osllmai/bge-base-en-v1.5-finetuned_osllmai_v1

SentenceTransformer based on BAAI/bge-base-en-v1.5 (Finetuned by OSLLMAI)

This is a SentenceTransformers model finetuned by OSLLMAI from BAAI/bge-base-en-v1.5. It maps sentences and paragraphs into a 768-dimensional dense vector space. The model is ideal for tasks such as semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more. It uses cosine similarity as the similarity function and supports a maximum sequence length of 512 tokens.

Connect with us on GitHub, LinkedIn, or email us at support@osllm.ai!

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: BAAI/bge-base-en-v1.5
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 tokens
• 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': 512, 'do_lower_case': True}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, '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("osllmai/bge-base-en-v1.5-finetuned_osllmai_v1")
# Run inference
sentences = [
    'Early this year, there was a buzz on Motorola inviting T-Mobile Moto X owners to take part in a soak test for possible future update. Motorola seemed skeptical in disclosing facts at that point of time but since Moto G was recently upgraded to Android 4.4.2; enthusiasts anticipated the same for T-Mobile Moto X. And it turned out to be true.\nNews Update\nThis T-Mobile version of Moto X is now receiving the upgrade which is a file size of 147.6 MB. The Android 4.4.2 is the latest version of KitKat that includes all the goodies from the earlier installments, plus a few additions. The good news is, Motorola has customized the whole package and made a few tweaks into the update. The Software Version bumped to 161.44.25 and the notable changes are listed as below:\n- It added substantial support for services like printing photos, Google Docs, Gmail messages and other such content via Wi-Fi, Bluetooth and hosted services such as HP ePrinters and Google Cloud Print.\n- It fixed all the bugs identified during the preliminary runs, including the ones that caused a few users to experience short battery life after upgrading to KitKat.\n- Another bug that caused delays in synchronizing email services like Microsoft Exchange was resolved, thus adding to the convenience of the user.\nThis is a noteworthy upgrade, considering the fact that bugs and errors were fixed. Mobile addicts across the world will rejoice, for they can experience the smartness of Android KitKat flawlessly in their devices. This is significant development in terms of update.\nThis variant is an unlocked GSM device so chances are, you can use it on networks of other service providers. In all probability, the update should not be affected and the installation should hardly take much time. The T-Mobile Moto X Android update is now available for manual download. It is accessible in the following sequential way:\n- Click on Settings\n- Click on About Phone\n- Click on System Updates\n- Click on Download\nRecommendations\nFor ensuring a successful installation, it is highly recommended to install this update with at least 50% battery and a strong connectivity; preferably Wi-Fi. Follow the notification message and select download-> once the download is over, select Install-> Once the installation is over, and the phone will automatically restart. This marks the completion of the installation process. The phone is now updated to 161.44.25 – This build is same as the soak test.\nThis upgrade is free in the carrier network and Motorola and Google has collaborated for a back up service for those in trouble. In case of distress, a user can contact them through the Moto X web interface and avail the service. There is still no news on other carrier variants of this update but we can safely hope that it will roll out very soon. Though the upgrade doesn’t appeal in terms of version number but it is definitely significant for users to live with the latest KitKat.',
    'What are some of the notable changes in the T-Mobile Moto X update?',
    'Who is the editor of "The Routledge Handbook of Tourism Geographies"?',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

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

Training Details

Training Dataset

Unnamed Dataset

• Size: 9,598 training samples
• Columns: positive and anchor
• Approximate statistics based on the first 1000 samples:

  	positive	anchor
  type	string	string
  details	min: 172 tokens mean: 467.87 tokens max: 512 tokens	min: 7 tokens mean: 18.68 tokens max: 43 tokens

• Samples:

  positive	anchor
  Caption: Tasmanian berry grower Nic Hansen showing Macau chef Antimo Merone around his property as part of export engagement activities. THE RISE and rise of the Australian strawberry, raspberry and blackberry industries has seen the sectors redouble their international trade focus, with the release of a dedicated export plan to grow their global presence over the next 10 years. Driven by significant grower input, the Berry Export Summary 2028 maps the sectors’ current position, where they want to be, high-opportunity markets and next steps. Hort Innovation trade manager Jenny Van de Meeberg said the value and volume of raspberry and blackberry exports rose by 100 per cent between 2016 and 2017. She said the Australian strawberry industry experienced similar success with an almost 30 per cent rise in export volume and a 26 per cent rise in value to $32.6M over the same period. “Australian berry sectors are in a firm position at the moment,” she said. “Production, adoption of protected substrate cropping, improved genetics and an expanding geographic footprint have all helped put Aussie berries on a positive trajectory. “We are seeing a real transition point. Broad industry interest and a strong commercial appetite for export market development combined with the potential to capitalise on existing trade agreements and build new trade partnerships has created this perfect environment for growth.” High-income countries across Europe, North America and Northern Asia have been identified as having a palate for Australian grown berries with more than 4244 tonnes of fresh berries exported in the last financial year alone. The strategy identified the best short-term prospect markets for the Australian blackberry and raspberry industry as Hong Kong, Singapore, The United Arab Emirates and Canada. The strongest short-term trade options identified for the strawberry sector were Thailand, Malaysia, New Zealand and Macau. The strategy focuses heavily on growing the existing strawberry export market from 4 per cent to at least 8 per cent of national production by volume, in markets with a capacity and willingness to pay a premium for quality fruit. For raspberries and blackberries, the sectors aim to achieve a 5 per cent boost in exports assessed by volume across identified markets by 2021. Tasmanian raspberry exporter Nic Hansen said Australia offers some of the sweetest and most attractive berries in the world, and this combined with our stringent food safety standards across all stages of the supply chain puts growers in a solid position. “We have a great product, we are hungry to expand trade and now with this new plan in place, we have a clear roadmap towards driving growth,” Mr Hansen said. He said it is exciting to see new export market prospects for raspberries: “The more options we have for export the better. Now we just have to get on with the job of ensuring industry has all the tools it needs, such as supporting data and relationship building opportunities, to thrive in new markets.” This project was commissioned by Hort Innovation, and developed by market analysts and research consultants Auspex Strategic Advisory and AgInfinity. Hort Innovation will work now with berry sectors to determine levy-funded activities to support trade. See a summary of the strategy on the Hort Innovation website. For more information on the berry industries, refer to the Horticulture Statistics Handbook and the Strategic Investment Plans for strawberries, raspberries and blackberries. Growers seeking more information should email trade@horticulture.com.au	What is the Berry Export Summary 2028 and what is its purpose?
  RWSN Collaborations Southern Africa Self-supply Study Review of Self-supply and its support services in African countries A lady in Zimbabwe proudly shows off her onions - watered from her self-supply well © 2015 André Olschewski • Skat Project starts: 2015 Project finished: 2016 Collaborators & Partners:. Project Description UNICEF and Skat have collaborated on a). Perspectives Reach and benefits: - Self-supply is practised by millions of rural households in Sub-Sahara Africa as well as in Europe, USA and other areas of the world. - Benefits reported from having access to Self-supply water sources include convenience, less time spent for fetching water and access to more and better quality water. In some areas, Self-supply sources offer important added values such as water for productive use, income generation, family safety and improved food security. - Sustainability of services from Self-supply is high as there is strong ownership by people investing in own sources. - As Self-supply sources are shared sources, many people, including poor and vulnerable households, benefit from investments in Self-supply, often at no costs. This means that Self-supply can be effective in reaching the hard-to-reach. - For millions of people in rural areas of Africa, supported Self-supply will be the most cost effective service delivery model to provide access to safe water. This also includes those parts of the population which actually have poor access as they e.g. cannot afford water from communal supplies. - However, in areas where external support for Self-supply is lacking, only marginal improvements can usually be achieved, and the quality of services is lower than in areas where a dedicated support effort was made. Costs and business model for supported Self-supply - In many rural contexts, supported Self-supply is the most cost effective approach for water service delivery. However, as it is not applicable in all contexts, a blended approach combining communal water supply and supported Self-supply models should be followed. - Based on a Life Cycle Cost (LCC) analysis of different service delivery approaches, the LCC for communal supplies are about 40 US$/capita served in the study countries, whereas the LCC for supported Self-supply is about 10 U$/capita. - In sparsely populated areas, communal supplies (e.g. handpumps) are even more costly (up to 100 U$/capita served) as only few people can be served with one additional unit. Serving all rural people with communal supply is therefore not financially viable. - Considering the applicability of Self-supply technologies, in Zambia and Zimbabwe, the cost saving of following a blended approach using both communal supplies and supported Self-supply is almost 50% of the total LCC for reaching 100% of the population by 2030. These cost savings are equivalent to more than 330 million US$ in Zambia and more than 260 million US$ in Zimbabwe. Support services needed - Supported Self-supply is a service delivery model putting support services in place to improve Self-supply, so it is not about a particular technology. - Supported Self-supply is aligned with the Human Rights to Water and Sanitation, which allows a progressive realisation of the universal access to safe water. However, supported Self-supply is not a way to exempt government from its duties: Government has specific roles to play to ensure that everybody will have access to safe water finally. - To sustain and to take Self-supply to scale there is need for contextualised support as well as long-term engagement, capacity development at all levels, M&E and technical support, reliable funding and learning and sharing. - Interministerial cooperation and champions within government agencies are needed to ensure sustainable embedding and for taking Self-supply further, particularly in remote rural areas. - There is no-one-size-fits-all solution for supported Self-supply – for each programme, it needs a contextualized design and follow-up to achieve desired impact. - Hygiene promotion, including Household Water Treatment and Safe Storage (HWTS), is highly recommended for any non-piped water supply services, including Self-supply water sources. - The huge potential for substantially improving the level of water supply for millions of people in rural areas should be accessed through supported Self-supply. Some countries have endorsed supported Self-supply as service delivery model, such as Zimbabwe or Sierra Leone, and in Ethiopia, Self-supply is now being scaled up at national level. More Information » Review of Self-supply and its support services in African countries: Synthesis Report).	What are some of the benefits reported from having access to Self-supply water sources?
  All Android applications categories Description Coolands for Twitter is a revolutionary twitter client. It has many unique features, gives you the best mobile twitter experience you never imagined before. The first unique feature is Real-Time. You can’t find any refresh button in this app, because you absolutely don’t need to. Every time you open it, you’ll get the latest tweets and while you’re reading, you’ll get incoming tweets in Real-Time. So if your friend mentioned you, you can reply instantly. The second unique feature is Avatar Indicator. Avatar Indicator is small avatars showed on the title bar to indicate that you’ve got new message/tweet/mention. Since it’s real-time, you’ll keep getting incoming tweets while you’re reading your older timeline, Avatar-Indicator will let you know who’s tweet you’ve just got, and decide whether to check it out right away. The third unique feature is Direct Link I think it is obviously the most intuitive and convenient way to open a link. When you want to open a link, just click it in the time line . You can also click a username to open a profile window, click a hash tag to open a search result window. Different kind of links displayed in different colors, you can change it to whatever color you like. The fourth unique feature is Smart Bookmark Have you ever experienced this scenario? When you are reading your home timeline, the app notified you that you’ve got some new tweets, you click “go-to-top” button to read the newest tweets, and then you want to get back to the previous position to continue your reading. How can you do this? In other twitter clients you have to scroll down all the way to find where you were, a lot of time wasted. But in this app, “go-to-top” button will appear when you scrolling up, click it, you can got to top, read the newest tweets. After that when you scrolling down, Smart Bookmark button will appear at the corner. Click it, you can get back to exactly where you were. The fifth unique feature is User Level Notification Notification for all your new tweets is meaningless, if you following more than a few users, you’ll get new tweets all the time. What if you only want to be notified when someone you most care about posted a new tweet? In this app, it’s easy. You can change your friend’s notification setting directly in his/her profile screen. You can also set different notification ringtone for your friends respectively. So when you heard a notification ringtone, you’ll know who he/she is without the need to open your phone. And you can manage all the enabled User-Level-Notification settings in one place. Also has most of the basic twitter client features, like post/delete tweets, retweet, retweet with comment, reply, quote, send/delete direct message, subscribe/unsubscribe lists, follow/unfollow user, multiple accounts support, append picture with your tweets, mention auto complete, recent search auto complete, conversation view. Please give me feedbacks if you have tried it, and I PROMISE to reply all your emails. 1.30-1.36 update: *Ad-Free. *3 times faster when launching and loading older tweets. *Support unlimited accounts rather than 3 accounts. *Support notifications for all accounts rather than only for the current account. *Improved mention suggestion feature. *Use URL link to do RT with comment, so you can comment more characters. *Conversation view, click the orange(you can change the color) username in replied tweet to show conversation view. *Support longer tweet, longer tweet will be converted to a picture automatically. You really should try it out your self. *Refined tweet composing view, to support longer tweet. *Support handle text shared by other app. from 103 reviews Download Coolands for Twitter Free - V1.38 - 298K Sorry ... This app is no longer available. Share this app Screenshots	What are the unique features of the Coolands for Twitter app?

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Evaluation Dataset

Unnamed Dataset

• Size: 500 evaluation samples
• Columns: positive and anchor
• Approximate statistics based on the first 500 samples:

  	positive	anchor
  type	string	string
  details	min: 188 tokens mean: 460.02 tokens max: 512 tokens	min: 10 tokens mean: 18.91 tokens max: 39 tokens

• Samples:

  positive	anchor
  Perhaps Not such a Good Idea I have found trying to run a blog is very time-consuming, and there are other calls on my time. I think it has been demonstrated that if enough people are unable to self-moderate, the nuggets of interest are swamped by the rubbish. Sadly I agree with Mark Frank's assessment. I had hoped more thread topics would be proposed nothing has been suggested by anyone for a while. My personal view is that, considering DaveScot's generally perceived blog persona, I have to admit that he hasn't been (on this site) quite the unmitigated disaster predicted. John Davison, on the other hand has conformed perfectly to predictions, which is a shame, but his choice. I am happy to let things run for a while, but would like to hear from anyone who has a suggestion for a thread topic. Post here or in the suggestions thread 23 comments: How about an "ID: show me the research" thread? OK Rich, put some meat on the bones and I'll paste it. Of course I have. I have no respect for you or your cronies from Psnda's Thumb. What did you expect kudos? What do you want another thread for? No one has even attempted to answer my four challenges yet. You know why? I do. It is because they can't, because everything you and your Darwimpian cronies stand for is a myth, an illusion and a hoax. That's why. You might as well close down this flame pit while you are behind as it isn't going anywhere. It is hard to believe isn't it? I love it so! " "How about an "ID: show me the research" thread? " Alan Fox said... OK Rich, put some meat on the bones and I'll paste it. " Why is there no ID research, even on Dembski's blog? Why does ID consist solely of an opinion that some parts of human biology are designed? How would IDers actually prove that some parts of human biology are in fact designed? Expand a bit on the theme, Wonderpants and I'll start a thread if you like. "My personal view is that, considering DaveScot's generally perceived blog persona, I have to admit that he hasn't been (on this site) quite the unmitigated disaster predicted. " At a guess, it's because he can't duck or delete awkward subjects. I note from skimming through the threads though that he's been rather selective as to which ones he posts in, namely the ones that don't pose awkward ID questions. ;-) Well, we can't torture a confession out of him. As Lenny points out frequently, an absence of an answer is in itself an answer. What about a thread from you, entitled " My awkard questions for DaveScot"? I'm not sure how much meat can be put on the bones of a non-existent project, but here's my thought: I would really like to hear about actual research projects that can be / are being done. Without knowing of any that are running currently, I'm not sure if it would be a good thread to start, but maybe you could ask for ideas. It could be a thread dedicated to lab experiments. If X is designed, we will find Y. Here's how we find Y in the lab. Then we watch for the landslide of X and Y that get suggested and, of course, the methods that actually find these things. Why would they start posting it now, though, after years of keeping it secret? JAD: I have no respect for you or your cronies from Psnda's Thumb. Which raises the obvious question of why you hang around in forums like this. Why not submit your work to a technical journal where real scientists will read it? Well, unless and until Wonderpants or Blipey want to expand on it, I have framed a thread along the line suggested. I think a good thread would be "Place A Vote For or Against the Banning of Professor Emeritus John Davison". I invade the ephemeral meaningless world of cyberdom for amusement on the outside chance that I might find a rational mind once in a while, one like johndarius for example. Mostly I encounter mentally impaired ideologues with IQs in the room temperature range or hostile, rabid, certifiably deranged schizophrenic sociopaths like Spravid Dinger. This particular blog seems to be blessed with both varieties. Naturally - I love it so! I'll try and think of something tomorrow. Been watching the footie tonight. Props to France for winning aghainst Spain, Alan. Ah, the World Cup. Something else Mrs Fox and I disagree on. Yes there would have been a few glum faces at work tomorrow. Now if only France can beat Brazil, and England beat Portugal. JAD, you didn't answer my second question: Why not submit your work to a technical journal where real scientists will read it? Why would real scientists want to read the nonsensical ramblings of a pseudoscientist? Give me a shout if you need some "help." Naturally - I love it so! But JAD won't publish in a scientific journal any longer. According to his second post, we can assume that he visits Nature's website "for amusement on the outside chance that [he] might find a rational mind once in a while, one like a [creationist] for example. Mostly [he] encounters mentally impaired [evolutionary biologists] " I can't think of a single great scientist who wouldn't describe himself as a creationist, not one. Can anyone? I love it so! Democritus, Sagan, Darwin, Edison, Feynman, Curie, just to name a handful. Of course, they never managed to publish in Rivista... [snicker] Feynman once described scientific discovery as a religious experience. I agree entirely as I have had the same experience. That anyone could describe Darwin as a scientist is unthinkable. I didn't know that about Curie and tend not to accept it without some documentation. I love it so! What a sad little weasel you are. Feynman was an avowed atheist. Curie was raised Catholic but became an atheist on the death of her mother. Darwin was 100 times the scientist you are. Creationism is all but dead among true scientists; critical inquiry is poison to that superstitious twaddle.	What is the author's personal view on DaveScot's blog persona?
  Age reduction Academic atmosphere Beef tendon bottom Straight buckle low-heel cowhide Lefu shoes Mary Jane shoes Spring and summer Women's shoes 0.73 ins Chaopai shoes Women's Shoes Academic atmosphere Versatile Graffiti Frenulum gym shoes Harajuku leisure time Hip hop jointly skate shoes Air force one Men's shoes Low Gang summer skate shoes student Korean version Versatile leisure time gym shoes female Reflection Little white shoes autumn Clover ozweego Daddy shoes Jackson Yi Same men and women Reflection motion Running shoes EE6999 Retro Britain Square head Frenulum Color matching motion Casual shoes 2021 new pattern Versatile Flat bottom Elastic band Little white shoes female Thick bottom British style Small leather shoes Women's shoes 2021 new pattern Big square head Spring and Autumn Lefu Autumn shoes black Single shoes U.S.A quality goods Jeffrey Campbell temperament crude high-heeled dollskill Buckles Low top shoes female widow quality goods Clover ozweego Black Warrior Dad Running shoes Night Walker Retro Men's Shoes Reflection increase Women's Shoes tide Internet celebrity Daddy shoes female 2021 summer new pattern ventilation comfortable leisure time gym shoes Retro Thick bottom increase Single shoes tide Sao Fen Paris Daddy shoes Three generations combination increase Thick bottom ins tide Single shoes Women's Shoes leisure time motion track3.0 Paris Home B Daddy shoes one three generation triple s Thick bottom increase men and women lovers leisure time motion Fashion shoes Dirty shoes U.S.A quality goods Jeffrey Campbell temperament crude high-heeled dollskill Buckles Low top shoes female widow D1G New products anniversary Graffiti high-heeled shoes Internet celebrity Show Sharp point Fine heel Women's shoes Europe Versatile Retro Women's Shoes 2021 Autumn and winter new pattern Low Gang Single shoes female genuine leather Flat bottom Frenulum Color matching motion Casual shoes male skate shoes tide Little white shoes female Josiny Spring and summer 2021 new pattern Korean version Versatile Leisure fashion ventilation student Flat bottom gym shoes European goods Forrest Gump Daddy shoes female tide 2021 autumn new pattern Small pretty waist gym shoes Frenulum Slope heel Single shoes Women's Shoes tide 【 goods in stock 】 devil sisters Sheep puff Lolita original Halloween Thick bottom Women's Shoes hottie high-heeled Women's Shoes Zhou Yangqing Same 2021 Spring and summer new pattern Thick bottom Shoe of sponge cake motion leisure time lovers P family Daddy shoes female ins tide 20 new pattern Internet celebrity Sharp point Single shoes female high-heeled genuine leather Fine heel Shallow mouth sexy Bridesmaid Women's Shoes Wedding shoes 6cm 10cm	What type of shoes are mentioned as being suitable for both men and women?
  I just started a new blog on my ultralight gear. My gear list in all it's glory is located on: each item of gear, I'm writing an in-depth review for the item and how we have used it. Would love to get feedback and the site and our gear and/or comments from people on how we can fine tune.Currently my wifes pack is 7.5 lbs base weight, and mine is 10.5 lbs.Thanks!-Brett Edited by brettmarl on 09/09/2006 15:59:48 MDT. Brett, Your BLOG looks good.You should put the size of your items where their is one such as pants, shoes, jacket etc. There is a golf like "handicap' for anyone that wears larger then size medium or size 9.5 shoe. Sure.I think you might recheck some of your math. Not sure but some totals look low. Don't trust the posted weightof gear, weigh it yourself if you haven't.Why is your pack list so heavy? I agree, nice looking blog. Bill is right on listing the sizes, other than that....looks great! Brett - nice list, and nice format!(One small typo: it currently says "Cloudburt" for the tent.) Edited by slnsf on 09/09/2006 18:08:48 MDT. Great site with good info. I'm trying to decide between the GoLite Infinity and Jam and I think after reading your blog, that the Jam should be plenty big.I'm interested to see what's in your first aid kit.Also, any issues with the water purification tablets? I currently use a MSR miniworks pump and I'm looking to lighten up... At first I thought you might not be warm enough as I reviewed your North Cascade hike, then I recounted your layers. Very nice site! What I was a little confused about was the opening statement of getting four days of backpacking gear into a pack, yet at the bottom the food for two was estimated for three days. However I now understand that these are not mutually exclusive statements. thanks for the feedback.i fixed the cloudburt typo (thanks), and the 4 vs. 3 days. ai also completed all my gear posts - including the innards of my first aid kit and my experiences with the MicroPUR tabs.the weights listed should be the ones that i weighed myself (unless, i've mis-typed in some areas)bill - you say to check my math with "Why is your pack list so heavy?". not sure what you are refering to here.great idea on including the sizes. You must login to post. MEMBERSHIP IS REQUIRED TO POST: You must be a Forum, Annual or Lifetime Member to post messages in the backpackinglight.com forums. SUBSCRIBE NOW »	What are the base weights of the blogger's and his wife's packs?

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 10
• per_device_eval_batch_size: 10
• num_train_epochs: 30
• warmup_ratio: 0.1
• bf16: True
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 10
• per_device_eval_batch_size: 10
• 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: 5e-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: 30
• 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: True
• 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}
• 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: 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
• eval_on_start: False
• eval_use_gather_object: False
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss	Validation Loss
0.0521	50	0.116	-
0.1042	100	0.1294	0.0527
0.1562	150	0.0813	-
0.2083	200	0.0813	0.0300
0.2604	250	0.0554	-
0.3125	300	0.0626	0.0192
0.3646	350	0.0667	-
0.4167	400	0.0564	0.0161
0.4688	450	0.0496	-
0.5208	500	0.0363	0.0146
0.5729	550	0.0465	-
0.625	600	0.0337	0.0141
0.6771	650	0.0485	-
0.7292	700	0.0568	0.0129
0.7812	750	0.0338	-
0.8333	800	0.0238	0.0111
0.8854	850	0.0537	-
0.9375	900	0.0419	0.0146
0.9896	950	0.0333	-
1.0417	1000	0.039	0.0108

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.2.1
• Transformers: 4.44.2
• PyTorch: 2.4.1+cu121
• Accelerate: 0.34.2
• 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",
}

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}
}