deebak14/embedding_gemma_ft_v1

SentenceTransformer based on google/embeddinggemma-300m

This is a sentence-transformers model finetuned from google/embeddinggemma-300m on the embedding_tuple_data_v1 dataset. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: google/embeddinggemma-300m
• Maximum Sequence Length: 2048 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • embedding_tuple_data_v1

Model Sources

• Documentation: Sentence Transformers Documentation
• Repository: Sentence Transformers on GitHub
• Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 2048, 'do_lower_case': False, 'architecture': 'Gemma3TextModel'})
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): Dense({'in_features': 768, 'out_features': 3072, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity'})
  (3): Dense({'in_features': 3072, 'out_features': 768, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity'})
  (4): Normalize()
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("deebak14/embedding_gemma_ft_v1")
# Run inference
queries = [
    "I need to find where a flat surface meets a sphere. How can I do that in Rhino?",
]
documents = [
    'You can use the following methods to complete the task.\nmethod: PlaneSphereIntersection\ndescription: Calculates the intersection of a plane and a sphere.\nsyntax: PlaneSphereIntersection(plane, sphere_plane, sphere_radius)\nparameters: plane (plane): The plane to intersect; sphere_plane (plane): Equatorial plane of the sphere (origin is center); sphere_radius (float): Radius of the sphere.\nreturns: list: [type, point/plane, radius] where type=0 for point, 1 for circle. None on error.\n\nFollowing is the code that uses this method to complete the task as per user query.\n\n```python\nimport rhinoscriptsyntax as rs\n\n# Define a flat surface as a plane\nplane = rs.WorldXYPlane()\n# Define the radius of the sphere\nradius = 10\n# Calculate the intersection between the plane and the sphere\nresults = rs.PlaneSphereIntersection(plane, plane, radius)\n\n# Check if there are results and handle them accordingly\nif results:\n    if results[0] == 0:\n        # If the intersection is a point, add it to the document\n        rs.AddPoint(results[1])\n    else:\n        # If the intersection is a circle, add it to the document\n        rs.AddCircle(results[1], results[2])\n```',
    'You cannot use the following methods LineSphereIntersection, IsSphere, AddSphere, LinePlaneIntersection, Angle, CircleCenterPoint, CurveCurveIntersection, CurveSurfaceIntersection, AddCircle3Pt',
    'plane',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 768] [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[ 0.6658,  0.4819, -0.1617]])

Evaluation

Metrics

Triplet

• Dataset: base-eval
• Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	1.0

Training Details

Training Dataset

embedding_tuple_data_v1

• Dataset: embedding_tuple_data_v1 at b592a1a
• Size: 15,565 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 9 tokens mean: 15.98 tokens max: 50 tokens	min: 42 tokens mean: 177.53 tokens max: 810 tokens

• Samples:

  anchor	positive
  Provide an example of using AddRectangle.	import rhinoscriptsyntax as rs plane = rs.WorldXYPlane() plane = rs.RotatePlane(plane, 45.0, [0,0,1]) rs.AddRectangle(plane, 5.0, 15.0) Metadata: Name: AddRectangle Category: curve Function Signature: rs.AddRectangle(plane: plane, width: number, height: number) -> guid Description: Add a rectangular curve to the document
  How do I search for the total number of linetypes in my document?	You can use the following method: Name: LinetypeCount Category: linetype Function Signature: rs.LinetypeCount() -> int Description: Description: Returns the number of linetypes in the document. Parameters: None Returns: int: The number of linetypes in the document.
  How do I maintain the shape of a curve while fitting it?	You can use the following method: Name: FitCurve Category: curve Function Signature: rs.FitCurve(curve_id: guid, degree: int = 3, distance_tolerance: float = -1, angle_tolerance: float = -1) -> guid Description: Description: Reduces the number of control points of a curve while maintaining its general shape. This function is useful for replacing curves with many control points. For more information, see the Rhino help for the FitCrv command. Parameters: curve_id (guid): Identifier of the curve object to be fitted. eg: '3D4F5A6B-7C8D-9E0F-1A2B-3C4D5E6F7A8B' degree (int, optional): The degree of the curve, which must be greater than 1. The default is 3. eg: 3 distance_tolerance (float, optional): The fitting tolerance. If not specified or <= 0.0, the document absolute tolerance is used. eg: 0.01 angle_tolerance (float, optional): The kink smoothing tolerance in degrees. If 0.0, all kinks are smoothed. If > 0.0, kinks smaller than this value are smoothed. If ...

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": false
  }
  

Evaluation Dataset

embedding_triplet_data_v1

• Dataset: embedding_triplet_data_v1 at 71ea1de
• Size: 476 evaluation samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 476 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 15 tokens mean: 23.13 tokens max: 38 tokens	min: 111 tokens mean: 252.44 tokens max: 1015 tokens	min: 32 tokens mean: 42.48 tokens max: 55 tokens

• Samples:

  anchor	positive	negative
  I need to flatten a curved surface for laser cutting. How can I do that?	You can use the following methods to complete the task. method: UnrollSurface description: Flattens a developable surface or polysurface syntax: UnrollSurface(surface_id, explode=False, following_geometry=None, absolute_tolerance=None, relative_tolerance=None) parameters: surface_id (guid): the surface's identifier
  explode (bool, optional): If True, the resulting surfaces ar not joined
  following_geometry ({guid, ...]): List of curves, dots, and points which
  should be unrolled with the surface returns: list(guid, ...): of unrolled surface ids
  tuple((guid, ...),(guid, ...)): if following_geometry is not None, a tuple
  [1] is the list of unrolled surface ids
  [2] is the list of unrolled following geometry Following is the code that uses this method to complete the task as per user query. ```python import rhinoscriptsyntax as rs # Flatten a curved surface for laser cutting surface = rs.GetObject("Select curved surface to flatten", rs.filter.surface) if surface: # Unrol...	You cannot use the following methods ConvertCurveToPolyline, MeshOutline, PullCurveToMesh, ExplodeText, MeshToNurb, IsCurvePlanar, Angle, AddFilletCurve, MeshVolume
  Can you show me how to rotate a plane by 30 degrees around its normal axis?	You can use the following methods to complete the task. method: PlaneTransform description: Transforms a plane. syntax: PlaneTransform(plane, xform) parameters: plane (plane): Plane to transform; xform (transform): Transformation to apply. returns: plane: Resulting plane if successful, otherwise None. Following is the code that uses this method to complete the task as per user query. python<br>import rhinoscriptsyntax as rs<br><br># Get the current construction plane<br>plane = rs.ViewCPlane()<br><br># Create a rotation transformation of 30 degrees around the normal axis (Z-axis)<br>xform = rs.XformRotation(30.0, plane.ZAxis, plane.Origin)<br><br># Apply the transformation to the plane<br>plane = rs.PlaneTransform(plane, xform)<br><br># Set the new construction plane<br>rs.ViewCPlane(None, plane)<br>	You cannot use the following methods RotatePlane, XformRotation1, PlaneFromNormal, VectorRotate, Angle, PlaneFromFrame, LinePlane, MovePlane, CreatePlane
  I want to change the height of a text dot I just created. How can I do that?	You can use the following methods to complete the task. method: TextDotHeight description: Returns or modified the font height of a text dot syntax: TextDotHeight(object_id, height=None) parameters: object_id (guid): identifier of a text dot object
  height (number, optional) new font height returns: number: If height is not specified, the current text dot height
  number: If height is specified, the previous text dot height
  None: on error Following is the code that uses this method to complete the task as per user query. python<br>import rhinoscriptsyntax as rs<br># Change the height of a text dot<br>obj = rs.GetObject("Select text dot")<br>if rs.IsTextDot(obj):<br> previous_height = rs.TextDotHeight(obj, 15.0) # Set new height to 15.0<br> print(f"Previous height was: {previous_height}")<br>	You cannot use the following methods TextDotPoint, TextDotFont, TextDotText, TextObjectHeight, IsTextDot, AddTextDot, TextObjectFont, PointCoordinates, ExplodeText

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": false
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• learning_rate: 2e-05
• warmup_ratio: 0.1
• bf16: True
• prompts: task: sentence similarity | query:
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 2e-05
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 3
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.1
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• save_safetensors: True
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• no_cuda: False
• use_cpu: False
• use_mps_device: False
• seed: 42
• data_seed: None
• jit_mode_eval: False
• use_ipex: False
• bf16: True
• fp16: False
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: 0
• ddp_backend: None
• tpu_num_cores: None
• tpu_metrics_debug: False
• debug: []
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_prefetch_factor: None
• past_index: -1
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: False
• ignore_data_skip: False
• fsdp: []
• fsdp_min_num_params: 0
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• fsdp_transformer_layer_cls_to_wrap: None
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• parallelism_config: None
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch_fused
• optim_args: None
• adafactor: False
• group_by_length: False
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• skip_memory_metrics: True
• use_legacy_prediction_loop: False
• push_to_hub: False
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: None
• hub_always_push: False
• hub_revision: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• include_for_metrics: []
• eval_do_concat_batches: True
• fp16_backend: auto
• push_to_hub_model_id: None
• push_to_hub_organization: None
• mp_parameters:
• auto_find_batch_size: False
• full_determinism: False
• torchdynamo: None
• ray_scope: last
• ddp_timeout: 1800
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• include_tokens_per_second: False
• include_num_input_tokens_seen: False
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• liger_kernel_config: None
• eval_use_gather_object: False
• average_tokens_across_devices: False
• prompts: task: sentence similarity | query:
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss	base-eval_cosine_accuracy
-1	-1	-	-	0.0147
0.1028	100	0.1601	-	-
0.2055	200	0.0474	0.2296	0.8971
0.3083	300	0.0749	-	-
0.4111	400	0.1037	0.1457	0.9265
0.5139	500	0.0564	-	-
0.6166	600	0.0706	0.3362	0.9475
0.7194	700	0.0549	-	-
0.8222	800	0.0427	0.2154	0.9538
0.9250	900	0.0599	-	-
1.0277	1000	0.0656	0.2439	0.9706
1.1305	1100	0.0409	-	-
1.2333	1200	0.0283	0.2422	0.9727
1.3361	1300	0.0336	-	-
1.4388	1400	0.0338	0.2397	0.9664
1.5416	1500	0.0384	-	-
1.6444	1600	0.0271	0.1048	0.9832
1.7472	1700	0.0305	-	-
1.8499	1800	0.024	0.1172	0.9916
1.9527	1900	0.014	-	-
2.0555	2000	0.018	0.0898	0.9958
2.1583	2100	0.0091	-	-
2.2610	2200	0.0154	0.0721	0.9916
2.3638	2300	0.0123	-	-
2.4666	2400	0.0119	0.0876	0.9937
2.5694	2500	0.0173	-	-
2.6721	2600	0.0091	0.0482	1.0
2.7749	2700	0.0211	-	-
2.8777	2800	0.0146	0.0550	1.0
2.9805	2900	0.0101	-	-
-1	-1	-	-	1.0

Framework Versions

• Python: 3.12.11
• Sentence Transformers: 5.1.0
• Transformers: 4.56.1
• PyTorch: 2.8.0+cu126
• Accelerate: 1.10.1
• Datasets: 4.0.0
• Tokenizers: 0.22.0

Citation

BibTeX

Sentence Transformers

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}