Instructions to use ArnavKewalram/bge-base-code-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ArnavKewalram/bge-base-code-v1 with sentence-transformers:

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("ArnavKewalram/bge-base-code-v1")

sentences = [
    "Sets the global variables $rects and $origRectSpecs",
    "def modify_ranking(tournament):\n    database = TinyDB('db.json')\n    # recuperation de tous les joueurs du tournoi\n    players_table = database.table('players')\n    list_players = get_player_list(tournament)\n    # Modification du rang joueur par joueur\n    for player in list_players:\n        new_ranking = view.modify_ranking_view(player)\n        players_table.update({'Classement': new_ranking}, doc_ids=[player.doc_id])",
    "function setConstants() {\n        const wrapItems = \".image-analysis-wrapper .face-wrap, .image-analysis-wrapper .score-wrap, .image-analysis-wrapper .attribute-wrap, .image-analysis-wrapper .region-block, .image-analysis-wrapper .region-block .word-block .word-wrap\";\n\n        $rects = jQuery(\".image-analysis-wrapper .rectangle\");\n\n        // Iterate over each rectangle and save the width, height, top position,\n        // left position, closest stats block element, and position of the closest\n        // stats block element to an object.  Each object is then added to the \n        // $origRectSpecs array for global use.\n        $origRectSpecs = $rects.map(function () {\n            closestWrapItems = jQuery(this).siblings(wrapItems);\n\n            const stats = closestWrapItems.map(function () {\n                return {\n                    origStatTop: jQuery(this).position().top || parseInt(jQuery(this).css(\"top\")),\n                    origStatLeft: jQuery(this).position().left || parseInt(jQuery(this).css(\"left\"))\n                }\n            })\n\n            return {\n                origRectWidth: jQuery(this).width(),\n                origRectHeight: jQuery(this).height(),\n                origRectTop: jQuery(this).position().top || parseInt(jQuery(this).css(\"top\")), // if the rect is on a tab that is currently not displayed it has a position of 0, so this check gets the css instead so we don't lose the value\n                origRectLeft: jQuery(this).position().left || parseInt(jQuery(this).css(\"left\")),\n                statBlock: closestWrapItems[0],\n                statPosition: stats[0]\n            }\n        })\n    }",
    "reset() {\n\n        // Set the initial crop to match any given fixed aspect ratio (or\n        // default to a square crop 1:1).\n        let aspectRatio = this._initialAspectRatio\n\n        // Calculate the initial crop size such that it fits within the bounds\n        let width = getWidth(this.bounds)\n        let height = getWidth(this.bounds) / aspectRatio\n\n        if (aspectRatio < width / getHeight(this.bounds)) {\n            width = getHeight(this.bounds) * aspectRatio\n            height = getHeight(this.bounds)\n        }\n\n        // Calculate the initial crop position to be central to the bounds\n        const x = (getWidth(this.bounds) - width) / 2\n        const y = (getHeight(this.bounds) - height) / 2\n\n        // Set the region\n        this.region = [\n            [x, y],\n            [x + width, y + height]\n        ]\n    }"
]
embeddings = model.encode(sentences)

similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [4, 4]

Notebooks
Google Colab
Kaggle

bge-base-code-v1 / README.md

ArnavKewalram

Upload bge-base-code-v1 (110M) trained on CoRNStack

43f8df7 verified 6 days ago

preview code

Raw

History Blame Contribute Delete

34.6 kB

metadata

tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:200000
  - loss:MatryoshkaLoss
  - loss:MultipleNegativesRankingLoss
base_model: BAAI/bge-base-en-v1.5
widget:
  - source_sentence: Sets the global variables $rects and $origRectSpecs
    sentences:
      - |-
        def modify_ranking(tournament):
            database = TinyDB('db.json')
            # recuperation de tous les joueurs du tournoi
            players_table = database.table('players')
            list_players = get_player_list(tournament)
            # Modification du rang joueur par joueur
            for player in list_players:
                new_ranking = view.modify_ranking_view(player)
                players_table.update({'Classement': new_ranking}, doc_ids=[player.doc_id])
      - |-
        function setConstants() {
                const wrapItems = ".image-analysis-wrapper .face-wrap, .image-analysis-wrapper .score-wrap, .image-analysis-wrapper .attribute-wrap, .image-analysis-wrapper .region-block, .image-analysis-wrapper .region-block .word-block .word-wrap";

                $rects = jQuery(".image-analysis-wrapper .rectangle");

                // Iterate over each rectangle and save the width, height, top position,
                // left position, closest stats block element, and position of the closest
                // stats block element to an object.  Each object is then added to the 
                // $origRectSpecs array for global use.
                $origRectSpecs = $rects.map(function () {
                    closestWrapItems = jQuery(this).siblings(wrapItems);

                    const stats = closestWrapItems.map(function () {
                        return {
                            origStatTop: jQuery(this).position().top || parseInt(jQuery(this).css("top")),
                            origStatLeft: jQuery(this).position().left || parseInt(jQuery(this).css("left"))
                        }
                    })

                    return {
                        origRectWidth: jQuery(this).width(),
                        origRectHeight: jQuery(this).height(),
                        origRectTop: jQuery(this).position().top || parseInt(jQuery(this).css("top")), // if the rect is on a tab that is currently not displayed it has a position of 0, so this check gets the css instead so we don't lose the value
                        origRectLeft: jQuery(this).position().left || parseInt(jQuery(this).css("left")),
                        statBlock: closestWrapItems[0],
                        statPosition: stats[0]
                    }
                })
            }
      - |-
        reset() {

                // Set the initial crop to match any given fixed aspect ratio (or
                // default to a square crop 1:1).
                let aspectRatio = this._initialAspectRatio

                // Calculate the initial crop size such that it fits within the bounds
                let width = getWidth(this.bounds)
                let height = getWidth(this.bounds) / aspectRatio

                if (aspectRatio < width / getHeight(this.bounds)) {
                    width = getHeight(this.bounds) * aspectRatio
                    height = getHeight(this.bounds)
                }

                // Calculate the initial crop position to be central to the bounds
                const x = (getWidth(this.bounds) - width) / 2
                const y = (getHeight(this.bounds) - height) / 2

                // Set the region
                this.region = [
                    [x, y],
                    [x + width, y + height]
                ]
            }
  - source_sentence: Returns true if the motor is at its lower limit.
    sentences:
      - |-
        def is_lower_limit(self):
            is_lower = self.get_raw_status() & self.STATUS_LLIM
            return bool(is_lower)
      - "def testLowerBound(self,time,accel):\n\t\tif (time - self.timestamp) > ParserSettings.TIME_DELTA:#tests lockout threshold of a flick event\n\t\t\tif accel > self.lower:#tests to see if the flick maximum is met yet, relative to the previous magnitude\n\t\t\t\tself.timestamp \t= time#set appropriate values when flick triggered\n\t\t\t\ttoReturn \t\t= self.lower\n\t\t\t\tself.lower\t \t= 0#reset flick for next magnitude test\n\t\t\t\treturn toReturn\n\t\t\telse:\n\t\t\t\tself.lower = accel#if no flick yet, update most recent flick to test\n\t\t\t\treturn 0\n\t\telse:\n\t\t\treturn 0"
      - "function checkVisibility(message)\n{\n\t// Scroll the view down a certain amount\n\t$chatlogs.stop().animate({scrollTop: $chatlogs[0].scrollHeight});\n}"
  - source_sentence: >-
      Extracts an object of params the given route cares about from the given
      params object.
    sentences:
      - "function getRouteParams(route, params) {\n\t  var routeParams = {};\n\t\n\t  if (!route.path) return routeParams;\n\t\n\t  (0, _PatternUtils.getParamNames)(route.path).forEach(function (p) {\n\t    if (Object.prototype.hasOwnProperty.call(params, p)) {\n\t      routeParams[p] = params[p];\n\t    }\n\t  });\n\t\n\t  return routeParams;\n\t}"
      - "public static void initImageLoader(Context context) {\n\t\tFile cacheDir = StorageUtils.getCacheDirectory(context);\n\t\tImageLoaderConfiguration config = new ImageLoaderConfiguration.Builder(context)\n\t\t\t\t.threadPriority(Thread.NORM_PRIORITY - 2)\n\t\t\t\t.denyCacheImageMultipleSizesInMemory() \n//\t\t\t\t.discCache(new UnlimitedDiscCache(cacheDir)) // default\n\t\t\t\t.discCacheFileNameGenerator(new Md5FileNameGenerator())\n\t\t\t\t.memoryCache(new LruMemoryCache(2 * 1024 * 1024))\n\t\t\t\t.tasksProcessingOrder(QueueProcessingType.LIFO)\n\t\t\t\t.writeDebugLogs() // Remove for release app\t\t\t\t\n\t\t\t\t.build();\n\t\t// Initialize ImageLoader with configuration.\n\t\tImageLoader.getInstance().init(config);\n\t}"
      - "function getRouteParams(route, params) {\n\t  var routeParams = {};\n\n\t  if (!route.path) return routeParams;\n\n\t  (0, _PatternUtils.getParamNames)(route.path).forEach(function (p) {\n\t    if (Object.prototype.hasOwnProperty.call(params, p)) {\n\t      routeParams[p] = params[p];\n\t    }\n\t  });\n\n\t  return routeParams;\n\t}"
  - source_sentence: select all as they all have a class of "section" return array of IDs
    sentences:
      - "func (o *BookBuyOK) WriteResponse(rw http.ResponseWriter, producer runtime.Producer) {\n\n\trw.WriteHeader(200)\n\tif o.Payload != nil {\n\t\tpayload := o.Payload\n\t\tif err := producer.Produce(rw, payload); err != nil {\n\t\t\tpanic(err) // let the recovery middleware deal with this\n\t\t}\n\t}\n}"
      - |-
        function getMenuCLassesIDs()
            {
                let IdsArray = [];
                const elements = document.querySelectorAll('.banner,.section');
                for(const elm of elements)
                    {
                        IdsArray.push(elm.id);
                    }
                
                return IdsArray;
            }
      - |-
        getAllSectionsForRegion(region) {
                if (!region) {
                    return null;
                }
                return region.getElementsByClassName(A11yClassNames.SECTION);
            }
  - source_sentence: For component xyz, returns "xyz[]"
    sentences:
      - "public String toString() {\n\t    \tif(size == 0) {\n\t    \t\treturn \"[]\";\n\t    \t}else {\n\t    \t\t\n\t    \t\tString result = \"[\" + elementData[0];\n\t    \t\tfor(int i = 1; i < size; i++) {\n\t    \t\t\tresult += \", \" + elementData[i];\n\t    \t\t}\n\t    \t\t\n\t    \t\tresult += \"]\";\n\t    \t\t\n\t    \t\treturn result;\n\t    \t}\n\t    }"
      - "def masterPath(self):\n\t\treturn fl.File( self._path + '/master.data' )"
      - |-
        private static final String getArrayTypeName(String typeName) {
            final String arrayTypeName = builtInArrayComponentName2ArrayTypeName.get(typeName);
            return (null == arrayTypeName) ? typeName + ARRAY_TYPE_SUFFIX : arrayTypeName;
          }
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - cosine_accuracy@1
  - cosine_accuracy@5
  - cosine_accuracy@10
  - cosine_precision@1
  - cosine_precision@3
  - cosine_precision@5
  - cosine_precision@10
  - cosine_recall@1
  - cosine_recall@3
  - cosine_recall@5
  - cosine_recall@10
  - cosine_ndcg@10
  - cosine_mrr@1
  - cosine_mrr@5
  - cosine_mrr@10
  - cosine_map@100
model-index:
  - name: SentenceTransformer based on BAAI/bge-base-en-v1.5
    results:
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: cornstack eval
          type: cornstack_eval
        metrics:
          - type: cosine_accuracy@1
            value: 0.718
            name: Cosine Accuracy@1
          - type: cosine_accuracy@5
            value: 0.884
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.924
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.718
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.2866666666666666
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.1768
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09240000000000001
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.718
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.86
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.884
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.924
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.8259696432548304
            name: Cosine Ndcg@10
          - type: cosine_mrr@1
            value: 0.718
            name: Cosine Mrr@1
          - type: cosine_mrr@5
            value: 0.7882666666666666
            name: Cosine Mrr@5
          - type: cosine_mrr@10
            value: 0.7940396825396825
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7967267778119025
            name: Cosine Map@100

SentenceTransformer based on BAAI/bge-base-en-v1.5

This is a sentence-transformers model finetuned from BAAI/bge-base-en-v1.5. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: BAAI/bge-base-en-v1.5
Maximum Sequence Length: 384 tokens
Output Dimensionality: 768 dimensions
Similarity Function: Cosine Similarity
Supported Modality: Text

Model Sources

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

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'BertModel'})
  (1): Pooling({'embedding_dimension': 768, 'pooling_mode': 'cls', '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("sentence_transformers_model_id")
# Run inference
queries = [
    'For component xyz, returns "xyz[]"',
]
documents = [
    'private static final String getArrayTypeName(String typeName) {\n    final String arrayTypeName = builtInArrayComponentName2ArrayTypeName.get(typeName);\n    return (null == arrayTypeName) ? typeName + ARRAY_TYPE_SUFFIX : arrayTypeName;\n  }',
    'public String toString() {\n\t    \tif(size == 0) {\n\t    \t\treturn "[]";\n\t    \t}else {\n\t    \t\t\n\t    \t\tString result = "[" + elementData[0];\n\t    \t\tfor(int i = 1; i < size; i++) {\n\t    \t\t\tresult += ", " + elementData[i];\n\t    \t\t}\n\t    \t\t\n\t    \t\tresult += "]";\n\t    \t\t\n\t    \t\treturn result;\n\t    \t}\n\t    }',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 768] [2, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[0.4454, 0.3780]])

Evaluation

Metrics

Information Retrieval

Dataset: cornstack_eval
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.718
cosine_accuracy@5	0.884
cosine_accuracy@10	0.924
cosine_precision@1	0.718
cosine_precision@3	0.2867
cosine_precision@5	0.1768
cosine_precision@10	0.0924
cosine_recall@1	0.718
cosine_recall@3	0.86
cosine_recall@5	0.884
cosine_recall@10	0.924
cosine_ndcg@10	0.826
cosine_mrr@1	0.718
cosine_mrr@5	0.7883
cosine_mrr@10	0.794
cosine_map@100	0.7967

Training Details

Training Dataset

Unnamed Dataset

Size: 200,000 training samples
Columns: anchor, positive, and negative

Approximate statistics based on the first 100 samples:

	anchor	positive	negative
type	string	string	string
modality	text	text	text
details	min: 6 tokens mean: 19.83 tokens max: 105 tokens	min: 21 tokens mean: 139.33 tokens max: 384 tokens	min: 11 tokens mean: 100.52 tokens max: 384 tokens

Samples:

anchor	positive	negative
`Fades all outputs to the given color and waits for it to complete.`	`def FadeOutputs(box, color, steps=50): for output in box: output.Fade(color=color, steps=steps) time.sleep(steps / (float(box.frequency) / len(box)))`	def _colour_loop(self, colours, seconds=None, milliseconds=None, fade=True): colours = self.convert_to_colour_list(colours) #Forces a list of colours into an actual python list if len(colours)<2: colours.append("#000000") #Blink between black and the specified colour if only one provided #Start with the first colour immediately: if fade: self.fade(colours[0]) else: self.set(colours[0]) step_time = self.clean_time_in_milliseconds(seconds, milliseconds, default_seconds=1, minimum_milliseconds=50) #Do the loop i = 1 #We're moving to the second colour now total_colours = len(colours) while not self._sequence_stop_signal: #Resolve our colour next_colour = colours[i] i = (i+1) % total_colours #ensures we are never asking for more colours than provided if fade: #Fading is a blocking process, thus we let the fade l...
`Sets the additional element count if buffer resize is required, defaults to initialElementCount of factory method.`	`public void setResizeElementCount(int v) { vboSet.setResizeElementCount(v); }`	`public int getResizeElementCount() { return vboSet.getResizeElementCount(); }`
`delete a specific incident`	`def delete_specific_incident(self, incident_id): self.cursor.execute("""DELETE FROM incidents WHERE incident_id ='%s' AND status='draft' """ %(incident_id)) self.commiting() return incident_id`	`def delete(openstack_resource): openstack_resource.delete()`

Loss: MatryoshkaLoss with these parameters:

{
    "loss": "MultipleNegativesRankingLoss",
    "matryoshka_dims": [
        768,
        512,
        256,
        128
    ],
    "matryoshka_weights": [
        1.0,
        1.0,
        1.0,
        1.0
    ],
    "n_dims_per_step": -1
}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 16
gradient_accumulation_steps: 4
learning_rate: 2e-05
num_train_epochs: 1
warmup_ratio: 0.05
fp16: True
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
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: 4
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: 1
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: None
warmup_ratio: 0.05
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: 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: 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: None
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	cornstack_eval_cosine_ndcg@10
0.016	50	4.7419	-
0.032	100	4.2347	-
0.048	150	3.9702	-
0.064	200	3.7572	-
0.08	250	3.7142	-
0.096	300	3.5756	-
0.112	350	3.6356	-
0.128	400	3.7447	-
0.144	450	3.4957	-
0.16	500	3.4753	-
0.176	550	3.3913	-
0.192	600	3.1695	-
0.208	650	3.2787	-
0.224	700	3.2488	-
0.24	750	3.2078	-
0.256	800	3.2942	-
0.272	850	3.0672	-
0.288	900	3.1279	-
0.304	950	3.1464	-
0.32	1000	3.1526	0.8336
0.336	1050	3.1064	-
0.352	1100	3.2408	-
0.368	1150	3.2562	-
0.384	1200	3.1835	-
0.4	1250	2.9471	-
0.416	1300	3.1631	-
0.432	1350	3.1428	-
0.448	1400	2.9445	-
0.464	1450	3.0482	-
0.48	1500	3.09	-
0.496	1550	3.0184	-
0.512	1600	3.031	-
0.528	1650	2.9703	-
0.544	1700	2.9743	-
0.56	1750	3.0344	-
0.576	1800	3.0521	-
0.592	1850	2.9177	-
0.608	1900	3.0357	-
0.624	1950	2.9277	-
0.64	2000	2.8525	0.8287
0.656	2050	2.978	-
0.672	2100	2.8437	-
0.688	2150	2.849	-
0.704	2200	2.9248	-
0.72	2250	2.8551	-
0.736	2300	2.8741	-
0.752	2350	2.7847	-
0.768	2400	2.8682	-
0.784	2450	2.7556	-
0.8	2500	2.8122	-
0.816	2550	2.9173	-
0.832	2600	2.8772	-
0.848	2650	2.8275	-
0.864	2700	2.7819	-
0.88	2750	2.8215	-
0.896	2800	2.9207	-
0.912	2850	2.8892	-
0.928	2900	2.7682	-
0.944	2950	2.868	-
0.96	3000	2.7926	0.8260
0.976	3050	2.8871	-
0.992	3100	2.8142	-

Training Time

Training: 2.5 days
Evaluation: 1.9 minutes
Total: 2.5 days

Framework Versions

Python: 3.13.7
Sentence Transformers: 5.6.0
Transformers: 4.57.6
PyTorch: 2.12.1+cu126
Accelerate: 1.13.0
Datasets: 4.8.5
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",
}

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{oord2019representationlearningcontrastivepredictive,
      title={Representation Learning with Contrastive Predictive Coding},
      author={Aaron van den Oord and Yazhe Li and Oriol Vinyals},
      year={2019},
      eprint={1807.03748},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1807.03748},
}