82 lines
3.1 KiB
Markdown
82 lines
3.1 KiB
Markdown
|
---
|
|||
|
|
language:
|
|||
|
|
- en
|
|||
|
|
- ja
|
|||
|
|
library_name: transformers
|
|||
|
|
pipeline_tag: text-generation
|
|||
|
|
license: other
|
|||
|
|
license_name: plamo-community-license
|
|||
|
|
license_link: LICENSE
|
|||
|
|
base_model:
|
|||
|
|
- Qwen/Qwen3-1.7B-Base
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
# Qwen3-1.7B-pfn-qfin
|
|||
|
|
|
|||
|
|
## Model Description
|
|||
|
|
Qwen3-1.7B-pfn-qfin is an fine-tuned model based on [Qwen/Qwen3-1.7B-Base](https://huggingface.co/Qwen/Qwen3-1.7B-Base).
|
|||
|
|
This is the base model, which is good at generating continuous sentences.
|
|||
|
|
Qwen3-1.7B-pfn-qfin is fine-tuned on about 400M tokens from multiple special datasets generated by Preferred Networks, which is clear to use for commercial usage.
|
|||
|
|
The fine-tuned were carried out at a 2048 context length.
|
|||
|
|
This model is released under [PLaMo Community License](https://www.preferred.jp/ja/plamo-community-license/).
|
|||
|
|
|
|||
|
|
# Benchmarking
|
|||
|
|
The benchmark score is obtained using [Japanese Language Model Financial Evaluation Harness](https://github.com/pfnet-research/japanese-lm-fin-harness)
|
|||
|
|
For the benchmark, 0-shot and default prompts are used.
|
|||
|
|
|
|||
|
|
| Task |Metric|Qwen3-1.7B| Ours |
|
|||
|
|
|----------------|------|------|------|
|
|||
|
|
|chabsa |f1 |0.5734|0.7116|
|
|||
|
|
|cma_basics |acc |0.3158|0.5263|
|
|||
|
|
|cpa_audit |acc |0.1583|0.1884|
|
|||
|
|
|fp2 |acc |0.4737|0.4912|
|
|||
|
|
|security_sales_1|acc |0.2421|0.3389|
|
|||
|
|
|----------------|------|------|------|
|
|||
|
|
|OVER ALL | |0.3527|0.4513|
|
|||
|
|
|
|||
|
|
## Usage
|
|||
|
|
Install the required libraries as follows:
|
|||
|
|
```sh
|
|||
|
|
>>> python -m pip install "transformers>=4.51.0"
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
Execute the following python code:
|
|||
|
|
```python
|
|||
|
|
import torch
|
|||
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM
|
|||
|
|
|
|||
|
|
tokenizer = AutoTokenizer.from_pretrained("pfnet/Qwen3-1.7B-pfn-qfin", trust_remote_code=True)
|
|||
|
|
model = AutoModelForCausalLM.from_pretrained("pfnet/Qwen3-1.7B-pfn-qfin", device_map="auto", trust_remote_code=True)
|
|||
|
|
text = "日本銀行は"
|
|||
|
|
input_ids = tokenizer(text, return_tensors="pt").input_ids.to(model.device)
|
|||
|
|
with torch.no_grad():
|
|||
|
|
generated_tokens = model.generate(
|
|||
|
|
inputs=input_ids,
|
|||
|
|
max_new_tokens=32,
|
|||
|
|
do_sample=True,
|
|||
|
|
top_k=50,
|
|||
|
|
top_p=0.95,
|
|||
|
|
temperature=1.0,
|
|||
|
|
pad_token_id=tokenizer.pad_token_id,
|
|||
|
|
bos_token_id=tokenizer.bos_token_id,
|
|||
|
|
eos_token_id=tokenizer.eos_token_id
|
|||
|
|
)[0]
|
|||
|
|
generated_text = tokenizer.decode(generated_tokens)
|
|||
|
|
print(generated_text)
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
## Bias, Risks, and Limitations
|
|||
|
|
Qwen3-1.7B-pfn-qfin is a new technology that carries risks with use.
|
|||
|
|
Testing conducted to date has been in English and Japanese, and has not covered, nor could it cover all scenarios.
|
|||
|
|
For these reasons, as with all LLMs, Qwen3-1.7B-pfn-qfin’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate, biased or other objectionable responses to user prompts.
|
|||
|
|
This model is not designed for legal, tax, investment, financial, or other advice.
|
|||
|
|
Therefore, before deploying any applications of Qwen3-1.7B-pfn-qfin, developers should perform safety testing and tuning tailored to their specific applications of the model.
|
|||
|
|
|
|||
|
|
## Authors
|
|||
|
|
Preferred Networks, Inc.
|
|||
|
|
- Masanori Hirano
|
|||
|
|
- Kentaro Imajo
|
|||
|
|
- Takeshi Masuko
|
|||
|
|
|
|||
|
|
# License
|
|||
|
|
[PLaMo Community License](https://www.preferred.jp/ja/plamo-community-license/)
|