Add new SentenceTransformer model

4676c9e verified 5 months ago

24.2 kB

tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - dense
  - generated_from_trainer
  - dataset_size:42280
  - loss:MultipleNegativesRankingLoss
base_model: google/embeddinggemma-300m
widget:
  - source_sentence: >-
      How is demand from blown film converters trending for natural-colour rLDPE
      pellets sourced from production scrap in Germany?
    sentences:
      - >-
        For a tender closing Friday, market participants indicated
        post-industrial, food-grade HDPE bales could be workable around
        €1,030-1,110/t DAP Valencia for prompt-to-March delivery, depending on
        lot size and delivery flexibility.
      - >-
        Demand from German blown-film converters for natural rLDPE pellets
        sourced from production scrap was steady to slightly firmer week on
        week, though buyers continued to push back on offers above the
        low-to-mid €1,200s/t FCA level.
      - >-
        Europe recycled high-density polyethylene (R-HDPE) blow-moulding natural
        pellet demand continues to increase on the back of new packaging
        projects and increased recycled content use from the packaging sector.
  - source_sentence: >-
      What is the current premium for clean PI white HDPE bales over
      mixed-colour PI HDPE bales on an FCA Germany basis?
    sentences:
      - >-
        Nevertheless, this is not yet considered representative of the bulk of
        material, or seen as achievable for high density polyethylene (HDPE)
        dominated bales, which continue to see a top end of €300/tonne ex-works.
      - >-
        UK supply of white rHDPE blow moulding pellets is described as
        comfortable to long, as recyclers are running close to normal rates
        while demand from blow moulders remains subdued, keeping prompt
        availability open.
      - >-
        Clean post-industrial white HDPE bales are currently at a €40-70/t
        premium to mixed-colour PI HDPE bales on an FCA Germany basis, with the
        spread widest where contamination guarantees are contractually enforced.
  - source_sentence: >-
      Are you hearing more quality claims (gels, black specks, odour) on natural
      transparent flexible rLDPE pellets in the current spot market?
    sentences:
      - >-
        ICIS assessed natural rLDPE pellets produced from post-consumer LDPE
        film at €1,140-1,230/tonne delivered FD Northwest Europe in the week to
        21 February, up €10/tonne week on week on tighter prompt availability.
      - >-
        How does the quality specification of rLDPE Pellet Flexible Natural
        Translucent affect its pricing in the spot market?
      - >-
        Spot market participants reported more frequent quality claims on
        natural transparent flexible rLDPE pellets—mainly gels and sporadic
        black specks—leading to discounts or load rejections in the €20-50/t
        range.
  - source_sentence: >-
      For blown film extrusion, what MFI (190°C/2.16 kg) range is most commonly
      traded for recycled LDPE pellet, flexible, natural transparent?
    sentences:
      - >-
        Demand is strongest for natural transparent pellet, which is for
        material with an MFI of 1.0 and higher, which gives a good transparency,
        low gels and low contamination, and is suitable for stretch film
        applications.
      - >-
        For post-consumer bulky rigid HDPE bales, typical buyer specs cap
        moisture at 10%, paper/labels at 5%, metals at 0.5% and non-HD plastics
        at 5%, with PVC expected to be near-zero (often ≤0.2%) to avoid
        wash-line issues.
      - >-
        For blown film applications, recycled LDPE flexible natural/transparent
        pellets are most commonly traded at around 0.3-0.8 g/10min MFI
        (190°C/2.16kg), with some grades offered up to about 1.0 g/10min for
        easier processing.
  - source_sentence: >-
      Any new import inspections, permits, or customs delays affecting inbound
      post-consumer black HDPE bales into Vietnam this quarter?
    sentences:
      - >-
        Market sources said Vietnam has maintained tighter inspection rates and
        documentation checks on inbound post-consumer black HDPE bales this
        quarter, extending customs clearance to roughly 10-15 working days in
        some cases, but without a major new permit requirement.
      - >-
        Deals and discussions for spot imports of HDPE have taken place at
        $960-970/tonne CFR (cost & freight) Vietnam, compared with the
        $1,030-1,050/tonne CFR Vietnam assessment range in the week ended 15
        September.
      - >-
        Container freight indications from Ningbo to Hamburg were heard at
        roughly $2,000-2,300 per 40ft this week (about $80-100/t), adding to CIF
        Europe ideas for imported white rHDPE blow moulding pellets even as
        netbacks stayed steady.
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("Netizine/icis_commodity_embedding")
# Run inference
queries = [
    "Any new import inspections, permits, or customs delays affecting inbound post-consumer black HDPE bales into Vietnam this quarter?",
]
documents = [
    'Market sources said Vietnam has maintained tighter inspection rates and documentation checks on inbound post-consumer black HDPE bales this quarter, extending customs clearance to roughly 10-15 working days in some cases, but without a major new permit requirement.',
    'Deals and discussions for spot imports of HDPE have taken place at $960-970/tonne CFR (cost & freight) Vietnam, compared with the $1,030-1,050/tonne CFR Vietnam assessment range in the week ended 15 September.',
    'Container freight indications from Ningbo to Hamburg were heard at roughly $2,000-2,300 per 40ft this week (about $80-100/t), adding to CIF Europe ideas for imported white rHDPE blow moulding pellets even as netbacks stayed steady.',
]
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([[0.9667, 0.0020, 0.0885]])

Training Details

Training Dataset

Unnamed Dataset

Size: 42,280 training samples
Columns: anchor, positive, and negative

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative
type	string	string	string
details	min: 18 tokens mean: 30.35 tokens max: 50 tokens	min: 33 tokens mean: 57.04 tokens max: 255 tokens	min: 10 tokens mean: 33.34 tokens max: 96 tokens

Samples:

anchor	positive	negative
`Can you give the latest ICIS range for rLDPE pellets, post-consumer, colourless, FCA Rotterdam/Antwerp?`	`ICIS assessed post-consumer colourless rLDPE pellets at €1,070-1,170/t FCA Rotterdam/Antwerp in the week to 21 February 2026.`	`ICIS began pricing R-HDPE natural blow-moulding pellets in May 2020.`
`What are traders paying this week for PCR LDPE colourless pellets (post-consumer) delivered to Lombardy, Italy (DDP)?`	`This week, post-consumer colourless PCR LDPE pellet business was heard around €1,180-1,280/t DDP Lombardy, with better-filtered, low-odour lots at the top of the range.`	`Colourless (C) polyethylene terephthalate (PET) post-consumer bottle bale prices have increased in Italy in the latest monthly auction, adding upwards pressure to both recycled PET (R-PET) C flake and food-grade pellet (FGP) prices in the country, which reflects a common theme for January being felt across the wider European market.`
`How did the weekly assessment for post-consumer colourless rLDPE pellets in NWE change versus last week?`	`The ICIS weekly assessment for post-consumer colourless rLDPE pellets FCA Rotterdam/Antwerp fell by €20/t week on week to €1,070-1,170/t as buyers resisted higher offers.`	`Consumption of flexible R-LDPE pellets has increased in April in NWE, although this remains counterbalanced by high stock levels.`

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: 16
learning_rate: 2e-05
num_train_epochs: 5
warmup_ratio: 0.1
prompts: {'anchor': 'task: search result | query: ', 'positive': 'title: none | text: ', 'negative': 'title: none | text: '}

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: no
prediction_loss_only: True
per_device_train_batch_size: 16
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
use_ipex: 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
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: {'anchor': 'task: search result | query: ', 'positive': 'title: none | text: ', 'negative': 'title: none | text: '}
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Click to expand

Epoch	Step	Training Loss
0.0378	100	0.0336
0.0757	200	0.0013
0.1135	300	0.0009
0.1513	400	0.0015
0.1892	500	0.0019
0.2270	600	0.0013
0.2649	700	0.0034
0.3027	800	0.0046
0.3405	900	0.0007
0.3784	1000	0.0009
0.4162	1100	0.0022
0.4540	1200	0.0107
0.4919	1300	0.0081
0.5297	1400	0.0111
0.5675	1500	0.0052
0.6054	1600	0.0013
0.6432	1700	0.0108
0.6810	1800	0.0055
0.7189	1900	0.0042
0.7567	2000	0.0056
0.7946	2100	0.0034
0.8324	2200	0.0051
0.8702	2300	0.0021
0.9081	2400	0.0022
0.9459	2500	0.0089
0.9837	2600	0.0036
1.0216	2700	0.0013
1.0594	2800	0.0019
1.0972	2900	0.0015
1.1351	3000	0.0008
1.1729	3100	0.001
1.2107	3200	0.003
1.2486	3300	0.0013
1.2864	3400	0.0017
1.3243	3500	0.0008
1.3621	3600	0.0004
1.3999	3700	0.0012
1.4378	3800	0.0022
1.4756	3900	0.0032
1.5134	4000	0.0011
1.5513	4100	0.0016
1.5891	4200	0.0014
1.6269	4300	0.0024
1.6648	4400	0.0038
1.7026	4500	0.0015
1.7404	4600	0.0008
1.7783	4700	0.001
1.8161	4800	0.0006
1.8540	4900	0.0011
1.8918	5000	0.001
1.9296	5100	0.001
1.9675	5200	0.0013
2.0053	5300	0.0011
2.0431	5400	0.0005
2.0810	5500	0.0004
2.1188	5600	0.0009
2.1566	5700	0.001
2.1945	5800	0.0005
2.2323	5900	0.0012
2.2701	6000	0.0024
2.3080	6100	0.0006
2.3458	6200	0.0002
2.3837	6300	0.0005
2.4215	6400	0.0003
2.4593	6500	0.0004
2.4972	6600	0.0003
2.5350	6700	0.0006
2.5728	6800	0.0005
2.6107	6900	0.0005
2.6485	7000	0.0004
2.6863	7100	0.0007
2.7242	7200	0.0005
2.7620	7300	0.0003
2.7998	7400	0.0005
2.8377	7500	0.0007
2.8755	7600	0.0009
2.9134	7700	0.0002
2.9512	7800	0.0001
2.9890	7900	0.0012
3.0269	8000	0.0004
3.0647	8100	0.0014
3.1025	8200	0.0003
3.1404	8300	0.0004
3.1782	8400	0.0003
3.2160	8500	0.0002
3.2539	8600	0.0003
3.2917	8700	0.0002
3.3295	8800	0.0001
3.3674	8900	0.0003
3.4052	9000	0.0002
3.4431	9100	0.0001
3.4809	9200	0.0001
3.5187	9300	0.0003
3.5566	9400	0.0006
3.5944	9500	0.0002
3.6322	9600	0.0001
3.6701	9700	0.0002
3.7079	9800	0.0007
3.7457	9900	0.0001
3.7836	10000	0.0001
3.8214	10100	0.0002
3.8593	10200	0.0003
3.8971	10300	0.0001
3.9349	10400	0.0002
3.9728	10500	0.0001
4.0106	10600	0.0002
4.0484	10700	0.0002
4.0863	10800	0.0001
4.1241	10900	0.0003
4.1619	11000	0.0001
4.1998	11100	0.0001
4.2376	11200	0.0002
4.2754	11300	0.0004
4.3133	11400	0.0003
4.3511	11500	0.0001
4.3890	11600	0.0001
4.4268	11700	0.0001
4.4646	11800	0.0001
4.5025	11900	0.0005
4.5403	12000	0.0004
4.5781	12100	0.0001
4.6160	12200	0.0002
4.6538	12300	0.0001
4.6916	12400	0.0001
4.7295	12500	0.0001
4.7673	12600	0.0001
4.8051	12700	0.0
4.8430	12800	0.0001
4.8808	12900	0.0007
4.9187	13000	0.0001
4.9565	13100	0.0008
4.9943	13200	0.0001

Framework Versions

Python: 3.12.12
Sentence Transformers: 5.2.3
Transformers: 4.57.0.dev0
PyTorch: 2.10.0+cu128
Accelerate: 1.12.0
Datasets: 4.5.0
Tokenizers: 0.22.2

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}
}