Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -55,10 +55,10 @@ curl -X POST "http://127.0.0.1:8080/v1/embeddings" \
|
|
| 55 |
-H "Content-Type: application/json" \
|
| 56 |
-d '{
|
| 57 |
"input": [
|
| 58 |
-
"A beautiful sunset over the beach",
|
| 59 |
-
"Un beau coucher de soleil sur la plage",
|
| 60 |
-
"海滩上美丽的日落",
|
| 61 |
-
"浜辺に沈む美しい夕日"
|
| 62 |
]
|
| 63 |
}'
|
| 64 |
```
|
|
@@ -72,12 +72,24 @@ curl -X POST "http://127.0.0.1:8080/v1/embeddings" \
|
|
| 72 |
"input": [
|
| 73 |
"Query: A beautiful sunset over the beach",
|
| 74 |
"Query: Un beau coucher de soleil sur la plage",
|
| 75 |
-
"
|
| 76 |
-
"
|
| 77 |
]
|
| 78 |
}'
|
| 79 |
```
|
| 80 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 81 |
You can also use `llama-embedding` for one-shot embedding:
|
| 82 |
|
| 83 |
```bash
|
|
|
|
| 55 |
-H "Content-Type: application/json" \
|
| 56 |
-d '{
|
| 57 |
"input": [
|
| 58 |
+
"Query: A beautiful sunset over the beach",
|
| 59 |
+
"Query: Un beau coucher de soleil sur la plage",
|
| 60 |
+
"Query: 海滩上美丽的日落",
|
| 61 |
+
"Query: 浜辺に沈む美しい夕日"
|
| 62 |
]
|
| 63 |
}'
|
| 64 |
```
|
|
|
|
| 72 |
"input": [
|
| 73 |
"Query: A beautiful sunset over the beach",
|
| 74 |
"Query: Un beau coucher de soleil sur la plage",
|
| 75 |
+
"Passage: 海滩上美丽的日落",
|
| 76 |
+
"Passage: 浜辺に沈む美しい夕日"
|
| 77 |
]
|
| 78 |
}'
|
| 79 |
```
|
| 80 |
|
| 81 |
+
To get fully consistent results as if you do AutoModel.from_pretrained("jinaai/jina-embeddings-v4")..., you need to be careful about the prefix and manually add them to your input to GGUF. Here's a reference table:
|
| 82 |
+
|
| 83 |
+
| Input Type | Task | `prompt_name` (Role) | Actual Input Processed by Model |
|
| 84 |
+
|------------|------|-------------|-------------------------------|
|
| 85 |
+
| **Text** | `retrieval` | `query` (default) | `Query: {original_text}` |
|
| 86 |
+
| **Text** | `retrieval` | `passage` | `Passage: {original_text}` |
|
| 87 |
+
| **Text** | `text-matching` | `query` (default) | `Query: {original_text}` |
|
| 88 |
+
| **Text** | `text-matching` | `passage` | `Query: {original_text}` ⚠️ |
|
| 89 |
+
| **Text** | `code` | `query` (default) | `Query: {original_text}` |
|
| 90 |
+
| **Text** | `code` | `passage` | `Passage: {original_text}` |
|
| 91 |
+
| **Image** | Any task | N/A | `<\|im_start\|>user\n<\|vision_start\|>\<\|image_pad\|>\<\|vision_end\|>Describe the image.\<\|im_end\|>` |
|
| 92 |
+
|
| 93 |
You can also use `llama-embedding` for one-shot embedding:
|
| 94 |
|
| 95 |
```bash
|