Update model card: full instTrans prompt documentation
Browse files
README.md
CHANGED
|
@@ -38,7 +38,7 @@ Measured on an NVIDIA A100 against the original BF16 checkpoint (greedy decoding
|
|
| 38 |
vllm serve IndexTeam/Index-Translate-2B-FP8 --host 127.0.0.1 --port 8000 --max-model-len 4096
|
| 39 |
```
|
| 40 |
|
| 41 |
-
Translation prompt format (greedy decoding, temperature=0 recommended):
|
| 42 |
|
| 43 |
```text
|
| 44 |
请将以下文本翻译为{target-language}，直接输出翻译结果，不要进行任何解释。
|
|
@@ -46,6 +46,16 @@ Translation prompt format (greedy decoding, temperature=0 recommended):
|
|
| 46 |
{source-text}
|
| 47 |
```
|
| 48 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 49 |
See the [base model card](https://huggingface.co/IndexTeam/Index-Translate-2B) for the full instTrans constrained-translation format and serving presets. GGUF builds for local inference are published in [Index-Translate-2B-GGUF](https://huggingface.co/IndexTeam/Index-Translate-2B-GGUF).
|
| 50 |
|
| 51 |
*Quantized and published by the Index team, 2026-10-03.*
|
|
|
|
| 38 |
vllm serve IndexTeam/Index-Translate-2B-FP8 --host 127.0.0.1 --port 8000 --max-model-len 4096
|
| 39 |
```
|
| 40 |
|
| 41 |
+
Translation prompt format (greedy decoding, temperature=0 recommended; chat template with `enable_thinking: false`):
|
| 42 |
|
| 43 |
```text
|
| 44 |
请将以下文本翻译为{target-language}，直接输出翻译结果，不要进行任何解释。
|
|
|
|
| 46 |
{source-text}
|
| 47 |
```
|
| 48 |
|
| 49 |
+
## Prompting & constrained translation (instTrans)
|
| 50 |
+
|
| 51 |
+
Beyond plain translation, the models follow the **instTrans** constrained-translation format. The official client wraps requests into the canonical structure `【源文】<text>` + numbered `1. 【硬性要求】<hard constraints>` + `2. 【注意】<soft constraints>` + suffix instructions:
|
| 52 |
+
|
| 53 |
+
- **Hard constraints** (binary, must hold): strict terminology glossary enforcement (e.g. `碳纤维:carbon fiber, 抗裂缝:crack resistance`), and format/structure preservation for JSON/CSV/code/placeholders.
|
| 54 |
+
- **Soft constraints** (graded): tone & style adaptation (e.g. formal business-email register), domain/word-sense disambiguation (e.g. *plant* → 工厂 in an industrial context), cross-sentence consistency, LaTeX preservation.
|
| 55 |
+
- **Syllable-controlled translation** (dubbing): use the Index-Homura checkpoints ([IndexTeam/Index-Homura-2B/9B](https://huggingface.co/IndexTeam/Index-Homura-2B)), which strictly respect a target syllable budget and can be combined with glossaries.
|
| 56 |
+
|
| 57 |
+
Full prompt reference: [github.com/bilibili/Index-Translate](https://github.com/bilibili/Index-Translate) (Instruction Following section, `docs/prompts.md`, `inference/llm/cases/`).
|
| 58 |
+
|
| 59 |
See the [base model card](https://huggingface.co/IndexTeam/Index-Translate-2B) for the full instTrans constrained-translation format and serving presets. GGUF builds for local inference are published in [Index-Translate-2B-GGUF](https://huggingface.co/IndexTeam/Index-Translate-2B-GGUF).
|
| 60 |
|
| 61 |
*Quantized and published by the Index team, 2026-10-03.*
|