Instructions to use huggingface-mic/embeddinggemma-300m-mic with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- sentence-transformers
How to use huggingface-mic/embeddinggemma-300m-mic with sentence-transformers:
from sentence_transformers import SentenceTransformer model = SentenceTransformer("huggingface-mic/embeddinggemma-300m-mic") sentences = [ "The weather is lovely today.", "It's so sunny outside!", "He drove to the stadium." ] embeddings = model.encode(sentences) similarities = model.similarity(embeddings, embeddings) print(similarities.shape) # [3, 3] - Notebooks
- Google Colab
- Kaggle
tarekziade HF Staff
Publish MIC usage, slim fixture and reviewed Transformers expectations
ffb7204 verified |
Download integrations/transformers/usage.md from huggingface-mic/embeddinggemma-300m-mic: direct link, hf CLI and curl.
- Browser
- Download file 4.97 kB
-
https://huggingface.co/huggingface-mic/embeddinggemma-300m-mic/resolve/main/integrations/transformers/usage.md
- Command line
-
hf download hf://huggingface-mic/embeddinggemma-300m-mic/integrations/transformers/usage.md
-
curl -L -o usage.md https://huggingface.co/huggingface-mic/embeddinggemma-300m-mic/resolve/main/integrations/transformers/usage.md
4.97 kB
| # EmbeddingGemma: sentence embeddings with Transformers | |
| This page is the Transformers contract of the EmbeddingGemma MIC and also a usage example. [doc-builder](https://github.com/huggingface/doc-builder/blob/main/docs/runnable-code-blocks.md) renders the `runnable` blocks below as documentation, and its pytest plugin runs them as written against `google/embeddinggemma-300m`. The MIC runner runs the same blocks against the slim fixture and the checkpoint in [`integration.json`](../../integration.json). Lines marked `# doc-builder: hide` only matter there: they swap in the pinned checkpoint or variant and print the result record. | |
| ## Embed queries and documents | |
| ```py runnable:embed | |
| import torch | |
| from transformers import AutoModel, AutoTokenizer, set_seed | |
| import json # doc-builder: hide | |
| import os # doc-builder: hide | |
| model_id = "google/embeddinggemma-300m" | |
| model_id = os.environ.get("MIC_MODEL", model_id) # doc-builder: hide | |
| variant = json.loads(os.environ.get("MIC_MODEL_KWARGS", "{}")) # doc-builder: hide | |
| device = "cuda" if torch.cuda.is_available() else "cpu" | |
| set_seed(0) # seeds torch, NumPy, and Python: loading and dropout draw random numbers | |
| tokenizer = AutoTokenizer.from_pretrained(model_id) | |
| model = AutoModel.from_pretrained(model_id).to(device) | |
| if variant: # doc-builder: hide | |
| del model # doc-builder: hide | |
| model = AutoModel.from_pretrained(model_id, **variant).to(device) # doc-builder: hide | |
| model.eval() | |
| query = "Which planet is known as the Red Planet?" | |
| documents = [ | |
| "Venus is often called Earth's twin because of its similar size and proximity.", | |
| "Mars, known for its reddish appearance, is often referred to as the Red Planet.", | |
| "Jupiter, the largest planet in our solar system, has a prominent red spot.", | |
| "Saturn, famous for its rings, is sometimes mistaken for the Red Planet.", | |
| ] | |
| # EmbeddingGemma uses a prompt prefix for queries; documents are embedded as-is. | |
| query_inputs = tokenizer(["task: search result | query: " + query], return_tensors="pt", padding=True).to(device) | |
| doc_inputs = tokenizer(documents, return_tensors="pt", padding=True).to(device) | |
| def mean_pool(outputs, attention_mask): | |
| hidden = outputs.last_hidden_state.float() | |
| mask = attention_mask.unsqueeze(-1).float() | |
| return (hidden * mask).sum(dim=1) / mask.sum(dim=1) | |
| with torch.no_grad(): | |
| query_embedding = mean_pool(model(**query_inputs), query_inputs["attention_mask"]) | |
| document_embeddings = mean_pool(model(**doc_inputs), doc_inputs["attention_mask"]) | |
| query_embedding = torch.nn.functional.normalize(query_embedding, dim=-1) | |
| document_embeddings = torch.nn.functional.normalize(document_embeddings, dim=-1) | |
| similarities = query_embedding @ document_embeddings.T | |
| print(query_embedding.shape, document_embeddings.shape) | |
| print(similarities) | |
| ``` | |
| ## One training step | |
| ```py runnable:embed:2 | |
| # A contrastive-style loss: pull the matching planet (index 1) towards the query. | |
| # Recompute the embeddings with gradients: the inference ones ran under no_grad. | |
| model.train() | |
| query = torch.nn.functional.normalize(mean_pool(model(**query_inputs), query_inputs["attention_mask"]), dim=-1) | |
| docs = torch.nn.functional.normalize(mean_pool(model(**doc_inputs), doc_inputs["attention_mask"]), dim=-1) | |
| target = torch.tensor([1], device=device) | |
| loss = torch.nn.functional.cross_entropy(query @ docs.T / 0.05, target) | |
| optimizer = torch.optim.SGD(model.parameters(), lr=1e-4) | |
| optimizer.zero_grad() | |
| loss.backward() | |
| before = {n: p.detach().cpu().clone() for n, p in model.named_parameters() if p.requires_grad} # doc-builder: hide | |
| optimizer.step() | |
| print(f"loss: {loss.item():.4f}") | |
| if os.environ.get("MIC_RUNNER"): # doc-builder: hide | |
| named = dict(model.named_parameters()) # doc-builder: hide | |
| checked = ("layers.0.self_attn.q_proj.weight", "layers.0.mlp.up_proj.weight") # doc-builder: hide | |
| result = { # doc-builder: hide | |
| "runtime": { # doc-builder: hide | |
| "device": model.device.type, # doc-builder: hide | |
| "dtype": str(model.dtype).removeprefix("torch."), # doc-builder: hide | |
| "attn_implementation": model.config._attn_implementation, # doc-builder: hide | |
| }, # doc-builder: hide | |
| "query_embedding": query_embedding[0, :8].tolist(), # doc-builder: hide | |
| "document_embeddings": document_embeddings[:, :8].tolist(), # doc-builder: hide | |
| "similarities": similarities.tolist(), # doc-builder: hide | |
| "training": { # doc-builder: hide | |
| "loss": loss.item(), # doc-builder: hide | |
| "gradients": {n: bool(named[n].grad is not None and torch.count_nonzero(named[n].grad) > 0 and torch.isfinite(named[n].grad).all()) for n in checked}, # doc-builder: hide | |
| "updated": any(not torch.equal(p.detach().cpu(), before[n]) for n, p in model.named_parameters() if n in before), # doc-builder: hide | |
| }, # doc-builder: hide | |
| } # doc-builder: hide | |
| print(json.dumps(result)) # doc-builder: hide | |
| ``` | |