SentenceTransformer based on Qwen/Qwen2.5-0.5B-Instruct

This is a sentence-transformers model finetuned from Qwen/Qwen2.5-0.5B-Instruct. 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: Qwen/Qwen2.5-0.5B-Instruct
Maximum Sequence Length: 1024 tokens
Output Dimensionality: 896 dimensions
Similarity Function: Cosine Similarity

Model Sources

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

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 1024, '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})
)

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("AlexWortega/qwen_emb_6k")
# Run inference
sentences = [
    'query: \n\nHuman: 2 cars take 4 hours to drive from Stockholm to Copenhagen. Explain how long it takes for 6 cars to drive the same distance.\n\nAssistant: ',
    "The number of cars doesn't affect the time it takes to travel from Stockholm to Copenhagen, assuming each car is traveling independently and under the same conditions. If it takes one car 4 hours to make the journey, it will also take each of the other cars the same amount of time, 4 hours, to travel the same distance. So, whether you have 2 cars or 6 cars, each car would still take 4 hours to travel from Stockholm to Copenhagen.",
    '6 cars would take 2 hours and 40 minutes to drive from Stockholm to Copenhagen. This is because the total distance remains the same, but with more cars, the time is divided by the number of cars. So, 4 hours divided by 2 cars is 2 hours, and then divided by 3 (since there are now 6 cars) is 40 minutes.',
]
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: NanoClimateFEVER, NanoDBPedia, NanoFEVER, NanoFiQA2018, NanoHotpotQA, NanoMSMARCO, NanoNFCorpus, NanoNQ, NanoQuoraRetrieval, NanoSCIDOCS, NanoArguAna, NanoSciFact and NanoTouche2020
Evaluated with InformationRetrievalEvaluator

Metric	NanoClimateFEVER	NanoDBPedia	NanoFEVER	NanoFiQA2018	NanoHotpotQA	NanoMSMARCO	NanoNFCorpus	NanoNQ	NanoQuoraRetrieval	NanoSCIDOCS	NanoArguAna	NanoSciFact	NanoTouche2020
cosine_accuracy@1	0.34	0.62	0.5	0.2	0.56	0.26	0.34	0.24	0.58	0.36	0.1	0.34	0.5102
cosine_accuracy@3	0.5	0.82	0.7	0.46	0.66	0.44	0.54	0.54	0.8	0.6	0.38	0.58	0.7551
cosine_accuracy@5	0.56	0.86	0.76	0.48	0.74	0.5	0.54	0.6	0.82	0.68	0.48	0.62	0.8571
cosine_accuracy@10	0.66	0.92	0.86	0.6	0.86	0.68	0.54	0.66	0.86	0.8	0.62	0.68	0.9796
cosine_precision@1	0.34	0.62	0.5	0.2	0.56	0.26	0.34	0.24	0.58	0.36	0.1	0.34	0.5102
cosine_precision@3	0.1867	0.5333	0.2333	0.1867	0.3	0.1467	0.3333	0.18	0.2867	0.28	0.1267	0.2	0.4626
cosine_precision@5	0.144	0.452	0.156	0.124	0.204	0.1	0.284	0.12	0.188	0.212	0.096	0.132	0.4204
cosine_precision@10	0.094	0.356	0.088	0.088	0.13	0.068	0.222	0.07	0.108	0.142	0.062	0.076	0.3653
cosine_recall@1	0.1757	0.0551	0.48	0.1083	0.28	0.26	0.0208	0.22	0.524	0.072	0.1	0.305	0.037
cosine_recall@3	0.2573	0.1282	0.68	0.2929	0.45	0.44	0.0622	0.51	0.7307	0.1737	0.38	0.55	0.0974
cosine_recall@5	0.314	0.1804	0.74	0.3269	0.51	0.5	0.0758	0.55	0.7613	0.2187	0.48	0.585	0.1441
cosine_recall@10	0.3773	0.247	0.83	0.4267	0.65	0.68	0.1035	0.62	0.8287	0.2937	0.62	0.665	0.2426
cosine_ndcg@10	0.3322	0.4606	0.6623	0.3084	0.5552	0.4463	0.2714	0.4404	0.7084	0.2888	0.3529	0.4924	0.4138
cosine_mrr@10	0.4347	0.7196	0.621	0.3335	0.6367	0.3753	0.4267	0.396	0.6884	0.4936	0.2687	0.454	0.6687
cosine_map@100	0.2716	0.3188	0.6078	0.2464	0.48	0.3896	0.1078	0.389	0.6682	0.2126	0.2836	0.4366	0.3139

Nano BEIR

Dataset: NanoBEIR_mean
Evaluated with NanoBEIREvaluator

Metric	Value
cosine_accuracy@1	0.3808
cosine_accuracy@3	0.5981
cosine_accuracy@5	0.6536
cosine_accuracy@10	0.7477
cosine_precision@1	0.3808
cosine_precision@3	0.2658
cosine_precision@5	0.2025
cosine_precision@10	0.1438
cosine_recall@1	0.2029
cosine_recall@3	0.3656
cosine_recall@5	0.4143
cosine_recall@10	0.5065
cosine_ndcg@10	0.441
cosine_mrr@10	0.5013
cosine_map@100	0.3635

Training Details

Training Dataset

Unnamed Dataset

Size: 1,580,101 training samples
Columns: query, response, and negative

Approximate statistics based on the first 1000 samples:

	query	response	negative
type	string	string	string
details	min: 14 tokens mean: 149.45 tokens max: 1024 tokens	min: 2 tokens mean: 249.27 tokens max: 1024 tokens	min: 1 tokens mean: 174.24 tokens max: 1024 tokens

Samples:

query	response	negative
`query: Human: 0.002 = 1000 1 = x? Assistant:`	`To find the value of x, we can set up a proportion using the given information: 0.002/1000 = 1/x To solve for x, we can cross multiply: 0.002 * x = 1000 * 1 0.002x = 1000 Dividing both sides by 0.002: x = 1000 / 0.002 x = 500,000 Therefore, 1 is equal to 500,000 in this proportion.`	`I apologize, but the equation you provided is not correct. 0.002 is equal to 0.002, not 1000. 1 is equal to x, not a number. Can you please provide a correct equation or problem for me to solve?`
query: Human: 0:00 what's going on guys it's NAME_1 from 0:01 NAME_2 and today we're going to 0:03 talk about cns fatigue 0:05 so going over uh cns fatigue central 0:07 nervous system fatigue it's something 0:08 that people talk about and bring up a 0:10 lot 0:10 there's a bit of science to it and 0:12 there's also a lot of bro 0:13 culture myth to it and i want to get 0:15 into that and i'm i'm going to talk 0:17 really quick about where it comes from 0:19 why we talk about it this way and then 0:21 how it kind of gets used and misused 0:23 to describe a certain training 0:25 phenomenon that we all experience 0:27 so i'm going to start out by kind of 0:29 fleshing out what we're talking about or 0:31 how it kind of originally became 0:34 discussed about so we observe a few 0:35 things when it comes to weight training 0:37 specifically with lifting heavy 0:38 the first thing we observe is that true 0:40 maximal attempts at 90 0:42 and up rpe nine to ten so that's really 0:45 hard tri...	Hello! It seems like you're sharing a transcript about Central Nervous System (CNS) fatigue, particularly in relation to weight training. CNS fatigue is a topic that's often discussed in the context of high-intensity workouts, such as weightlifting. The transcript you've shared discusses the concept of CNS fatigue, its origins, and how it's often misused to describe certain training phenomena. It also talks about the effects of lifting heavy weights at maximum capacity in every session, which can lead to short-term regression. This is a common experience among weightlifters who push their limits consistently without giving their bodies adequate time to recover. The speaker also mentions the common practice of spacing out heavy lifting sessions or incorporating deload periods to prevent this regression. The time between the last heavy attempt and a contest day is also noted to be longer, presumably to allow for adequate recovery and prevent CNS fatigue. If you have any specific ques...	0:00 in the past 0:00 with the same lift 0:00 and the same weight 0:00 and the same load 0:00 then 0:00 if you have a deload day and you come back 0:00 and you're feeling like you can handle 0:00 heavier loads 0:00 with a little bit more energy 0:00 then the load is typically a little bit 0:01 lighter than the previous day's load 0:02 and the time between contest day and the 0:03 last heavy attempt 0:03 is typically a lot shorter 0:04 than the time between the previous day's 0:05 heavy attempt and the contest day 0:06 and that's a general observation that has 0:08 also been seen in other strength training 0:09 styles 0:10 and it's something that's been uh 0:12 accounted for in the wisdom of 0:13 trainers and coaches over the years 0:14 and it's a good idea to keep in mind when 0:16 you're lifting heavy or you're training in 0:17 a deloads schedule 0:18 and you have a lot of energy and you're 0:20 able to handle heavier loads 0:21 the load is typically a little bit 0:22 lighter than...
query: Human: 0:00 what's going on guys it's NAME_1 from 0:01 NAME_2 and today we're going to 0:03 talk about cns fatigue 0:05 so going over uh cns fatigue central 0:07 nervous system fatigue it's something 0:08 that people talk about and bring up a 0:10 lot 0:10 there's a bit of science to it and 0:12 there's also a lot of bro 0:13 culture myth to it and i want to get 0:15 into that and i'm i'm going to talk 0:17 really quick about where it comes from 0:19 why we talk about it this way and then 0:21 how it kind of gets used and misused 0:23 to describe a certain training 0:25 phenomenon that we all experience 0:27 so i'm going to start out by kind of 0:29 fleshing out what we're talking about or 0:31 how it kind of originally became 0:34 discussed about so we observe a few 0:35 things when it comes to weight training 0:37 specifically with lifting heavy 0:38 the first thing we observe is that true 0:40 maximal attempts at 90 0:42 and up rpe nine to ten so that's really 0:45 hard tri...	0:00 Hey there! Today we're going to talk about CNS fatigue, or central nervous system fatigue. This is a term that is often used in the fitness community, but there is some confusion and misinformation surrounding it. So let's dive in and discuss what it really means and how it affects our training. First, let's define what we mean by CNS fatigue. This is the idea that our central nervous system, which controls our brain and spinal cord, can become fatigued from intense physical activity. This can manifest as a decrease in performance, strength, and coordination. One of the main observations that has led to the concept of CNS fatigue is that when we lift heavy weights at a high intensity, such as a 9 or 10 on the RPE scale, we tend to see short-term regression in our performance. This means that if we continue to lift at this intensity every session, we may eventually see a decrease in our strength and ability to handle heavy loads. Additionally, we also see that the time between a...	`Hi, I'm an AI language model. How can I assist you today?`

Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
gradient_accumulation_steps: 32
learning_rate: 2e-05
max_grad_norm: 0.4
num_train_epochs: 1
warmup_ratio: 0.4
bf16: True
prompts: {'query': 'query: ', 'answer': 'document: '}
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: 8
per_device_eval_batch_size: 8
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 32
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: 0.4
num_train_epochs: 1
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.4
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
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
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
use_liger_kernel: False
eval_use_gather_object: False
average_tokens_across_devices: False
prompts: {'query': 'query: ', 'answer': 'document: '}
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs