File size: 3,922 Bytes
db5f960
 
 
 
 
 
 
 
 
 
 
 
 
 
 
12482b0
db5f960
 
 
5da5d03
 
 
 
 
 
 
 
 
 
db5f960
 
 
12482b0
db5f960
6032eb5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
db5f960
 
5c83f1a
db5f960
 
 
5c83f1a
 
8427f91
 
 
 
 
5c83f1a
db5f960
 
 
 
1b49982
db5f960
5c83f1a
db5f960
 
 
 
 
 
12482b0
5c83f1a
db5f960
 
 
12482b0
 
db5f960
 
 
 
 
 
 
1
2
3
4
5
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
---
license: other
base_model: microsoft/phi-1_5
tags:
- generated_from_trainer
model-index:
- name: phi-1_5-pl-v_0_1
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# phi-1_5-pl-v_0_1

This model is based on [microsoft/phi-1_5](https://huggingface.co/microsoft/phi-1_5). It was trained from scratch on the 20231201 Polish Wikipedia dump.

## Model description

The model was trained for a context length of 1024 tokens. In addition, while the original model has a hidden size of 2048 (1.3B parameters), this model has a hidden size of 1024 (450.3M parameters).

The model used for training was as follows:
```
model_config = AutoConfig.from_pretrained(
    'microsoft/phi-1_5', vocab_size=len(trained_tokenizer), max_position_embeddings=1024,
    hidden_size=1024, attn_implementation="flash_attention_2", trust_remote_code=True
)
model = AutoModelForCausalLM.from_config(model_config, trust_remote_code=True)
```

## Intended uses & limitations

The model is intended for research purposes only. It may generate fictitious, incorrect, unethical, or biased texts. At its current state, it is not suitable for production purposes.

Example:
```
tokenizer = AutoTokenizer.from_pretrained(
    model_name, trust_remote_code=True, use_fast=True
)
# to use flash_attention_2, may need to load the original microsoft phi-1.5 and load weights from this model
model = AutoModelForCausalLM.from_pretrained(
    model_name, vocab_size=len(tokenizer), # attn_implementation="flash_attention_2",
    trust_remote_code=True, torch_dtype=torch.bfloat16
).to(torch.device('cuda'))
model.eval()

generation_config = GenerationConfig.from_pretrained(
    model_name, do_sample=False, repetition_penalty=1.5,
    min_new_tokens=1, max_new_tokens=128
)

test_input = tokenizer("Wrocław to polski miasto. Wrocław jest ", return_tensors='pt').to(torch.device('cuda'))
test_output = model.generate(**test_input, generation_config=generation_config)
test_preds = tokenizer.batch_decode(sequences=test_output, skip_special_tokens=True, clean_up_tokenization_spaces=True)
print(test_preds)
```

Output:
```
['Wrocław to polski miasto. Wrocław jest  stolicą województwa dolnośląskiego, a także siedzibą władz powiatu wrocławskiego i gminy miejsko-wiejskiej Wrocław\n\nMiasto leży w południowo–zachodniej części Dolnego Śląska na Przedgórzu Sudeckim nad rzeką Odrą (odnoga Odry). Przez miasto przebiega droga krajowa nr 94 łącząca Berlin z Wrocławiem oraz linia kolejowa do Wrocławia Głównego przez Wrocław Główny – Kłodzko Główne/Szczecin Zachodni - Legnica. Miasto posiada połączenie kolejowe ze stacją kolejową Wrocław Gądów Mały lub Gądowem Małym poprzez węzeł kolejowy Wrocław Gądów Wielki. W mieście znajduje się stacja towarowa Wrocław Gądów Mały.\nW latach 1975−1998 miejscowość administracyjnie należała do woj. wałbrzyskiego. Od 1']
```

## Training and evaluation data

The 20231201 Polish Wikipedia dump.

## Training procedure

### Training environment

- GPU: 4 x RTX4090 (24GB per GPU, 96GB total)
- CPU: AMD EPYC 75F3 32-core (128 virtual cores)
- RAM: 258GB
- Motherboard: ROME2D32GM PCLe 4.0, 16x
- Storage: nvme 194.0GB

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 0.0002
- distributed_type: multi-GPU (DDP)
- num_devices: 4
- train_batch_size: 2
- gradient_accumulation_steps: 8
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-07
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 2
- precision: bf16
- seed: 42

### Training results

- runtime: 2d 21h 26m 36s
- train_loss: 2.727

### Framework versions

- Transformers 4.36.2
- Pytorch 2.1.2
- Datasets 2.14.7
- Tokenizers 0.15.0