Skip to content

Commit

Permalink
Add adepter selection argument.
Browse files Browse the repository at this point in the history
  • Loading branch information
cryscan committed Aug 2, 2023
1 parent efdf98a commit 3120154
Show file tree
Hide file tree
Showing 4 changed files with 41 additions and 10 deletions.
11 changes: 10 additions & 1 deletion README.md
Original file line number Diff line number Diff line change
Expand Up @@ -96,7 +96,16 @@ QQ Group for communication: 30920262
* `--model`: Model path
* `--tokenizer`: Tokenizer path
* `--port`: Running port
* `--quant`: Specify number of layers to quantize
* `--quant`: Specify the number of quantization layers
* `--adapter`: Adapter (GPU and backend) selection options

### Example

The server listens on port 3000, loads the full-layer quantized (32 > 24) 0.4B model, and selects adapter 0 (to get the specific adapter number, you can first not add this parameter, and the program will enter the adapter selection page).

```bash
$ cargo run --release -- --model assets/models/RWKV-4-World-0.4B-v1-20230529-ctx4096.st --port 3000 --quant 32 --adapter 0
```

## 📙Currently Available APIs

Expand Down
13 changes: 11 additions & 2 deletions README_jp.md
Original file line number Diff line number Diff line change
Expand Up @@ -92,9 +92,18 @@ OpenAIのChatGPT APIインターフェースと互換性があります。
## 📝サポートされている起動パラメーター

* `--model`: モデルのパス
* `--tokenizer`: トークナイザーの
* `--tokenizer`: トークナイザーのパス
* `--port`: 実行ポート
* `--quant`: クオンタイズするレイヤーの数を指定します
* `--quant`: 量子化レイヤーの数を指定
* `--adapter`: アダプター(GPUおよびバックエンド)の選択オプション

###

サーバーはポート3000でリッスンし、全レイヤー量子化(32 > 24)の0.4Bモデルをロードし、アダプター0を選択します(特定のアダプター番号を取得するには、最初にこのパラメーターを追加せず、プログラムがアダプター選択ページに入るまで待ちます)。

```bash
$ cargo run --release -- --model assets/models/RWKV-4-World-0.4B-v1-20230529-ctx4096.st --port 3000 --quant 32 --adapter 0
```

## 📙現在利用可能なAPI

Expand Down
8 changes: 8 additions & 0 deletions README_zh.md
Original file line number Diff line number Diff line change
Expand Up @@ -104,6 +104,14 @@
- `--tokenizer`: 词表路径
- `--port`: 运行端口
- `--quant`: 指定量化层数
- `--adapter`: 适配器(GPU和后端)选择项

### 示例

服务器监听3000端口,加载全部层量化(32 > 24)的0.4B模型,选择0号适配器(要查看具体适配器编号可以先不加该参数,程序会先进入选择页面)。
```bash
$ cargo run --release -- --model assets/models/RWKV-4-World-0.4B-v1-20230529-ctx4096.st --port 3000 --quant 32 --adapter 0
```


## 📙目前可用的API
Expand Down
19 changes: 12 additions & 7 deletions src/main.rs
Original file line number Diff line number Diff line change
Expand Up @@ -119,14 +119,17 @@ pub struct ReloadRequest {
pub quantized_layers: Vec<usize>,
}

async fn create_environment() -> Result<Environment> {
async fn create_environment(selection: Option<usize>) -> Result<Environment> {
let instance = Instance::new();
let adapters = instance.adapters();
let selection = Select::with_theme(&ColorfulTheme::default())
.with_prompt("Please select an adapter")
.default(0)
.items(&adapters)
.interact()?;
let selection = match selection {
Some(selection) => selection,
None => Select::with_theme(&ColorfulTheme::default())
.with_prompt("Please select an adapter")
.default(0)
.items(&adapters)
.interact()?,
};

let adapter = instance.select_adapter(selection)?;
let env = Environment::new(adapter).await?;
Expand Down Expand Up @@ -347,6 +350,8 @@ fn model_task(model: Model, tokenizer: Tokenizer, receiver: Receiver<ThreadReque
#[derive(Parser, Debug, Clone)]
#[command(author, version, about, long_about = None)]
struct Args {
#[arg(long, short)]
adepter: Option<usize>,
#[arg(long, short, value_name = "FILE")]
model: Option<String>,
#[arg(long, short, value_name = "FILE")]
Expand Down Expand Up @@ -385,7 +390,7 @@ async fn main() -> Result<()> {
);

let (sender, receiver) = flume::unbounded::<ThreadRequest>();
let env = create_environment().await?;
let env = create_environment(args.adepter).await?;
let tokenizer = load_tokenizer(&tokenizer_path)?;

log::info!("{:#?}", env.adapter.get_info());
Expand Down

0 comments on commit 3120154

Please sign in to comment.