Upload folder using huggingface_hub
Browse files- README.md +61 -0
- config.json +29 -0
- generation_config.json +6 -0
- model.safetensors +3 -0
- pytorch_model.bin +3 -0
- tokenizer.model +3 -0
- tokenizer.vocab +0 -0
- tokenizer_config.json +12 -0
README.md
ADDED
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
base_model: llama
|
4 |
+
library_name: transformers
|
5 |
+
pipeline_tag: text-generation
|
6 |
+
tags:
|
7 |
+
- one-way-polyglot
|
8 |
+
- japanese
|
9 |
+
- english
|
10 |
+
- bilingual
|
11 |
+
- small-model
|
12 |
+
---
|
13 |
+
|
14 |
+
# one-way-polyglot-8m-tied
|
15 |
+
|
16 |
+
A one-way polyglot language model trained to understand Japanese but generate only English.
|
17 |
+
|
18 |
+
## Model Details
|
19 |
+
|
20 |
+
- **Architecture**: LLaMA-based transformer
|
21 |
+
- **Parameters**: 8,519,936 (8.5M)
|
22 |
+
- **Vocabulary**: 16,384 tokens (bilingual SentencePiece)
|
23 |
+
- **Context Length**: 512 tokens
|
24 |
+
- **Embedding Strategy**: Tied
|
25 |
+
|
26 |
+
## Capabilities
|
27 |
+
|
28 |
+
- **Semantic Transfer**: Understands Japanese input and generates contextually appropriate English
|
29 |
+
- **One-Way Constraint**: Strong bias toward English-only generation
|
30 |
+
- **Name Transliteration**: Can transliterate Japanese names to English (context-dependent)
|
31 |
+
|
32 |
+
## Training Data
|
33 |
+
|
34 |
+
Trained on bilingual Japanese-English story data with masked loss on Japanese prefixes to enforce one-way generation.
|
35 |
+
|
36 |
+
## Usage
|
37 |
+
|
38 |
+
```python
|
39 |
+
from transformers import LlamaForCausalLM, LlamaTokenizer
|
40 |
+
|
41 |
+
model = LlamaForCausalLM.from_pretrained("one-way-polyglot-8m-tied")
|
42 |
+
tokenizer = LlamaTokenizer.from_pretrained("one-way-polyglot-8m-tied")
|
43 |
+
|
44 |
+
# Japanese input, English output
|
45 |
+
prompt = "ζγ
γθ΅€γεγζγ£γε°ε₯³γγγΎγγγ"
|
46 |
+
inputs = tokenizer(prompt, return_tensors="pt")
|
47 |
+
outputs = model.generate(**inputs, max_new_tokens=50, temperature=0.7)
|
48 |
+
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
|
49 |
+
```
|
50 |
+
|
51 |
+
## Model Variants
|
52 |
+
|
53 |
+
This is part of a series exploring one-way polyglot capabilities:
|
54 |
+
- 1.25M parameters (tied embeddings)
|
55 |
+
- 8.5M parameters (tied embeddings)
|
56 |
+
- 12.7M parameters (untied embeddings)
|
57 |
+
- 15.7M parameters (tied embeddings)
|
58 |
+
|
59 |
+
## License
|
60 |
+
|
61 |
+
Apache 2.0
|
config.json
ADDED
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"architectures": [
|
3 |
+
"LlamaForCausalLM"
|
4 |
+
],
|
5 |
+
"attention_bias": false,
|
6 |
+
"attention_dropout": 0.0,
|
7 |
+
"bos_token_id": 1,
|
8 |
+
"eos_token_id": 2,
|
9 |
+
"head_dim": 64,
|
10 |
+
"hidden_act": "silu",
|
11 |
+
"hidden_size": 256,
|
12 |
+
"initializer_range": 0.02,
|
13 |
+
"intermediate_size": 682,
|
14 |
+
"max_position_embeddings": 512,
|
15 |
+
"mlp_bias": false,
|
16 |
+
"model_type": "llama",
|
17 |
+
"num_attention_heads": 4,
|
18 |
+
"num_hidden_layers": 6,
|
19 |
+
"num_key_value_heads": 2,
|
20 |
+
"pretraining_tp": 1,
|
21 |
+
"rms_norm_eps": 1e-06,
|
22 |
+
"rope_scaling": null,
|
23 |
+
"rope_theta": 10000.0,
|
24 |
+
"tie_word_embeddings": true,
|
25 |
+
"torch_dtype": "float32",
|
26 |
+
"transformers_version": "4.54.1",
|
27 |
+
"use_cache": true,
|
28 |
+
"vocab_size": 16384
|
29 |
+
}
|
generation_config.json
ADDED
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_from_model_config": true,
|
3 |
+
"bos_token_id": 1,
|
4 |
+
"eos_token_id": 2,
|
5 |
+
"transformers_version": "4.54.1"
|
6 |
+
}
|
model.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:1804e76de82b0380509a595273b62af4117bf200d42183ba15d433243bfbac91
|
3 |
+
size 34085832
|
pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:4359c7d373bfaa6bc9ec396e10c9cde2d0e7604a0821075d8529b3561b7c2778
|
3 |
+
size 34102703
|
tokenizer.model
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:18fd3bcd564240df1343dda502457ea5db419f4061ca442a9b204796d3d79301
|
3 |
+
size 581374
|
tokenizer.vocab
ADDED
The diff for this file is too large to render.
See raw diff
|
|
tokenizer_config.json
ADDED
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"tokenizer_class": "LlamaTokenizer",
|
3 |
+
"vocab_size": 16384,
|
4 |
+
"model_max_length": 512,
|
5 |
+
"pad_token": "[PAD]",
|
6 |
+
"bos_token": "[BOS]",
|
7 |
+
"eos_token": "[EOS]",
|
8 |
+
"unk_token": "[UNK]",
|
9 |
+
"add_bos_token": true,
|
10 |
+
"add_eos_token": false,
|
11 |
+
"clean_up_tokenization_spaces": false
|
12 |
+
}
|