Add dimensions parameter to embeddings endpoint

README.md

@@ -82,6 +82,14 @@ response = client.embeddings.create(
 )
 
 embedding = response.data[0].embedding
+
+# Optional: Specify embedding dimensions (requires Ollama v0.11.11+)
+# Example with a model that supports dynamic dimensions
+response = client.embeddings.create(
+    input="The quick brown fox jumps over the lazy dog",
+    model="qwen3-embedding",  # Use a model with MRL support
+    dimensions=512  # Custom dimension size
+)
 ```
 
 **Response (OpenAI format):**
@@ -93,6 +101,11 @@ embedding = response.data[0].embedding
 }
 ```
 
+**Note on dimensions parameter:** The `dimensions` parameter is supported in Ollama v0.11.11 and later. However, model support varies:
+- **Models with MRL (Matryoshka Representation Learning)** like Qwen3 Embedding support dynamic dimensions
+- **Fixed-dimension models** like `nomic-embed-text` (768) and `all-minilm` (384) will use their default dimensions regardless of this parameter
+- Check your model's documentation to see if it supports custom dimensions
+
 The server waits up to 30 seconds for the result. If processing takes longer, it returns a task ID for polling.
 
 #### Chat Completions
@@ -165,6 +178,7 @@ All endpoints require `Authorization: Bearer your-secret-token` header.
 |-------|------|---------|-------------|
 | `input` | string | required | Text to embed |
 | `model` | string | `nomic-embed-text` | Model name (optional) |
+| `dimensions` | integer | null | Embedding dimension size (optional, must be positive). Requires Ollama v0.11.11+. Support varies by model. |
 
 `POST /v1/chat/completions`
 

server/main.py

@@ -164,6 +164,8 @@ async def openai_embeddings(request: OpenAIEmbeddingRequest, token: str = Depend
         {"id": "task-id"} - poll GET /tasks/{id} for result
     """
     payload = {"text": request.input, "model": request.model}
+    if request.dimensions is not None:
+        payload["dimensions"] = request.dimensions
     task_id = await database.create_task("embedding", payload)
     max_wait = 30  # Fixed wait time for embeddings
     deadline = time.time() + max_wait

server/models.py

@@ -19,6 +19,7 @@ class OpenAIEmbeddingResponse(BaseModel):
 class OpenAIEmbeddingRequest(BaseModel):
     input: str  # text to embed
     model: str  # Remove default value
+    dimensions: Optional[int] = Field(default=None, gt=0)  # Optional embedding dimension size
 
 
 # Chat completion models

worker/embedder.py

@@ -2,14 +2,18 @@
 from config import OLLAMA_URL, EMBEDDING_MODEL
 
 
-def get_embedding(text: str, model: str = None) -> list[float]:
+def get_embedding(text: str, model: str = None, dimensions: int = None) -> list[float]:
     """Get embedding from Ollama API."""
+    payload = {
+        "model": model or EMBEDDING_MODEL,
+        "prompt": text,
+    }
+    if dimensions is not None:
+        payload["dimensions"] = dimensions
+
     response = requests.post(
         f"{OLLAMA_URL}/api/embeddings",
-        json={
-            "model": model or EMBEDDING_MODEL,
-            "prompt": text,
-        },
+        json=payload,
         timeout=60,
     )
     response.raise_for_status()

worker/worker.py

@@ -50,9 +50,10 @@ def process_embedding_task(task_id: str, payload: dict):
     if not isinstance(text, str):
         raise ValueError("Embedding task payload must include a 'text' field of type string.")
     model = payload.get("model")
+    dimensions = payload.get("dimensions")
     print(f"[embedding] {task_id}: {text[:50]}...")
 
-    embedding = get_embedding(text, model)
+    embedding = get_embedding(text, model, dimensions)
     complete_task(task_id, embedding)
     print(f"[embedding] {task_id} completed (dim: {len(embedding)})")