cross-encoder/ms-marco-MiniLM-L2-v2 Finetuned on PV211 HomeWork

This is a Cross Encoder model finetuned from cross-encoder/ms-marco-MiniLM-L2-v2 using the sentence-transformers library. It computes scores for pairs of texts, which can be used for text reranking and semantic search.

Model Details

Model Description

Model Type: Cross Encoder
Base model: cross-encoder/ms-marco-MiniLM-L2-v2
Maximum Sequence Length: 512 tokens
Number of Output Labels: 1 label
Language: en
License: apache-2.0

Model Sources

Documentation: Sentence Transformers Documentation
Documentation: Cross Encoder Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Cross Encoders on Hugging Face

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 CrossEncoder

# Download from the 🤗 Hub
model = CrossEncoder("maennyn/pv211_beir_cqadupstack_crossencoder")
# Get scores for pairs of texts
pairs = [
    ['Do elevator upgrades increase your passive credit earnings, too?', 'I searched for a solution for this problem, but cannot find an answer (or exact replica of the problem) Basically, I set up Multisite on MAMP Pro (Apache port 80, MySQL Port 3306). The set up was smooth, and I created a new site via a subdirectory. The parent theme loads fine. I created a child theme, and it activates (it doesn\'t show a broken message). On the Appearance page it shows the message "This theme requires the parent theme", but underneath the Theme Description. However when I view the front page of the site, the page is blank, and there is no html at all. Would could possibly be the error? I spent a few hours on this already and it\'s not going really well. Code of child theme, only CSS, no functions.php or other php files in the child theme folder.               /*     Theme Name: Confit Child Theme     Author: Automattic     Template: confit     Description: Confit Child Theme 1     Version: 1.0     */          @import url(\'../confit/style.css\');        * Should also mention that the parent functions are not loading either.  Thanks!'],
    ['Traceback (most recent call last) error appears on terminal', "I've got a binary characteristic and a population $S$ with size $n$ and $P[X] = p$ such that $p$ may be small and $n$ is extremely large. Within this population are subpopulations of various sizes $S_0, S_1, \\dots, S_k \\subset S$. I'd like to be able to select each subpopulation in which $p_i < p$ with some concept of statistical significance. My first inclination is to observe that the standard error on each $p_i$ is $SE_i = \\sqrt{\\frac{\\hat{p_i}(1-\\hat{p_i})}{n}}$ and to compare upper bounds on confidence intervals. $\\{S_i \\; | \\; \\hat{p_i} + 3 \\cdot SE_i < p\\}$, for example. But when $\\hat{p_i} = 0$, then $SE_i = 0$, and this upper bound is 0 even for the smallest subpopulations (like those where $n_i = 1$). Is there any way to express uncertainty in $p_i$ when $\\hat{p_i} = 0$? Maybe through use of $p$ as a prior? **Edit:** It looks like the Jeffreys interval as described in Brown et al. is about what I'm after, though I'm not as-of-yet sure how to apply it."],
    ['Do I have to install a custom ROM if I root?', 'What is the difference between a battery and a charged capacitor? I can see lot of similarities between capacitor and battery. In both these charges are separated and When not connected in a circuit both can have same Potential difference `V`. The only difference is that battery runs for longer time but a capacitor discharges almost instantaneously. Why this difference? What is the exact cause for the difference in the discharge times?'],
    ['How to seprate words into two lines in one cell?', 'To me the word "curious" would be something you can be i.e. > I am curious what tomorrow will bring I recently read a text of a student I was supervising which used it as follows > A curious phenomenon is ... With which he meant to say that the phenomenon was peculiar, odd or strange. The only other case I have ever seen this is in the movie title: "The Curious Case of Benjamin Button", but that might be \'artistic freedom\' (since Curious Case has the nice C.. C..). My question is: is the usage of the word "curious" in the meaning of peculiar correct?'],
    ["Bought game on Steam, but it's not in my Library", "I'm looking to choose open source project hosting site for an F# project using SVN. CodePlex is where the .NET community in general and most F# projects are hosted, but I'm worried TFS + SvnBridge is going to give me headaches. So I'm looking elsewhere and seeking advice here. Or if you think CodePlex is still the best choice in my scenario, I'd like to hear that too. So far, Google Code is looking appealing to me. They have a clean interface and true SVN hosting. But there are close to no F# projects currently hosted (it's not even in their search by programming language list), so I'm wondering if there are any notable downsides besides the lack of community I might encounter. If there is yet another option, I'd like to hear that too. Thanks!"],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'Do elevator upgrades increase your passive credit earnings, too?',
    [
        'I searched for a solution for this problem, but cannot find an answer (or exact replica of the problem) Basically, I set up Multisite on MAMP Pro (Apache port 80, MySQL Port 3306). The set up was smooth, and I created a new site via a subdirectory. The parent theme loads fine. I created a child theme, and it activates (it doesn\'t show a broken message). On the Appearance page it shows the message "This theme requires the parent theme", but underneath the Theme Description. However when I view the front page of the site, the page is blank, and there is no html at all. Would could possibly be the error? I spent a few hours on this already and it\'s not going really well. Code of child theme, only CSS, no functions.php or other php files in the child theme folder.               /*     Theme Name: Confit Child Theme     Author: Automattic     Template: confit     Description: Confit Child Theme 1     Version: 1.0     */          @import url(\'../confit/style.css\');        * Should also mention that the parent functions are not loading either.  Thanks!',
        "I've got a binary characteristic and a population $S$ with size $n$ and $P[X] = p$ such that $p$ may be small and $n$ is extremely large. Within this population are subpopulations of various sizes $S_0, S_1, \\dots, S_k \\subset S$. I'd like to be able to select each subpopulation in which $p_i < p$ with some concept of statistical significance. My first inclination is to observe that the standard error on each $p_i$ is $SE_i = \\sqrt{\\frac{\\hat{p_i}(1-\\hat{p_i})}{n}}$ and to compare upper bounds on confidence intervals. $\\{S_i \\; | \\; \\hat{p_i} + 3 \\cdot SE_i < p\\}$, for example. But when $\\hat{p_i} = 0$, then $SE_i = 0$, and this upper bound is 0 even for the smallest subpopulations (like those where $n_i = 1$). Is there any way to express uncertainty in $p_i$ when $\\hat{p_i} = 0$? Maybe through use of $p$ as a prior? **Edit:** It looks like the Jeffreys interval as described in Brown et al. is about what I'm after, though I'm not as-of-yet sure how to apply it.",
        'What is the difference between a battery and a charged capacitor? I can see lot of similarities between capacitor and battery. In both these charges are separated and When not connected in a circuit both can have same Potential difference `V`. The only difference is that battery runs for longer time but a capacitor discharges almost instantaneously. Why this difference? What is the exact cause for the difference in the discharge times?',
        'To me the word "curious" would be something you can be i.e. > I am curious what tomorrow will bring I recently read a text of a student I was supervising which used it as follows > A curious phenomenon is ... With which he meant to say that the phenomenon was peculiar, odd or strange. The only other case I have ever seen this is in the movie title: "The Curious Case of Benjamin Button", but that might be \'artistic freedom\' (since Curious Case has the nice C.. C..). My question is: is the usage of the word "curious" in the meaning of peculiar correct?',
        "I'm looking to choose open source project hosting site for an F# project using SVN. CodePlex is where the .NET community in general and most F# projects are hosted, but I'm worried TFS + SvnBridge is going to give me headaches. So I'm looking elsewhere and seeking advice here. Or if you think CodePlex is still the best choice in my scenario, I'd like to hear that too. So far, Google Code is looking appealing to me. They have a clean interface and true SVN hosting. But there are close to no F# projects currently hosted (it's not even in their search by programming language list), so I'm wondering if there are any notable downsides besides the lack of community I might encounter. If there is yet another option, I'd like to hear that too. Thanks!",
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Correlation

Dataset: sts_dev
Evaluated with CrossEncoderCorrelationEvaluator

Metric	Value
pearson	0.8392
spearman	0.7298

Cross Encoder Reranking

Datasets: NanoMSMARCO_R100, NanoNFCorpus_R100 and NanoNQ_R100

Evaluated with CrossEncoderRerankingEvaluator with these parameters:

{
    "at_k": 10,
    "always_rerank_positives": true
}

Metric	NanoMSMARCO_R100	NanoNFCorpus_R100	NanoNQ_R100
map	0.5685 (+0.0790)	0.3511 (+0.0901)	0.5917 (+0.1721)
mrr@10	0.5570 (+0.0795)	0.5391 (+0.0392)	0.6017 (+0.1750)
ndcg@10	0.6146 (+0.0742)	0.3779 (+0.0529)	0.6450 (+0.1444)

Cross Encoder Nano BEIR

Dataset: NanoBEIR_R100_mean

Evaluated with CrossEncoderNanoBEIREvaluator with these parameters:

{
    "dataset_names": [
        "msmarco",
        "nfcorpus",
        "nq"
    ],
    "rerank_k": 100,
    "at_k": 10,
    "always_rerank_positives": true
}

Metric	Value
map	0.5038 (+0.1137)
mrr@10	0.5659 (+0.0979)
ndcg@10	0.5459 (+0.0905)

Training Details

Training Dataset

Unnamed Dataset

Size: 69,699 training samples
Columns: query, document, and label

Approximate statistics based on the first 1000 samples:

	query	document	label
type	string	string	int
details	min: 15 characters mean: 49.33 characters max: 125 characters	min: 45 characters mean: 793.68 characters max: 18801 characters	0: ~74.50% 1: ~25.50%

Samples:

query	document	label
`Do elevator upgrades increase your passive credit earnings, too?`	I searched for a solution for this problem, but cannot find an answer (or exact replica of the problem) Basically, I set up Multisite on MAMP Pro (Apache port 80, MySQL Port 3306). The set up was smooth, and I created a new site via a subdirectory. The parent theme loads fine. I created a child theme, and it activates (it doesn't show a broken message). On the Appearance page it shows the message "This theme requires the parent theme", but underneath the Theme Description. However when I view the front page of the site, the page is blank, and there is no html at all. Would could possibly be the error? I spent a few hours on this already and it's not going really well. Code of child theme, only CSS, no functions.php or other php files in the child theme folder. /* Theme Name: Confit Child Theme Author: Automattic Template: confit Description: Confit Child Theme 1 Version: 1.0 / @import url('../confit/style.css'); Should also menti...	`0`
`Traceback (most recent call last) error appears on terminal`	I've got a binary characteristic and a population $S$ with size $n$ and $P[X] = p$ such that $p$ may be small and $n$ is extremely large. Within this population are subpopulations of various sizes $S_0, S_1, \dots, S_k \subset S$. I'd like to be able to select each subpopulation in which $p_i < p$ with some concept of statistical significance. My first inclination is to observe that the standard error on each $p_i$ is $SE_i = \sqrt{\frac{\hat{p_i}(1-\hat{p_i})}{n}}$ and to compare upper bounds on confidence intervals. ${S_i ;	; \hat{p_i} + 3 \cdot SE_i < p}$, for example. But when $\hat{p_i} = 0$, then $SE_i = 0$, and this upper bound is 0 even for the smallest subpopulations (like those where $n_i = 1$). Is there any way to express uncertainty in $p_i$ when $\hat{p_i} = 0$? Maybe through use of $p$ as a prior? Edit: It looks like the Jeffreys interval as described in Brown et al. is about what I'm after, though I'm not as-of-yet sure how to apply it.
`Do I have to install a custom ROM if I root?`	`What is the difference between a battery and a charged capacitor? I can see lot of similarities between capacitor and battery. In both these charges are separated and When not connected in a circuit both can have same Potential difference V. The only difference is that battery runs for longer time but a capacitor discharges almost instantaneously. Why this difference? What is the exact cause for the difference in the discharge times?`	`0`

Loss: BinaryCrossEntropyLoss with these parameters:

{
    "activation_fn": "torch.nn.modules.linear.Identity",
    "pos_weight": null
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: epoch
per_device_train_batch_size: 32
per_device_eval_batch_size: 32
learning_rate: 2e-05
warmup_ratio: 0.1
save_only_model: True
fp16: True
load_best_model_at_end: True

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: epoch
prediction_loss_only: True
per_device_train_batch_size: 32
per_device_eval_batch_size: 32
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: True
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: 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: True
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}
tp_size: 0
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
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
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
eval_use_gather_object: False
average_tokens_across_devices: False
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	sts_dev_spearman	NanoMSMARCO_R100_ndcg@10	NanoNFCorpus_R100_ndcg@10	NanoNQ_R100_ndcg@10	NanoBEIR_R100_mean_ndcg@10
-1	-1	-	0.5982	0.6519 (+0.1115)	0.3749 (+0.0498)	0.6497 (+0.1490)	0.5588 (+0.1035)
0.4589	1000	0.4015	-	-	-	-	-
0.9179	2000	0.191	-	-	-	-	-
1.0	2179	-	0.7298	0.6146 (+0.0742)	0.3779 (+0.0529)	0.6450 (+0.1444)	0.5459 (+0.0905)
1.3768	3000	0.163	-	-	-	-	-
1.8357	4000	0.1524	-	-	-	-	-
2.0	4358	-	0.7312	0.5951 (+0.0547)	0.3808 (+0.0557)	0.6490 (+0.1484)	0.5416 (+0.0863)
2.2946	5000	0.1369	-	-	-	-	-
2.7536	6000	0.1297	-	-	-	-	-
3.0	6537	-	0.7335	0.5994 (+0.0590)	0.3743 (+0.0492)	0.6500 (+0.1494)	0.5412 (+0.0859)
-1	-1	-	0.7298	0.6146 (+0.0742)	0.3779 (+0.0529)	0.6450 (+0.1444)	0.5459 (+0.0905)

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.10.12
Sentence Transformers: 4.1.0
Transformers: 4.51.3
PyTorch: 2.1.0+cu118
Accelerate: 1.6.0
Datasets: 3.5.0
Tokenizers: 0.21.1

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",
}

maennyn
/

pv211_beir_cqadupstack_crossencoder