Spaces:
Running
Running
Update README.md
Browse files
README.md
CHANGED
|
@@ -48,7 +48,21 @@ app_file: README.md
|
|
| 48 |
|
| 49 |
> **(先阅读语境 -> 分析语境 -> 落笔翻译)**
|
| 50 |
|
| 51 |
-
这种机制精准解决了 ACGN
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
|
| 53 |
---
|
| 54 |
|
|
@@ -56,8 +70,8 @@ app_file: README.md
|
|
| 56 |
|
| 57 |
| Model Name | Type | Size | VRAM | Description |
|
| 58 |
| :--- | :--- | :--- | :--- | :--- |
|
| 59 |
-
| **[Murasaki-8B-v0.1](https://huggingface.co/Murasaki-Project/Murasaki-8B-v0.1)** | **BF16** | 15.
|
| 60 |
-
| **[Murasaki-8B-v0.1-GGUF](https://huggingface.co/Murasaki-Project/Murasaki-8B-v0.1-GGUF)** | **GGUF** |
|
| 61 |
|
| 62 |
---
|
| 63 |
|
|
|
|
| 48 |
|
| 49 |
> **(先阅读语境 -> 分析语境 -> 落笔翻译)**
|
| 50 |
|
| 51 |
+
这种机制精准解决了 ACGN 翻译中"主语省略"、"人称混淆"及"风格漂移"的三大难题。
|
| 52 |
+
|
| 53 |
+
---
|
| 54 |
+
|
| 55 |
+
## 🏆 Benchmark Performance
|
| 56 |
+
|
| 57 |
+
在 **Murasaki-ACGN Benchmark** 的综合评测中，Murasaki-8B 以 4-bit 量化版本取得了 **第一名** 的成绩。
|
| 58 |
+
|
| 59 |
+
| Rank | Model | Avg COMET |
|
| 60 |
+
| :--- | :--- | :--- |
|
| 61 |
+
| 🥇 | **Murasaki-8B-v0.1** | **0.8523** |
|
| 62 |
+
| 2 | Gemini-3-Flash | 0.8512 |
|
| 63 |
+
| 3 | Sakura-Qwen-2.5-14B | 0.8509 |
|
| 64 |
+
| 4 | GPT-5 | 0.8503 |
|
| 65 |
+
| 5 | Gemini-2.5-Flash | 0.8502 |
|
| 66 |
|
| 67 |
---
|
| 68 |
|
|
|
|
| 70 |
|
| 71 |
| Model Name | Type | Size | VRAM | Description |
|
| 72 |
| :--- | :--- | :--- | :--- | :--- |
|
| 73 |
+
| **[Murasaki-8B-v0.1](https://huggingface.co/Murasaki-Project/Murasaki-8B-v0.1)** | **BF16** | 15.3GB | 24GB+ | **科研/服务器版**。全精度权重。 |
|
| 74 |
+
| **[Murasaki-8B-v0.1-GGUF](https://huggingface.co/Murasaki-Project/Murasaki-8B-v0.1-GGUF)** | **GGUF** | 3.6~6.3GB | 6GB+ | **推荐/本地版**。多量化精度可选。 |
|
| 75 |
|
| 76 |
---
|
| 77 |
|