
def fmt(examples):
|
4 |
+
print(len(examples))
|
5 |
+
return examples
|
6 |
+
|
7 |
+
# 'lora_r' is the dimension of the LoRA attention.
|
8 |
+
lora_r = 32
|
9 |
+
|
10 |
+
# 'lora_alpha' is the alpha parameter for LoRA scaling.
|
11 |
+
lora_alpha = 16
|
12 |
+
|
13 |
+
# 'lora_dropout' is the dropout probability for LoRA layers.
|
14 |
+
lora_dropout = 0.05
|
15 |
+
|
16 |
+
# 'target_modules' is a list of the modules that should be targeted by LoRA.
|
17 |
+
target_modules= ['k_proj', 'q_proj', 'v_proj', 'o_proj', "gate_proj", "down_proj", "up_proj"]
|
18 |
+
|
19 |
+
# 'se
|
20 |
+
|
21 |
+
peft_config = LoraConfig(
|
22 |
+
r=lora_r,
|
23 |
+
lora_alpha=lora_alpha,
|
24 |
+
lora_dropout=lora_dropout,
|
25 |
+
task_type=TaskType.CAUSAL_LM,
|
26 |
+
target_modules=target_modules,
|
27 |
+
)
|
28 |
+
|
29 |
+
trainer = SFTTrainer(
|
30 |
+
model = model,
|
31 |
+
tokenizer = tokenizer,
|
32 |
+
train_dataset = qa_dataset['train'],
|
33 |
+
eval_dataset = qa_dataset['test'],
|
34 |
+
dataset_text_field = "text",
|
35 |
+
max_seq_length = max_seq_length,
|
36 |
+
dataset_num_proc = 4,
|
37 |
+
data_collator = collator,
|
38 |
+
# formatting_func = fmt,
|
39 |
+
# peft_config=peft_config,
|
40 |
+
args = TrainingArguments(
|
41 |
+
per_device_train_batch_size = 1,
|
42 |
+
gradient_checkpointing = True,
|
43 |
+
gradient_accumulation_steps = 4,
|
44 |
+
per_device_eval_batch_size = 40,
|
45 |
+
do_eval = True,
|
46 |
+
eval_strategy = 'steps',
|
47 |
+
eval_steps = 50,
|
48 |
+
# save_strategy = 'steps',
|
49 |
+
save_steps = 1000,
|
50 |
+
|
51 |
+
# Use num_train_epochs and warmup_ratio for longer runs!
|
52 |
+
# max_steps = 70,
|
53 |
+
# warmup_steps = 10,
|
54 |
+
# warmup_ratio = 0.1,
|
55 |
+
num_train_epochs = 2,
|
56 |
+
|
57 |
+
# Select a 2 to 10x smaller learning rate for the embedding matrices!
|
58 |
+
learning_rate = 1e-5,
|
59 |
+
# embedding_learning_rate = 1e-6,
|
60 |
+
|
61 |
+
# fp16 = not is_bfloat16_supported(),
|
62 |
+
bf16 = True,
|
63 |
+
logging_steps = 1,
|
64 |
+
optim = "adamw_torch",
|
65 |
+
weight_decay = 0.00,
|
66 |
+
lr_scheduler_type = "linear",
|
67 |
+
# seed = 3407,
|
68 |
+
|
69 |
+
output_dir = "llama_3b_step2_batch_v1",
|
70 |
+
),
|
71 |
+
)
|
README.md
CHANGED
@@ -3,199 +3,174 @@ library_name: transformers
|
|
3 |
tags:
|
4 |
- trl
|
5 |
- sft
|
|
|
|
|
|
|
|
|
6 |
---
|
7 |
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
-
|
35 |
-
-
|
36 |
-
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
|
167 |
-
|
168 |
-
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
## Citation [optional]
|
174 |
-
|
175 |
-
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
|
176 |
-
|
177 |
-
**BibTeX:**
|
178 |
-
|
179 |
-
[More Information Needed]
|
180 |
-
|
181 |
-
**APA:**
|
182 |
-
|
183 |
-
[More Information Needed]
|
184 |
-
|
185 |
-
## Glossary [optional]
|
186 |
-
|
187 |
-
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
|
188 |
-
|
189 |
-
[More Information Needed]
|
190 |
-
|
191 |
-
## More Information [optional]
|
192 |
-
|
193 |
-
[More Information Needed]
|
194 |
-
|
195 |
-
## Model Card Authors [optional]
|
196 |
-
|
197 |
-
[More Information Needed]
|
198 |
-
|
199 |
-
## Model Card Contact
|
200 |
-
|
201 |
-
[More Information Needed]
|
|
|
3 |
tags:
|
4 |
- trl
|
5 |
- sft
|
6 |
+
- generated_from_trainer
|
7 |
+
model-index:
|
8 |
+
- name: llama_3b_step2_batch_v1
|
9 |
+
results: []
|
10 |
---
|
11 |
|
12 |
+
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
13 |
+
should probably proofread and complete it, then remove this comment. -->
|
14 |
+
|
15 |
+
# llama_3b_step2_batch_v1
|
16 |
+
|
17 |
+
This model was trained from scratch on an unknown dataset.
|
18 |
+
It achieves the following results on the evaluation set:
|
19 |
+
- Loss: 0.5060
|
20 |
+
|
21 |
+
## Model description
|
22 |
+
|
23 |
+
More information needed
|
24 |
+
|
25 |
+
## Intended uses & limitations
|
26 |
+
|
27 |
+
More information needed
|
28 |
+
|
29 |
+
## Training and evaluation data
|
30 |
+
|
31 |
+
More information needed
|
32 |
+
|
33 |
+
## Training procedure
|
34 |
+
|
35 |
+
### Training hyperparameters
|
36 |
+
|
37 |
+
The following hyperparameters were used during training:
|
38 |
+
- learning_rate: 1e-05
|
39 |
+
- train_batch_size: 1
|
40 |
+
- eval_batch_size: 40
|
41 |
+
- seed: 42
|
42 |
+
- gradient_accumulation_steps: 4
|
43 |
+
- total_train_batch_size: 4
|
44 |
+
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
|
45 |
+
- lr_scheduler_type: linear
|
46 |
+
- num_epochs: 2
|
47 |
+
|
48 |
+
### Training results
|
49 |
+
|
50 |
+
| Training Loss | Epoch | Step | Validation Loss |
|
51 |
+
|:-------------:|:------:|:----:|:---------------:|
|
52 |
+
| 1.0531 | 0.0170 | 50 | 1.2007 |
|
53 |
+
| 1.0336 | 0.0341 | 100 | 1.1242 |
|
54 |
+
| 0.9428 | 0.0511 | 150 | 1.0800 |
|
55 |
+
| 1.4386 | 0.0682 | 200 | 1.0408 |
|
56 |
+
| 0.8375 | 0.0852 | 250 | 1.0127 |
|
57 |
+
| 0.9193 | 0.1023 | 300 | 0.9817 |
|
58 |
+
| 1.0368 | 0.1193 | 350 | 0.9573 |
|
59 |
+
| 1.2018 | 0.1364 | 400 | 0.9319 |
|
60 |
+
| 1.2749 | 0.1534 | 450 | 0.9072 |
|
61 |
+
| 0.9881 | 0.1704 | 500 | 0.8820 |
|
62 |
+
| 0.9707 | 0.1875 | 550 | 0.8599 |
|
63 |
+
| 1.2377 | 0.2045 | 600 | 0.8412 |
|
64 |
+
| 0.9024 | 0.2216 | 650 | 0.8180 |
|
65 |
+
| 0.5889 | 0.2386 | 700 | 0.8024 |
|
66 |
+
| 0.8046 | 0.2557 | 750 | 0.7899 |
|
67 |
+
| 0.83 | 0.2727 | 800 | 0.7710 |
|
68 |
+
| 0.6852 | 0.2898 | 850 | 0.7548 |
|
69 |
+
| 0.8512 | 0.3068 | 900 | 0.7422 |
|
70 |
+
| 0.8377 | 0.3238 | 950 | 0.7345 |
|
71 |
+
| 0.5361 | 0.3409 | 1000 | 0.7220 |
|
72 |
+
| 0.7696 | 0.3579 | 1050 | 0.7105 |
|
73 |
+
| 0.8175 | 0.3750 | 1100 | 0.7013 |
|
74 |
+
| 0.6144 | 0.3920 | 1150 | 0.6886 |
|
75 |
+
| 0.3598 | 0.4091 | 1200 | 0.6809 |
|
76 |
+
| 0.7176 | 0.4261 | 1250 | 0.6692 |
|
77 |
+
| 0.5281 | 0.4432 | 1300 | 0.6644 |
|
78 |
+
| 0.3555 | 0.4602 | 1350 | 0.6547 |
|
79 |
+
| 0.9024 | 0.4772 | 1400 | 0.6471 |
|
80 |
+
| 0.7713 | 0.4943 | 1450 | 0.6386 |
|
81 |
+
| 0.6172 | 0.5113 | 1500 | 0.6322 |
|
82 |
+
| 0.6325 | 0.5284 | 1550 | 0.6266 |
|
83 |
+
| 0.7503 | 0.5454 | 1600 | 0.6206 |
|
84 |
+
| 0.349 | 0.5625 | 1650 | 0.6136 |
|
85 |
+
| 0.7 | 0.5795 | 1700 | 0.6085 |
|
86 |
+
| 0.5014 | 0.5966 | 1750 | 0.6023 |
|
87 |
+
| 0.6441 | 0.6136 | 1800 | 0.5975 |
|
88 |
+
| 0.5066 | 0.6306 | 1850 | 0.5921 |
|
89 |
+
| 0.6036 | 0.6477 | 1900 | 0.5883 |
|
90 |
+
| 0.6549 | 0.6647 | 1950 | 0.5840 |
|
91 |
+
| 0.3903 | 0.6818 | 2000 | 0.5789 |
|
92 |
+
| 0.8864 | 0.6988 | 2050 | 0.5754 |
|
93 |
+
| 0.7164 | 0.7159 | 2100 | 0.5709 |
|
94 |
+
| 0.5504 | 0.7329 | 2150 | 0.5687 |
|
95 |
+
| 0.4216 | 0.7500 | 2200 | 0.5646 |
|
96 |
+
| 0.4241 | 0.7670 | 2250 | 0.5618 |
|
97 |
+
| 0.6452 | 0.7840 | 2300 | 0.5590 |
|
98 |
+
| 0.7067 | 0.8011 | 2350 | 0.5558 |
|
99 |
+
| 0.4536 | 0.8181 | 2400 | 0.5537 |
|
100 |
+
| 0.8657 | 0.8352 | 2450 | 0.5508 |
|
101 |
+
| 0.7452 | 0.8522 | 2500 | 0.5483 |
|
102 |
+
| 0.3444 | 0.8693 | 2550 | 0.5458 |
|
103 |
+
| 0.2889 | 0.8863 | 2600 | 0.5437 |
|
104 |
+
| 0.2415 | 0.9034 | 2650 | 0.5401 |
|
105 |
+
| 0.5393 | 0.9204 | 2700 | 0.5385 |
|
106 |
+
| 0.4866 | 0.9374 | 2750 | 0.5372 |
|
107 |
+
| 0.9233 | 0.9545 | 2800 | 0.5347 |
|
108 |
+
| 0.4623 | 0.9715 | 2850 | 0.5318 |
|
109 |
+
| 0.4211 | 0.9886 | 2900 | 0.5299 |
|
110 |
+
| 0.4308 | 1.0056 | 2950 | 0.5283 |
|
111 |
+
| 0.618 | 1.0227 | 3000 | 0.5285 |
|
112 |
+
| 0.7693 | 1.0397 | 3050 | 0.5262 |
|
113 |
+
| 0.2893 | 1.0568 | 3100 | 0.5266 |
|
114 |
+
| 0.461 | 1.0738 | 3150 | 0.5273 |
|
115 |
+
| 0.3648 | 1.0908 | 3200 | 0.5230 |
|
116 |
+
| 0.4981 | 1.1079 | 3250 | 0.5253 |
|
117 |
+
| 0.5005 | 1.1249 | 3300 | 0.5222 |
|
118 |
+
| 0.4117 | 1.1420 | 3350 | 0.5217 |
|
119 |
+
| 0.3319 | 1.1590 | 3400 | 0.5188 |
|
120 |
+
| 0.2549 | 1.1761 | 3450 | 0.5190 |
|
121 |
+
| 0.3758 | 1.1931 | 3500 | 0.5186 |
|
122 |
+
| 0.2889 | 1.2102 | 3550 | 0.5173 |
|
123 |
+
| 0.6341 | 1.2272 | 3600 | 0.5167 |
|
124 |
+
| 0.3217 | 1.2442 | 3650 | 0.5155 |
|
125 |
+
| 0.4406 | 1.2613 | 3700 | 0.5150 |
|
126 |
+
| 0.7445 | 1.2783 | 3750 | 0.5148 |
|
127 |
+
| 0.5511 | 1.2954 | 3800 | 0.5133 |
|
128 |
+
| 0.3933 | 1.3124 | 3850 | 0.5125 |
|
129 |
+
| 0.39 | 1.3295 | 3900 | 0.5134 |
|
130 |
+
| 0.3015 | 1.3465 | 3950 | 0.5126 |
|
131 |
+
| 0.8124 | 1.3636 | 4000 | 0.5118 |
|
132 |
+
| 0.6512 | 1.3806 | 4050 | 0.5111 |
|
133 |
+
| 0.7011 | 1.3976 | 4100 | 0.5106 |
|
134 |
+
| 0.4556 | 1.4147 | 4150 | 0.5103 |
|
135 |
+
| 0.4563 | 1.4317 | 4200 | 0.5100 |
|
136 |
+
| 0.2651 | 1.4488 | 4250 | 0.5100 |
|
137 |
+
| 0.5674 | 1.4658 | 4300 | 0.5090 |
|
138 |
+
| 0.2869 | 1.4829 | 4350 | 0.5093 |
|
139 |
+
| 0.5327 | 1.4999 | 4400 | 0.5088 |
|
140 |
+
| 0.726 | 1.5170 | 4450 | 0.5086 |
|
141 |
+
| 0.2619 | 1.5340 | 4500 | 0.5084 |
|
142 |
+
| 0.6597 | 1.5510 | 4550 | 0.5081 |
|
143 |
+
| 0.4848 | 1.5681 | 4600 | 0.5083 |
|
144 |
+
| 0.412 | 1.5851 | 4650 | 0.5080 |
|
145 |
+
| 0.6712 | 1.6022 | 4700 | 0.5077 |
|
146 |
+
| 0.5523 | 1.6192 | 4750 | 0.5076 |
|
147 |
+
| 0.5105 | 1.6363 | 4800 | 0.5077 |
|
148 |
+
| 0.5315 | 1.6533 | 4850 | 0.5071 |
|
149 |
+
| 0.4166 | 1.6704 | 4900 | 0.5069 |
|
150 |
+
| 0.4081 | 1.6874 | 4950 | 0.5065 |
|
151 |
+
| 0.3154 | 1.7044 | 5000 | 0.5063 |
|
152 |
+
| 0.396 | 1.7215 | 5050 | 0.5063 |
|
153 |
+
| 0.6121 | 1.7385 | 5100 | 0.5064 |
|
154 |
+
| 0.379 | 1.7556 | 5150 | 0.5063 |
|
155 |
+
| 0.4534 | 1.7726 | 5200 | 0.5061 |
|
156 |
+
| 0.5572 | 1.7897 | 5250 | 0.5060 |
|
157 |
+
| 0.3847 | 1.8067 | 5300 | 0.5059 |
|
158 |
+
| 0.3751 | 1.8238 | 5350 | 0.5060 |
|
159 |
+
| 0.4346 | 1.8408 | 5400 | 0.5061 |
|
160 |
+
| 0.4928 | 1.8578 | 5450 | 0.5061 |
|
161 |
+
| 0.5215 | 1.8749 | 5500 | 0.5060 |
|
162 |
+
| 0.6156 | 1.8919 | 5550 | 0.5060 |
|
163 |
+
| 0.4041 | 1.9090 | 5600 | 0.5060 |
|
164 |
+
| 0.5604 | 1.9260 | 5650 | 0.5059 |
|
165 |
+
| 0.424 | 1.9431 | 5700 | 0.5060 |
|
166 |
+
| 0.1856 | 1.9601 | 5750 | 0.5060 |
|
167 |
+
| 0.3701 | 1.9772 | 5800 | 0.5061 |
|
168 |
+
| 0.4201 | 1.9942 | 5850 | 0.5060 |
|
169 |
+
|
170 |
+
|
171 |
+
### Framework versions
|
172 |
+
|
173 |
+
- Transformers 4.46.1
|
174 |
+
- Pytorch 2.1.0+cu118
|
175 |
+
- Datasets 3.0.2
|
176 |
+
- Tokenizers 0.20.1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
args_v1.txt
ADDED
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
max_seq_length = 500
|
2 |
+
|
3 |
+
def fmt(examples):
|
4 |
+
print(len(examples))
|
5 |
+
return examples
|
6 |
+
|
7 |
+
# 'lora_r' is the dimension of the LoRA attention.
|
8 |
+
lora_r = 32
|
9 |
+
|
10 |
+
# 'lora_alpha' is the alpha parameter for LoRA scaling.
|
11 |
+
lora_alpha = 16
|
12 |
+
|
13 |
+
# 'lora_dropout' is the dropout probability for LoRA layers.
|
14 |
+
lora_dropout = 0.05
|
15 |
+
|
16 |
+
# 'target_modules' is a list of the modules that should be targeted by LoRA.
|
17 |
+
target_modules= ['k_proj', 'q_proj', 'v_proj', 'o_proj', "gate_proj", "down_proj", "up_proj"]
|
18 |
+
|
19 |
+
# 'se
|
20 |
+
|
21 |
+
peft_config = LoraConfig(
|
22 |
+
r=lora_r,
|
23 |
+
lora_alpha=lora_alpha,
|
24 |
+
lora_dropout=lora_dropout,
|
25 |
+
task_type=TaskType.CAUSAL_LM,
|
26 |
+
target_modules=target_modules,
|
27 |
+
)
|
28 |
+
|
29 |
+
trainer = SFTTrainer(
|
30 |
+
model = model,
|
31 |
+
tokenizer = tokenizer,
|
32 |
+
train_dataset = qa_dataset['train'],
|
33 |
+
eval_dataset = qa_dataset['test'],
|
34 |
+
dataset_text_field = "text",
|
35 |
+
max_seq_length = max_seq_length,
|
36 |
+
dataset_num_proc = 4,
|
37 |
+
data_collator = collator,
|
38 |
+
# formatting_func = fmt,
|
39 |
+
# peft_config=peft_config,
|
40 |
+
args = TrainingArguments(
|
41 |
+
per_device_train_batch_size = 1,
|
42 |
+
gradient_checkpointing = True,
|
43 |
+
gradient_accumulation_steps = 4,
|
44 |
+
per_device_eval_batch_size = 40,
|
45 |
+
do_eval = True,
|
46 |
+
eval_strategy = 'steps',
|
47 |
+
eval_steps = 50,
|
48 |
+
# save_strategy = 'steps',
|
49 |
+
save_steps = 1000,
|
50 |
+
|
51 |
+
# Use num_train_epochs and warmup_ratio for longer runs!
|
52 |
+
# max_steps = 70,
|
53 |
+
# warmup_steps = 10,
|
54 |
+
# warmup_ratio = 0.1,
|
55 |
+
num_train_epochs = 2,
|
56 |
+
|
57 |
+
# Select a 2 to 10x smaller learning rate for the embedding matrices!
|
58 |
+
learning_rate = 1e-5,
|
59 |
+
# embedding_learning_rate = 1e-6,
|
60 |
+
|
61 |
+
# fp16 = not is_bfloat16_supported(),
|
62 |
+
bf16 = True,
|
63 |
+
logging_steps = 1,
|
64 |
+
optim = "adamw_torch",
|
65 |
+
weight_decay = 0.00,
|
66 |
+
lr_scheduler_type = "linear",
|
67 |
+
# seed = 3407,
|
68 |
+
|
69 |
+
output_dir = "llama_3b_step2_batch_v1",
|
70 |
+
),
|
71 |
+
)
|
training_args.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:4d1dda7f4027338634ae65de01b3aae9e2f272bfcb78432f7bb9f9dbbec1c69d
|
3 |
+
size 5240
|