akin-em7

This is a sentence-transformers model finetuned from zerbaUst/cs-em6 on the json dataset. It maps sentences & paragraphs to a 896-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: zerbaUst/cs-em6
Maximum Sequence Length: 512 tokens
Output Dimensionality: 896 dimensions
Similarity Function: Cosine Similarity
Training Dataset:
- json
Language: en
License: apache-2.0

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': 512, 'do_lower_case': False}) with Transformer model: Qwen2Model 
  (1): Pooling({'word_embedding_dimension': 896, '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): 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("zerbaUst/akin-em7")
# Run inference
sentences = [
    'ive lost my account due to my mum making my account and shes lost the details\n\nlost email address: [EMAIL]',
    "issue no1:the player is looking to recover their account. - if there is no relevant target account information, please redirect the player to the recovery form available on the help portal.- if you received a recovery form case and have found the player's account information, please follow the dedicated processes you can find in signavio (step 1a below)issue no2:the player is looking to recover their account, but, there is suspicion of account reselling/sharing.there are a few factors you can look out for to determine whether there is a suspicion of account reselling or sharing >- the log-in country in the p360 activity log, can be split into different continuous sections between the account owner and other accesses- the owner will be reaching out to us to recover the account at all times over many cases- the owner disables 2fa, and changes the email address and the display name on the account, just before the activity country changes- there are previous account recovery cases on the account already- the owner is usually able to provide all the information required for aov- you may see duplicate game activations from purchases made by the buyer/other people who access the account from different ip'splease also note:- if a player admits they are not the original owner, we should refuse the recovery request. this includes mentioning the email address on their account belongs to someone else.- if a player mentions that they share the account with other people, we should also refuse recovery in this situation, as this account sharing breaks our terms of service.- if a player admits that they sold their account, we also deny recovery.- if a player also directly admits to purchasing an account, you may deny further support as they have admitted not being the original owner of the account (please leave a private note on the account mentioning player admitting to purchasing account the account)- if they admit to selling this account, we deny recovery also as they broke our terms of service (please leave a private note on the account mentioning the player admitting to selling the account)----------------------------------------------------------case handling:1) determine which of the 2x issues noted above, your contact falls into issue 1 or issue 2.a) for issue 1, please follow this process > https://editor.signavio.com/p/hub/model/ad240bbab2d5496c8e9e9e285b253258b) for issue 2, if you suspect the account is being shared or resold, please make sure to check the salesforce notes. if you can see an existing note mentioning account reselling or sharing on the account (placed after 08/10/2020), deny account recovery and do not proceed with further steps.if you suspect the account being shared or resold and there are no salesforce notes please perform aov (they need to pass aov) and escalate to tier 2.if you are a tier 2 specialist please follow process:\xa0https://editor.signavio.com/p/hub/model/094692df25534e7389cf86aaa89f73f8note: only escalate cases where we suspect the original account owner has sold the account. no need to escalate cases where the account selling was initiated by a hacker.2) next check if the player can pass aov and recover the account by following the process: https://goto.ubisoft.org/jtguf3) if the player passes the aov process, and we suspect that the account is being resold or shared, please make sure the player is aware of all account security measures available to them, and let them know that we may not be able to recover the account again in future. please also make sure to provide the following account security faq: ubisoft.com/help/article/0000627644) place a salesforce note on the account, that we suspect is being resold or shared.please note the usual <[account] recovery> subject line still applies to these cases.---------------------------------------------------------additional information:as we are unable to prove beyond doubt that an account is being deliberately resold or shared, please make sure you do not accuse the player directly of account reselling or sharing. rather than focusing on the reselling/sharing assumption, we would like the focus to be on account security and the fact that our previous security instructions were not followed.",
    'issue: player is reporting a cheater/hacker in gamecase handling: please thank the user for reporting the player. and advise we cannot communicate the outcome of any such investigation.please make sure that the reported username field in the ticket is filled out with the username of the suspected cheater.please advise the player to report the cheater within the game.additional information:if the player mentions being ddos during his game, please use the following kb : [report a player] cheating / ddos - 000084534 - i would like to report a player ddosing the game',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 896]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Information Retrieval

Datasets: dim_896, dim_768, dim_512, dim_256, dim_128 and dim_64
Evaluated with InformationRetrievalEvaluator

Metric	dim_896	dim_768	dim_512	dim_256	dim_128	dim_64
cosine_accuracy@1	0.0158	0.0173	0.0152	0.0149	0.0122	0.0099
cosine_accuracy@3	0.0346	0.0349	0.0331	0.0328	0.029	0.0224
cosine_accuracy@5	0.0463	0.046	0.0448	0.0439	0.0388	0.0319
cosine_accuracy@10	0.066	0.0657	0.0624	0.0615	0.0582	0.0472
cosine_precision@1	0.0158	0.0173	0.0152	0.0149	0.0122	0.0099
cosine_precision@3	0.0115	0.0116	0.011	0.0109	0.0097	0.0075
cosine_precision@5	0.0093	0.0092	0.009	0.0088	0.0078	0.0064
cosine_precision@10	0.0066	0.0066	0.0062	0.0061	0.0058	0.0047
cosine_recall@1	0.0158	0.0173	0.0152	0.0149	0.0122	0.0099
cosine_recall@3	0.0346	0.0349	0.0331	0.0328	0.029	0.0224
cosine_recall@5	0.0463	0.046	0.0448	0.0439	0.0388	0.0319
cosine_recall@10	0.066	0.0657	0.0624	0.0615	0.0582	0.0472
cosine_ndcg@10	0.0376	0.0382	0.0361	0.0354	0.0321	0.0258
cosine_mrr@10	0.029	0.0299	0.0281	0.0274	0.0242	0.0192
cosine_map@100	0.0326	0.0334	0.0318	0.0309	0.0276	0.0226

Training Details

Training Dataset

json

Dataset: json
Size: 13,396 training samples
Columns: anchor and positive
Approximate statistics based on the first 1000 samples:
anchor positive
type string string
details
min: 4 tokens
mean: 64.85 tokens
max: 512 tokens

min: 0 tokens
mean: 398.47 tokens
max: 512 tokens

	anchor	positive
type	string	string
details	min: 4 tokens mean: 64.85 tokens max: 512 tokens	min: 0 tokens mean: 398.47 tokens max: 512 tokens

Samples:

anchor	positive
`the email i use for this account got hacked and i can no longer access the email. so i can not login into ubisoft and i no longer feel comfortable with that email on my account. lost email address: [EMAIL]`	issue no1:the player is looking to recover their account. - if there is no relevant target account information, please redirect the player to the recovery form available on the help portal.- if you received a recovery form case and have found the player's account information, please follow the dedicated processes you can find in signavio (step 1a below)issue no2:the player is looking to recover their account, but, there is suspicion of account reselling/sharing.there are a few factors you can look out for to determine whether there is a suspicion of account reselling or sharing >- the log-in country in the p360 activity log, can be split into different continuous sections between the account owner and other accesses- the owner will be reaching out to us to recover the account at all times over many cases- the owner disables 2fa, and changes the email address and the display name on the account, just before the activity country changes- there are previous account recovery cases on the a...
`i am unable to change my email on my account as the old email i no longer have access to it. the old email is showing on my account as [EMAIL] but i need it to be changed to [EMAIL].`	issue no1:the player is looking to recover their account. - if there is no relevant target account information, please redirect the player to the recovery form available on the help portal.- if you received a recovery form case and have found the player's account information, please follow the dedicated processes you can find in signavio (step 1a below)issue no2:the player is looking to recover their account, but, there is suspicion of account reselling/sharing.there are a few factors you can look out for to determine whether there is a suspicion of account reselling or sharing >- the log-in country in the p360 activity log, can be split into different continuous sections between the account owner and other accesses- the owner will be reaching out to us to recover the account at all times over many cases- the owner disables 2fa, and changes the email address and the display name on the account, just before the activity country changes- there are previous account recovery cases on the a...
`It seems like there is no text provided for me to process. Please provide the support ticket text that you would like me to mask.`	issuebased on the customer's description, we are unclear on what the issue might be and we need to ask them for more information, to help us find the correct kbcase handling1) check what information the player has provided us with in their first message, and check keywords in sowa to see if you can find a relevant kbif the player and issue is unclear, please ask for further information, such as >is this issue related to their ubisoft account / a purchase or subscription / missing content / a ban or player report / a game bug / a technical issue / feedback on a game or our servicescan the player further explain what the problem is?do they have any screenshots or videos they can provide us with? are they seeing any error messages on their side they can share with us?2) note: this is a placeholder subject line and the sl should always be updated on the case, once we receive enough info to confirm if we have a relevant kb we can use.if the issue turns out to be the player is reporting an u...

Loss: MatryoshkaLoss with these parameters:

{
    "loss": "MultipleNegativesRankingLoss",
    "matryoshka_dims": [
        896,
        768,
        512,
        256,
        128,
        64
    ],
    "matryoshka_weights": [
        1,
        1,
        1,
        1,
        1,
        1
    ],
    "n_dims_per_step": -1
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: epoch
per_device_train_batch_size: 2
per_device_eval_batch_size: 2
gradient_accumulation_steps: 32
learning_rate: 2e-05
num_train_epochs: 4
lr_scheduler_type: cosine
warmup_ratio: 0.1
fp16: True
tf32: False
optim: adamw_torch_fused
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: epoch
prediction_loss_only: True
per_device_train_batch_size: 2
per_device_eval_batch_size: 2
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 32
eval_accumulation_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: 4
max_steps: -1
lr_scheduler_type: cosine
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: True
fp16_opt_level: O1
half_precision_backend: auto
bf16_full_eval: False
fp16_full_eval: False
tf32: False
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}
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: False
hub_always_push: False
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
include_inputs_for_metrics: False
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
dispatch_batches: None
split_batches: None
include_tokens_per_second: False
include_num_input_tokens_seen: False
neftune_noise_alpha: None
optim_target_modules: None
batch_eval_metrics: False
prompts: None
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	dim_896_cosine_ndcg@10	dim_768_cosine_ndcg@10	dim_512_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.0478	10	0.7342	-	-	-	-	-	-
0.0956	20	0.359	-	-	-	-	-	-
0.1433	30	0.6206	-	-	-	-	-	-
0.1911	40	0.3286	-	-	-	-	-	-
0.2389	50	0.4635	-	-	-	-	-	-
0.2867	60	0.4779	-	-	-	-	-	-
0.3344	70	0.6539	-	-	-	-	-	-
0.3822	80	0.5646	-	-	-	-	-	-
0.4300	90	0.5571	-	-	-	-	-	-
0.4778	100	0.4717	-	-	-	-	-	-
0.5255	110	0.3666	-	-	-	-	-	-
0.5733	120	0.692	-	-	-	-	-	-
0.6211	130	0.6166	-	-	-	-	-	-
0.6689	140	0.618	-	-	-	-	-	-
0.7166	150	0.4731	-	-	-	-	-	-
0.7644	160	0.5375	-	-	-	-	-	-
0.8122	170	0.4384	-	-	-	-	-	-
0.8600	180	0.4214	-	-	-	-	-	-
0.9077	190	0.7847	-	-	-	-	-	-
0.9555	200	0.7723	-	-	-	-	-	-
0.9985	209	-	0.0381	0.0366	0.0382	0.0369	0.0343	0.0271
1.0033	210	0.5171	-	-	-	-	-	-
1.0511	220	0.5229	-	-	-	-	-	-
1.0988	230	0.3208	-	-	-	-	-	-
1.1466	240	0.361	-	-	-	-	-	-
1.1944	250	0.1921	-	-	-	-	-	-
1.2422	260	0.2428	-	-	-	-	-	-
1.2899	270	0.214	-	-	-	-	-	-
1.3377	280	0.5747	-	-	-	-	-	-
1.3855	290	0.4278	-	-	-	-	-	-
1.4333	300	0.2921	-	-	-	-	-	-
1.4810	310	0.3406	-	-	-	-	-	-
1.5288	320	0.3055	-	-	-	-	-	-
1.5766	330	0.4052	-	-	-	-	-	-
1.6244	340	0.3753	-	-	-	-	-	-
1.6721	350	0.2922	-	-	-	-	-	-
1.7199	360	0.324	-	-	-	-	-	-
1.7677	370	0.2779	-	-	-	-	-	-
1.8155	380	0.3366	-	-	-	-	-	-
1.8632	390	0.4493	-	-	-	-	-	-
1.9110	400	0.3796	-	-	-	-	-	-
1.9588	410	0.4291	-	-	-	-	-	-
1.9970	418	-	0.0378	0.0387	0.0361	0.0346	0.0309	0.0257
2.0066	420	0.3842	-	-	-	-	-	-
2.0543	430	0.4343	-	-	-	-	-	-
2.1021	440	0.3238	-	-	-	-	-	-
2.1499	450	0.2563	-	-	-	-	-	-
2.1977	460	0.3092	-	-	-	-	-	-
2.2454	470	0.2376	-	-	-	-	-	-
2.2932	480	0.2644	-	-	-	-	-	-
2.3410	490	0.5582	-	-	-	-	-	-
2.3888	500	0.3216	-	-	-	-	-	-
2.4365	510	0.2821	-	-	-	-	-	-
2.4843	520	0.2969	-	-	-	-	-	-
2.5321	530	0.2768	-	-	-	-	-	-
2.5799	540	0.3804	-	-	-	-	-	-
2.6277	550	0.3968	-	-	-	-	-	-
2.6754	560	0.2676	-	-	-	-	-	-
2.7232	570	0.3127	-	-	-	-	-	-
2.7710	580	0.2596	-	-	-	-	-	-
2.8188	590	0.3421	-	-	-	-	-	-
2.8665	600	0.493	-	-	-	-	-	-
2.9143	610	0.3426	-	-	-	-	-	-
2.9621	620	0.4613	-	-	-	-	-	-
2.9955	627	-	0.0363	0.0368	0.0358	0.0348	0.0319	0.0253
3.0099	630	0.3526	-	-	-	-	-	-
3.0576	640	0.4347	-	-	-	-	-	-
3.1054	650	0.3257	-	-	-	-	-	-
3.1532	660	0.2329	-	-	-	-	-	-
3.2010	670	0.3199	-	-	-	-	-	-
3.2487	680	0.2374	-	-	-	-	-	-
3.2965	690	0.2711	-	-	-	-	-	-
3.3443	700	0.5732	-	-	-	-	-	-
3.3921	710	0.293	-	-	-	-	-	-
3.4398	720	0.2809	-	-	-	-	-	-
3.4876	730	0.3323	-	-	-	-	-	-
3.5354	740	0.2609	-	-	-	-	-	-
3.5832	750	0.3763	-	-	-	-	-	-
3.6309	760	0.3886	-	-	-	-	-	-
3.6787	770	0.2631	-	-	-	-	-	-
3.7265	780	0.3211	-	-	-	-	-	-
3.7743	790	0.2488	-	-	-	-	-	-
3.8220	800	0.3503	-	-	-	-	-	-
3.8698	810	0.4986	-	-	-	-	-	-
3.9176	820	0.3986	-	-	-	-	-	-
3.9654	830	0.4216	-	-	-	-	-	-
3.9940	836	-	0.0376	0.0382	0.0361	0.0354	0.0321	0.0258

Framework Versions

Python: 3.10.11
Sentence Transformers: 3.3.1
Transformers: 4.41.2
PyTorch: 2.1.2+cu121
Accelerate: 0.33.0
Datasets: 3.0.1
Tokenizers: 0.19.1

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",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning},
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

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}
}

zerbaUst
/

akin-em7