poc-gemma / README.md

blachang28

Add new SentenceTransformer model

7c241b9 verified about 2 months ago

preview code

raw

history blame contribute delete

23.2 kB

metadata

tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - dense
  - generated_from_trainer
  - dataset_size:200
  - loss:MultipleNegativesRankingLoss
base_model: google/embeddinggemma-300m
widget:
  - source_sentence: >-
      Alice and Bob play a game involving a circle whose circumference is
      divided by 12 equally-spaced points. The points are numbered clockwise,
      from 1 to 12. Both start on point 12. Alice moves clockwise and Bob,
      counterclockwise. In a turn of the game, Alice moves 5 points clockwise
      and Bob moves 9 points counterclockwise. The game ends when they stop on
      the same point. How many turns will this take?
    sentences:
      - >-
        How many distinct triangles can be drawn using three of the dots below
        as vertices? [asy]dot(origin^^(1,0)^^(2,0)^^(0,1)^^(1,1)^^(2,1));[/asy]
      - >-
        Tess runs counterclockwise around rectangular block $JKLM$. She lives at
        corner $J$. Which graph could represent her straight-line distance from
        home?
      - Which triplet of numbers has a sum NOT equal to 1?
  - source_sentence: Which of these numbers is less than its reciprocal?
    sentences:
      - >-
        A white cylindrical silo has a diameter of 30 feet and a height of 80
        feet. A red stripe with a horizontal width of 3 feet is painted on the
        silo, as shown, making two complete revolutions around it. What is the
        area of the stripe in square feet? [asy]
        size(250);defaultpen(linewidth(0.8)); draw(ellipse(origin, 3, 1));
        fill((3,0)--(3,2)--(-3,2)--(-3,0)--cycle, white);
        draw((3,0)--(3,16)^^(-3,0)--(-3,16)); draw((0, 15)--(3, 12)^^(0,
        16)--(3, 13)); filldraw(ellipse((0, 16), 3, 1), white, black);
        draw((-3,11)--(3, 5)^^(-3,10)--(3, 4));
        draw((-3,2)--(0,-1)^^(-3,1)--(-1,-0.89)); draw((0,-1)--(0,15), dashed);
        draw((3,-2)--(3,-4)^^(-3,-2)--(-3,-4));
        draw((-7,0)--(-5,0)^^(-7,16)--(-5,16)); draw((3,-3)--(-3,-3),
        Arrows(6)); draw((-6,0)--(-6,16), Arrows(6)); draw((-2,9)--(-1,9),
        Arrows(3)); label("$3$", (-1.375,9.05), dir(260), fontsize(7));
        label("$A$", (0,15), N); label("$B$", (0,-1), NE); label("$30$", (0,
        -3), S); label("$80$", (-6, 8), W);[/asy]
      - >-
        Aunt Anna is $42$ years old. Caitlin is $5$ years younger than Brianna,
        and Brianna is half as old as Aunt Anna. How old is Caitlin?
      - >-
        Tess runs counterclockwise around rectangular block $JKLM$. She lives at
        corner $J$. Which graph could represent her straight-line distance from
        home?
  - source_sentence: >-
      Problems 8,9 and 10 use the data found in the accompanying paragraph and
      table: Juan organizes the stamps in his collection by country and by the
      decade in which they were issued. The prices he paid for them at a stamp
      shop were: Brazil and France, 6 cents each, Peru 4 cents each, and Spain 5
      cents each. (Brazil and Peru are South American countries and France and
      Spain are in Europe.) [asy] /* AMC8 2002 #8, 9, 10 Problem */ size(3inch,
      1.5inch); for ( int y = 0; y <= 5; ++y ) { draw((0,y)--(18,y)); }
      draw((0,0)--(0,5)); draw((6,0)--(6,5)); draw((9,0)--(9,5));
      draw((12,0)--(12,5)); draw((15,0)--(15,5)); draw((18,0)--(18,5));
      draw(scale(0.8)*"50s", (7.5,4.5)); draw(scale(0.8)*"4", (7.5,3.5));
      draw(scale(0.8)*"8", (7.5,2.5)); draw(scale(0.8)*"6", (7.5,1.5));
      draw(scale(0.8)*"3", (7.5,0.5)); draw(scale(0.8)*"60s", (10.5,4.5));
      draw(scale(0.8)*"7", (10.5,3.5)); draw(scale(0.8)*"4", (10.5,2.5));
      draw(scale(0.8)*"4", (10.5,1.5)); draw(scale(0.8)*"9", (10.5,0.5));
      draw(scale(0.8)*"70s", (13.5,4.5)); draw(scale(0.8)*"12", (13.5,3.5));
      draw(scale(0.8)*"12", (13.5,2.5)); draw(scale(0.8)*"6", (13.5,1.5));
      draw(scale(0.8)*"13", (13.5,0.5)); draw(scale(0.8)*"80s", (16.5,4.5));
      draw(scale(0.8)*"8", (16.5,3.5)); draw(scale(0.8)*"15", (16.5,2.5));
      draw(scale(0.8)*"10", (16.5,1.5)); draw(scale(0.8)*"9", (16.5,0.5));
      label(scale(0.8)*"Country", (3,4.5)); label(scale(0.8)*"Brazil", (3,3.5));
      label(scale(0.8)*"France", (3,2.5)); label(scale(0.8)*"Peru", (3,1.5));
      label(scale(0.8)*"Spain", (3,0.5)); label(scale(0.9)*"Juan's Stamp
      Collection", (9,0), S); label(scale(0.9)*"Number of Stamps by Decade",
      (9,5), N);[/asy] The average price of his '70s stamps is closest to
    sentences:
      - >-
        A rectangular garden 60 feet long and 20 feet wide is enclosed by a
        fence. To make the garden larger, while using the same fence, its shape
        is changed to a square. By how many square feet does this enlarge the
        garden?
      - >-
        Problems 8,9 and 10 use the data found in the accompanying paragraph and
        table: Juan organizes the stamps in his collection by country and by the
        decade in which they were issued. The prices he paid for them at a stamp
        shop were: Brazil and France, 6 cents each, Peru 4 cents each, and Spain
        5 cents each. (Brazil and Peru are South American countries and France
        and Spain are in Europe.) [asy] /* AMC8 2002 #8, 9, 10 Problem */
        size(3inch, 1.5inch); for ( int y = 0; y <= 5; ++y ) {
        draw((0,y)--(18,y)); } draw((0,0)--(0,5)); draw((6,0)--(6,5));
        draw((9,0)--(9,5)); draw((12,0)--(12,5)); draw((15,0)--(15,5));
        draw((18,0)--(18,5)); draw(scale(0.8)*"50s", (7.5,4.5));
        draw(scale(0.8)*"4", (7.5,3.5)); draw(scale(0.8)*"8", (7.5,2.5));
        draw(scale(0.8)*"6", (7.5,1.5)); draw(scale(0.8)*"3", (7.5,0.5));
        draw(scale(0.8)*"60s", (10.5,4.5)); draw(scale(0.8)*"7", (10.5,3.5));
        draw(scale(0.8)*"4", (10.5,2.5)); draw(scale(0.8)*"4", (10.5,1.5));
        draw(scale(0.8)*"9", (10.5,0.5)); draw(scale(0.8)*"70s", (13.5,4.5));
        draw(scale(0.8)*"12", (13.5,3.5)); draw(scale(0.8)*"12", (13.5,2.5));
        draw(scale(0.8)*"6", (13.5,1.5)); draw(scale(0.8)*"13", (13.5,0.5));
        draw(scale(0.8)*"80s", (16.5,4.5)); draw(scale(0.8)*"8", (16.5,3.5));
        draw(scale(0.8)*"15", (16.5,2.5)); draw(scale(0.8)*"10", (16.5,1.5));
        draw(scale(0.8)*"9", (16.5,0.5)); label(scale(0.8)*"Country", (3,4.5));
        label(scale(0.8)*"Brazil", (3,3.5)); label(scale(0.8)*"France",
        (3,2.5)); label(scale(0.8)*"Peru", (3,1.5)); label(scale(0.8)*"Spain",
        (3,0.5)); label(scale(0.9)*"Juan's Stamp Collection", (9,0), S);
        label(scale(0.9)*"Number of Stamps by Decade", (9,5), N);[/asy] In
        dollars and cents, how much did his South American stamps issued before
        the ’70s cost him?
      - >-
        Three mutually tangent spheres of radius $1$ rest on a horizontal plane.
        A sphere of radius $2$ rests on them. What is the distance from the
        plane to the top of the larger sphere?
  - source_sentence: >-
      Two-thirds of the people in a room are seated in three-fourths of the
      chairs. The rest of the people are standing. If there are $6$ empty
      chairs, how many people are in the room?
    sentences:
      - >-
        Spinners $A$ and $B$ are spun. On each spinner, the arrow is equally
        likely to land on each number. What is the probability that the product
        of the two spinners' numbers is even?
      - >-
        On the AMC 8 contest Billy answers 13 questions correctly, answers 7
        questions incorrectly and doesn't answer the last 5. What is his score?
      - >-
        How many different combinations of \$5 bills and \$2 bills can be used
        to make a total of \$17? Order does not matter in this problem.
  - source_sentence: When $1999^{2000}$ is divided by $5$, the remainder is
    sentences:
      - >-
        Square $ABCD$ has sides of length 3. Segments $CM$ and $CN$ divide the
        square's area into three equal parts. How long is segment $CM$?
      - >-
        Problems 14, 15 and 16 involve Mrs. Reed's English assignment. A Novel
        Assignment The students in Mrs. Reed's English class are reading the
        same 760-page novel. Three friends, Alice, Bob and Chandra, are in the
        class. Alice reads a page in 20 seconds, Bob reads a page in 45 seconds
        and Chandra reads a page in 30 seconds. Chandra and Bob, who each have a
        copy of the book, decide that they can save time by "team reading" the
        novel. In this scheme, Chandra will read from page 1 to a certain page
        and Bob will read from the next page through page 760, finishing the
        book. When they are through they will tell each other about the part
        they read. What is the last page that Chandra should read so that she
        and Bob spend the same amount of time reading the novel?
      - >-
        $(6?3) + 4 - (2 - 1) = 5.$ To make this statement true, the question
        mark between the 6 and the 3 should be replaced by
pipeline_tag: sentence-similarity
library_name: sentence-transformers

SentenceTransformer based on google/embeddinggemma-300m

This is a sentence-transformers model finetuned from google/embeddinggemma-300m. 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

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("blachang28/poc-gemma")
# Run inference
queries = [
    "When $1999^{2000}$ is divided by $5$, the remainder is",
]
documents = [
    "Square $ABCD$ has sides of length 3. Segments $CM$ and $CN$ divide the square's area into three equal parts. How long is segment $CM$?",
    '$(6?3) + 4 - (2 - 1) = 5.$ To make this statement true, the question mark between the 6 and the 3 should be replaced by',
    'Problems 14, 15 and 16 involve Mrs. Reed\'s English assignment. A Novel Assignment The students in Mrs. Reed\'s English class are reading the same 760-page novel. Three friends, Alice, Bob and Chandra, are in the class. Alice reads a page in 20 seconds, Bob reads a page in 45 seconds and Chandra reads a page in 30 seconds. Chandra and Bob, who each have a copy of the book, decide that they can save time by "team reading" the novel. In this scheme, Chandra will read from page 1 to a certain page and Bob will read from the next page through page 760, finishing the book. When they are through they will tell each other about the part they read. What is the last page that Chandra should read so that she and Bob spend the same amount of time reading the novel?',
]
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([[ 0.7710, -0.4438, -0.0152]])

Training Details

Training Dataset

Unnamed Dataset

Size: 200 training samples
Columns: anchor, positive, and negative

Approximate statistics based on the first 200 samples:

	anchor	positive	negative
type	string	string	string
details	min: 11 tokens mean: 81.76 tokens max: 806 tokens	min: 11 tokens mean: 80.08 tokens max: 806 tokens	min: 12 tokens mean: 79.01 tokens max: 797 tokens

Samples:

anchor	positive	negative
`$(6?3) + 4 - (2 - 1) = 5.$ To make this statement true, the question mark between the 6 and the 3 should be replaced by`	`What is the degree measure of the smaller angle formed by the hands of a clock at 10 o'clock?`	`An insect lives on the surface of a regular tetrahedron with edges of length 1. It wishes to travel on the surface of the tetrahedron from the midpoint of one edge to the midpoint of the opposite edge. What is the length of the shortest such trip? (Note: Two edges of a tetrahedron are opposite if they have no common endpoint.)`
`What is the degree measure of the smaller angle formed by the hands of a clock at 10 o'clock?`	`Which triplet of numbers has a sum NOT equal to 1?`	`Corners are sliced off a unit cube so that the six faces each become regular octagons. What is the total volume of the removed tetrahedra?`
`Which triplet of numbers has a sum NOT equal to 1?`	`What is the degree measure of the smaller angle formed by the hands of a clock at 10 o'clock?`	`How many pairs of positive integers $(a,b)$ are there such that $\text{gcd}(a,b)=1$ and $\frac{a}{b} + \frac{14b}{9a}$ is an integer? $\mathrm {(A)}\ 4\quad\mathrm {(B)}\ 6\quad\mathrm {(C)}\ 9\quad\mathrm {(D)}\ 12\quad\mathrm {(E)}\ \text{infinitely many}$`

Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 1
learning_rate: 2e-05
num_train_epochs: 5
warmup_ratio: 0.1
prompts: task: classification | query:

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: no
prediction_loss_only: True
per_device_train_batch_size: 1
per_device_eval_batch_size: 8
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: 5
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
bf16: False
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
project: huggingface
trackio_space_id: trackio
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: no
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: True
prompts: task: classification | query:
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss
1.0	200	1.3954
2.0	400	0.8112
3.0	600	0.0855
4.0	800	0.0529
5.0	1000	0.0018

Framework Versions

Python: 3.12.12
Sentence Transformers: 5.1.2
Transformers: 4.57.2
PyTorch: 2.9.0+cu126
Accelerate: 1.12.0
Datasets: 4.0.0
Tokenizers: 0.22.1

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}
}